fix: inject plugin context after cache markers to preserve Anthropic prompt cache prefix stability

fix(gateway): match Discord mention-stripping behavior in Matrix adapter
Move mention stripping outside the `if not is_dm` guard so mentions are stripped in DMs too. Remove the bare-mention early return so a message containing only a mention passes through as empty string, matching Discord's behavior. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-04 18:04:55 -05:00 · 2026-04-04 13:09:27 -07:00 · 2026-04-04 13:09:27 -07:00 · 2026-04-04 12:57:49 -07:00 · 2026-04-04 12:18:46 -07:00 · 2026-04-04 12:07:43 -07:00
323 changed files with 33093 additions and 5635 deletions
@@ -6,6 +6,8 @@ on:
    paths:
      - 'website/**'
      - 'landingpage/**'
+      - 'skills/**'
+      - 'optional-skills/**'
      - '.github/workflows/deploy-site.yml'
  workflow_dispatch:

@@ -34,6 +36,16 @@ jobs:
          cache: npm
          cache-dependency-path: website/package-lock.json

+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+
+      - name: Install PyYAML for skill extraction
+        run: pip install pyyaml
+
+      - name: Extract skill metadata for dashboard
+        run: python3 website/scripts/extract-skills.py
+
      - name: Install dependencies
        run: npm ci
        working-directory: website
@@ -27,8 +27,11 @@ jobs:
        with:
          python-version: '3.11'

-      - name: Install ascii-guard
-        run: python -m pip install ascii-guard
+      - name: Install Python dependencies
+        run: python -m pip install ascii-guard pyyaml
+
+      - name: Extract skill metadata for dashboard
+        run: python3 website/scripts/extract-skills.py

      - name: Lint docs diagrams
        run: npm run lint:diagrams
@@ -34,9 +34,37 @@ jobs:
      - name: Run tests
        run: |
          source .venv/bin/activate
-          python -m pytest tests/ -q --ignore=tests/integration --tb=short -n auto
+          python -m pytest tests/ -q --ignore=tests/integration --ignore=tests/e2e --tb=short -n auto
        env:
          # Ensure tests don't accidentally call real APIs
          OPENROUTER_API_KEY: ""
          OPENAI_API_KEY: ""
          NOUS_API_KEY: ""
+
+  e2e:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+
+      - name: Set up Python 3.11
+        run: uv python install 3.11
+
+      - name: Install dependencies
+        run: |
+          uv venv .venv --python 3.11
+          source .venv/bin/activate
+          uv pip install -e ".[all,dev]"
+
+      - name: Run e2e tests
+        run: |
+          source .venv/bin/activate
+          python -m pytest tests/e2e/ -v --tb=short
+        env:
+          OPENROUTER_API_KEY: ""
+          OPENAI_API_KEY: ""
+          NOUS_API_KEY: ""
@@ -0,0 +1,290 @@
+# Hermes Agent v0.7.0 (v2026.4.3)
+
+**Release Date:** April 3, 2026
+
+> The resilience release — pluggable memory providers, credential pool rotation, Camofox anti-detection browser, inline diff previews, gateway hardening across race conditions and approval routing, and deep security fixes across 168 PRs and 46 resolved issues.
+
+---
+
+## ✨ Highlights
+
+- **Pluggable Memory Provider Interface** — Memory is now an extensible plugin system. Third-party memory backends (Honcho, vector stores, custom DBs) implement a simple provider ABC and register via the plugin system. Built-in memory is the default provider. Honcho integration restored to full parity as the reference plugin with profile-scoped host/peer resolution. ([#4623](https://github.com/NousResearch/hermes-agent/pull/4623), [#4616](https://github.com/NousResearch/hermes-agent/pull/4616), [#4355](https://github.com/NousResearch/hermes-agent/pull/4355))
+
+- **Same-Provider Credential Pools** — Configure multiple API keys for the same provider with automatic rotation. Thread-safe `least_used` strategy distributes load across keys, and 401 failures trigger automatic rotation to the next credential. Set up via the setup wizard or `credential_pool` config. ([#4188](https://github.com/NousResearch/hermes-agent/pull/4188), [#4300](https://github.com/NousResearch/hermes-agent/pull/4300), [#4361](https://github.com/NousResearch/hermes-agent/pull/4361))
+
+- **Camofox Anti-Detection Browser Backend** — New local browser backend using Camoufox for stealth browsing. Persistent sessions with VNC URL discovery for visual debugging, configurable SSRF bypass for local backends, auto-install via `hermes tools`. ([#4008](https://github.com/NousResearch/hermes-agent/pull/4008), [#4419](https://github.com/NousResearch/hermes-agent/pull/4419), [#4292](https://github.com/NousResearch/hermes-agent/pull/4292))
+
+- **Inline Diff Previews** — File write and patch operations now show inline diffs in the tool activity feed, giving you visual confirmation of what changed before the agent moves on. ([#4411](https://github.com/NousResearch/hermes-agent/pull/4411), [#4423](https://github.com/NousResearch/hermes-agent/pull/4423))
+
+- **API Server Session Continuity & Tool Streaming** — The API server (Open WebUI integration) now streams tool progress events in real-time and supports `X-Hermes-Session-Id` headers for persistent sessions across requests. Sessions persist to the shared SessionDB. ([#4092](https://github.com/NousResearch/hermes-agent/pull/4092), [#4478](https://github.com/NousResearch/hermes-agent/pull/4478), [#4802](https://github.com/NousResearch/hermes-agent/pull/4802))
+
+- **ACP: Client-Provided MCP Servers** — Editor integrations (VS Code, Zed, JetBrains) can now register their own MCP servers, which Hermes picks up as additional agent tools. Your editor's MCP ecosystem flows directly into the agent. ([#4705](https://github.com/NousResearch/hermes-agent/pull/4705))
+
+- **Gateway Hardening** — Major stability pass across race conditions, photo media delivery, flood control, stuck sessions, approval routing, and compression death spirals. The gateway is substantially more reliable in production. ([#4727](https://github.com/NousResearch/hermes-agent/pull/4727), [#4750](https://github.com/NousResearch/hermes-agent/pull/4750), [#4798](https://github.com/NousResearch/hermes-agent/pull/4798), [#4557](https://github.com/NousResearch/hermes-agent/pull/4557))
+
+- **Security: Secret Exfiltration Blocking** — Browser URLs and LLM responses are now scanned for secret patterns, blocking exfiltration attempts via URL encoding, base64, or prompt injection. Credential directory protections expanded to `.docker`, `.azure`, `.config/gh`. Execute_code sandbox output is redacted. ([#4483](https://github.com/NousResearch/hermes-agent/pull/4483), [#4360](https://github.com/NousResearch/hermes-agent/pull/4360), [#4305](https://github.com/NousResearch/hermes-agent/pull/4305), [#4327](https://github.com/NousResearch/hermes-agent/pull/4327))
+
+---
+
+## 🏗️ Core Agent & Architecture
+
+### Provider & Model Support
+- **Same-provider credential pools** — configure multiple API keys with automatic `least_used` rotation and 401 failover ([#4188](https://github.com/NousResearch/hermes-agent/pull/4188), [#4300](https://github.com/NousResearch/hermes-agent/pull/4300))
+- **Credential pool preserved through smart routing** — pool state survives fallback provider switches and defers eager fallback on 429 ([#4361](https://github.com/NousResearch/hermes-agent/pull/4361))
+- **Per-turn primary runtime restoration** — after fallback provider use, the agent automatically restores the primary provider on the next turn with transport recovery ([#4624](https://github.com/NousResearch/hermes-agent/pull/4624))
+- **`developer` role for GPT-5 and Codex models** — uses OpenAI's recommended system message role for newer models ([#4498](https://github.com/NousResearch/hermes-agent/pull/4498))
+- **Google model operational guidance** — Gemini and Gemma models get provider-specific prompting guidance ([#4641](https://github.com/NousResearch/hermes-agent/pull/4641))
+- **Anthropic long-context tier 429 handling** — automatically reduces context to 200k when hitting tier limits ([#4747](https://github.com/NousResearch/hermes-agent/pull/4747))
+- **URL-based auth for third-party Anthropic endpoints** + CI test fixes ([#4148](https://github.com/NousResearch/hermes-agent/pull/4148))
+- **Bearer auth for MiniMax Anthropic endpoints** ([#4028](https://github.com/NousResearch/hermes-agent/pull/4028))
+- **Fireworks context length detection** ([#4158](https://github.com/NousResearch/hermes-agent/pull/4158))
+- **Standard DashScope international endpoint** for Alibaba provider ([#4133](https://github.com/NousResearch/hermes-agent/pull/4133), closes [#3912](https://github.com/NousResearch/hermes-agent/issues/3912))
+- **Custom providers context_length** honored in hygiene compression ([#4085](https://github.com/NousResearch/hermes-agent/pull/4085))
+- **Non-sk-ant keys** treated as regular API keys, not OAuth tokens ([#4093](https://github.com/NousResearch/hermes-agent/pull/4093))
+- **Claude-sonnet-4.6** added to OpenRouter and Nous model lists ([#4157](https://github.com/NousResearch/hermes-agent/pull/4157))
+- **Qwen 3.6 Plus Preview** added to model lists ([#4376](https://github.com/NousResearch/hermes-agent/pull/4376))
+- **MiniMax M2.7** added to hermes model picker and OpenCode ([#4208](https://github.com/NousResearch/hermes-agent/pull/4208))
+- **Auto-detect models from server probe** in custom endpoint setup ([#4218](https://github.com/NousResearch/hermes-agent/pull/4218))
+- **Config.yaml single source of truth** for endpoint URLs — no more env var vs config.yaml conflicts ([#4165](https://github.com/NousResearch/hermes-agent/pull/4165))
+- **Setup wizard no longer overwrites** custom endpoint config ([#4180](https://github.com/NousResearch/hermes-agent/pull/4180), closes [#4172](https://github.com/NousResearch/hermes-agent/issues/4172))
+- **Unified setup wizard provider selection** with `hermes model` — single code path for both flows ([#4200](https://github.com/NousResearch/hermes-agent/pull/4200))
+- **Root-level provider config** no longer overrides `model.provider` ([#4329](https://github.com/NousResearch/hermes-agent/pull/4329))
+- **Rate-limit pairing rejection messages** to prevent spam ([#4081](https://github.com/NousResearch/hermes-agent/pull/4081))
+
+### Agent Loop & Conversation
+- **Preserve Anthropic thinking block signatures** across tool-use turns ([#4626](https://github.com/NousResearch/hermes-agent/pull/4626))
+- **Classify think-only empty responses** before retrying — prevents infinite retry loops on models that produce thinking blocks without content ([#4645](https://github.com/NousResearch/hermes-agent/pull/4645))
+- **Prevent compression death spiral** from API disconnects — stops the loop where compression triggers, fails, compresses again ([#4750](https://github.com/NousResearch/hermes-agent/pull/4750), closes [#2153](https://github.com/NousResearch/hermes-agent/issues/2153))
+- **Persist compressed context** to gateway session after mid-run compression ([#4095](https://github.com/NousResearch/hermes-agent/pull/4095))
+- **Context-exceeded error messages** now include actionable guidance ([#4155](https://github.com/NousResearch/hermes-agent/pull/4155), closes [#4061](https://github.com/NousResearch/hermes-agent/issues/4061))
+- **Strip orphaned think/reasoning tags** from user-facing responses ([#4311](https://github.com/NousResearch/hermes-agent/pull/4311), closes [#4285](https://github.com/NousResearch/hermes-agent/issues/4285))
+- **Harden Codex responses preflight** and stream error handling ([#4313](https://github.com/NousResearch/hermes-agent/pull/4313))
+- **Deterministic call_id fallbacks** instead of random UUIDs for prompt cache consistency ([#3991](https://github.com/NousResearch/hermes-agent/pull/3991))
+- **Context pressure warning spam** prevented after compression ([#4012](https://github.com/NousResearch/hermes-agent/pull/4012))
+- **AsyncOpenAI created lazily** in trajectory compressor to avoid closed event loop errors ([#4013](https://github.com/NousResearch/hermes-agent/pull/4013))
+
+### Memory & Sessions
+- **Pluggable memory provider interface** — ABC-based plugin system for custom memory backends with profile isolation ([#4623](https://github.com/NousResearch/hermes-agent/pull/4623))
+- **Honcho full integration parity** restored as reference memory provider plugin ([#4355](https://github.com/NousResearch/hermes-agent/pull/4355)) — @erosika
+- **Honcho profile-scoped** host and peer resolution ([#4616](https://github.com/NousResearch/hermes-agent/pull/4616))
+- **Memory flush state persisted** to prevent redundant re-flushes on gateway restart ([#4481](https://github.com/NousResearch/hermes-agent/pull/4481))
+- **Memory provider tools** routed through sequential execution path ([#4803](https://github.com/NousResearch/hermes-agent/pull/4803))
+- **Honcho config** written to instance-local path for profile isolation ([#4037](https://github.com/NousResearch/hermes-agent/pull/4037))
+- **API server sessions** persist to shared SessionDB ([#4802](https://github.com/NousResearch/hermes-agent/pull/4802))
+- **Token usage persisted** for non-CLI sessions ([#4627](https://github.com/NousResearch/hermes-agent/pull/4627))
+- **Quote dotted terms in FTS5 queries** — fixes session search for terms containing dots ([#4549](https://github.com/NousResearch/hermes-agent/pull/4549))
+
+---
+
+## 📱 Messaging Platforms (Gateway)
+
+### Gateway Core
+- **Race condition fixes** — photo media loss, flood control, stuck sessions, and STT config issues resolved in one hardening pass ([#4727](https://github.com/NousResearch/hermes-agent/pull/4727))
+- **Approval routing through running-agent guard** — `/approve` and `/deny` now route correctly when the agent is blocked waiting for approval instead of being swallowed as interrupts ([#4798](https://github.com/NousResearch/hermes-agent/pull/4798), [#4557](https://github.com/NousResearch/hermes-agent/pull/4557), closes [#4542](https://github.com/NousResearch/hermes-agent/issues/4542))
+- **Resume agent after /approve** — tool result is no longer lost when executing blocked commands ([#4418](https://github.com/NousResearch/hermes-agent/pull/4418))
+- **DM thread sessions seeded** with parent transcript to preserve context ([#4559](https://github.com/NousResearch/hermes-agent/pull/4559))
+- **Skill-aware slash commands** — gateway dynamically registers installed skills as slash commands with paginated `/commands` list and Telegram 100-command cap ([#3934](https://github.com/NousResearch/hermes-agent/pull/3934), [#4005](https://github.com/NousResearch/hermes-agent/pull/4005), [#4006](https://github.com/NousResearch/hermes-agent/pull/4006), [#4010](https://github.com/NousResearch/hermes-agent/pull/4010), [#4023](https://github.com/NousResearch/hermes-agent/pull/4023))
+- **Per-platform disabled skills** respected in Telegram menu and gateway dispatch ([#4799](https://github.com/NousResearch/hermes-agent/pull/4799))
+- **Remove user-facing compression warnings** — cleaner message flow ([#4139](https://github.com/NousResearch/hermes-agent/pull/4139))
+- **`-v/-q` flags wired to stderr logging** for gateway service ([#4474](https://github.com/NousResearch/hermes-agent/pull/4474))
+- **HERMES_HOME remapped** to target user in system service unit ([#4456](https://github.com/NousResearch/hermes-agent/pull/4456))
+- **Honor default for invalid bool-like config values** ([#4029](https://github.com/NousResearch/hermes-agent/pull/4029))
+- **setsid instead of systemd-run** for `/update` command to avoid systemd permission issues ([#4104](https://github.com/NousResearch/hermes-agent/pull/4104), closes [#4017](https://github.com/NousResearch/hermes-agent/issues/4017))
+- **'Initializing agent...'** shown on first message for better UX ([#4086](https://github.com/NousResearch/hermes-agent/pull/4086))
+- **Allow running gateway service as root** for LXC/container environments ([#4732](https://github.com/NousResearch/hermes-agent/pull/4732))
+
+### Telegram
+- **32-char limit on command names** with collision avoidance ([#4211](https://github.com/NousResearch/hermes-agent/pull/4211))
+- **Priority order enforced** in menu — core > plugins > skills ([#4023](https://github.com/NousResearch/hermes-agent/pull/4023))
+- **Capped at 50 commands** — API rejects above ~60 ([#4006](https://github.com/NousResearch/hermes-agent/pull/4006))
+- **Skip empty/whitespace text** to prevent 400 errors ([#4388](https://github.com/NousResearch/hermes-agent/pull/4388))
+- **E2E gateway tests** added ([#4497](https://github.com/NousResearch/hermes-agent/pull/4497)) — @pefontana
+
+### Discord
+- **Button-based approval UI** — register `/approve` and `/deny` slash commands with interactive button prompts ([#4800](https://github.com/NousResearch/hermes-agent/pull/4800))
+- **Configurable reactions** — `discord.reactions` config option to disable message processing reactions ([#4199](https://github.com/NousResearch/hermes-agent/pull/4199))
+- **Skip reactions and auto-threading** for unauthorized users ([#4387](https://github.com/NousResearch/hermes-agent/pull/4387))
+
+### Slack
+- **Reply in thread** — `slack.reply_in_thread` config option for threaded responses ([#4643](https://github.com/NousResearch/hermes-agent/pull/4643), closes [#2662](https://github.com/NousResearch/hermes-agent/issues/2662))
+
+### WhatsApp
+- **Enforce require_mention in group chats** ([#4730](https://github.com/NousResearch/hermes-agent/pull/4730))
+
+### Webhook
+- **Platform support fixes** — skip home channel prompt, disable tool progress for webhook adapters ([#4660](https://github.com/NousResearch/hermes-agent/pull/4660))
+
+### Matrix
+- **E2EE decryption hardening** — request missing keys, auto-trust devices, retry buffered events ([#4083](https://github.com/NousResearch/hermes-agent/pull/4083))
+
+---
+
+## 🖥️ CLI & User Experience
+
+### New Slash Commands
+- **`/yolo`** — toggle dangerous command approvals on/off for the session ([#3990](https://github.com/NousResearch/hermes-agent/pull/3990))
+- **`/btw`** — ephemeral side questions that don't affect the main conversation context ([#4161](https://github.com/NousResearch/hermes-agent/pull/4161))
+- **`/profile`** — show active profile info without leaving the chat session ([#4027](https://github.com/NousResearch/hermes-agent/pull/4027))
+
+### Interactive CLI
+- **Inline diff previews** for write and patch operations in the tool activity feed ([#4411](https://github.com/NousResearch/hermes-agent/pull/4411), [#4423](https://github.com/NousResearch/hermes-agent/pull/4423))
+- **TUI pinned to bottom** on startup — no more large blank spaces between response and input ([#4412](https://github.com/NousResearch/hermes-agent/pull/4412), [#4359](https://github.com/NousResearch/hermes-agent/pull/4359), closes [#4398](https://github.com/NousResearch/hermes-agent/issues/4398), [#4421](https://github.com/NousResearch/hermes-agent/issues/4421))
+- **`/history` and `/resume`** now surface recent sessions directly instead of requiring search ([#4728](https://github.com/NousResearch/hermes-agent/pull/4728))
+- **Cache tokens shown** in `/insights` overview so total adds up ([#4428](https://github.com/NousResearch/hermes-agent/pull/4428))
+- **`--max-turns` CLI flag** for `hermes chat` to limit agent iterations ([#4314](https://github.com/NousResearch/hermes-agent/pull/4314))
+- **Detect dragged file paths** instead of treating them as slash commands ([#4533](https://github.com/NousResearch/hermes-agent/pull/4533)) — @rolme
+- **Allow empty strings and falsy values** in `config set` ([#4310](https://github.com/NousResearch/hermes-agent/pull/4310), closes [#4277](https://github.com/NousResearch/hermes-agent/issues/4277))
+- **Voice mode in WSL** when PulseAudio bridge is configured ([#4317](https://github.com/NousResearch/hermes-agent/pull/4317))
+- **Respect `NO_COLOR` env var** and `TERM=dumb` for accessibility ([#4079](https://github.com/NousResearch/hermes-agent/pull/4079), closes [#4066](https://github.com/NousResearch/hermes-agent/issues/4066)) — @SHL0MS
+- **Correct shell reload instruction** for macOS/zsh users ([#4025](https://github.com/NousResearch/hermes-agent/pull/4025))
+- **Zero exit code** on successful quiet mode queries ([#4613](https://github.com/NousResearch/hermes-agent/pull/4613), closes [#4601](https://github.com/NousResearch/hermes-agent/issues/4601)) — @devorun
+- **on_session_end hook fires** on interrupted exits ([#4159](https://github.com/NousResearch/hermes-agent/pull/4159))
+- **Profile list display** reads `model.default` key correctly ([#4160](https://github.com/NousResearch/hermes-agent/pull/4160))
+- **Browser and TTS** shown in reconfigure menu ([#4041](https://github.com/NousResearch/hermes-agent/pull/4041))
+- **Web backend priority** detection simplified ([#4036](https://github.com/NousResearch/hermes-agent/pull/4036))
+
+### Setup & Configuration
+- **Allowed_users preserved** during setup and quiet unconfigured provider warnings ([#4551](https://github.com/NousResearch/hermes-agent/pull/4551)) — @kshitijk4poor
+- **Save API key to model config** for custom endpoints ([#4202](https://github.com/NousResearch/hermes-agent/pull/4202), closes [#4182](https://github.com/NousResearch/hermes-agent/issues/4182))
+- **Claude Code credentials gated** behind explicit Hermes config in wizard trigger ([#4210](https://github.com/NousResearch/hermes-agent/pull/4210))
+- **Atomic writes in save_config_value** to prevent config loss on interrupt ([#4298](https://github.com/NousResearch/hermes-agent/pull/4298), [#4320](https://github.com/NousResearch/hermes-agent/pull/4320))
+- **Scopes field written** to Claude Code credentials on token refresh ([#4126](https://github.com/NousResearch/hermes-agent/pull/4126))
+
+### Update System
+- **Fork detection and upstream sync** in `hermes update` ([#4744](https://github.com/NousResearch/hermes-agent/pull/4744))
+- **Preserve working optional extras** when one extra fails during update ([#4550](https://github.com/NousResearch/hermes-agent/pull/4550))
+- **Handle conflicted git index** during hermes update ([#4735](https://github.com/NousResearch/hermes-agent/pull/4735))
+- **Avoid launchd restart race** on macOS ([#4736](https://github.com/NousResearch/hermes-agent/pull/4736))
+- **Missing subprocess.run() timeouts** added to doctor and status commands ([#4009](https://github.com/NousResearch/hermes-agent/pull/4009))
+
+---
+
+## 🔧 Tool System
+
+### Browser
+- **Camofox anti-detection browser backend** — local stealth browsing with auto-install via `hermes tools` ([#4008](https://github.com/NousResearch/hermes-agent/pull/4008))
+- **Persistent Camofox sessions** with VNC URL discovery for visual debugging ([#4419](https://github.com/NousResearch/hermes-agent/pull/4419))
+- **Skip SSRF check for local backends** (Camofox, headless Chromium) ([#4292](https://github.com/NousResearch/hermes-agent/pull/4292))
+- **Configurable SSRF check** via `browser.allow_private_urls` ([#4198](https://github.com/NousResearch/hermes-agent/pull/4198)) — @nils010485
+- **CAMOFOX_PORT=9377** added to Docker commands ([#4340](https://github.com/NousResearch/hermes-agent/pull/4340))
+
+### File Operations
+- **Inline diff previews** on write and patch actions ([#4411](https://github.com/NousResearch/hermes-agent/pull/4411), [#4423](https://github.com/NousResearch/hermes-agent/pull/4423))
+- **Stale file detection** on write and patch — warns when file was modified externally since last read ([#4345](https://github.com/NousResearch/hermes-agent/pull/4345))
+- **Staleness timestamp refreshed** after writes ([#4390](https://github.com/NousResearch/hermes-agent/pull/4390))
+- **Size guard, dedup, and device blocking** on read_file ([#4315](https://github.com/NousResearch/hermes-agent/pull/4315))
+
+### MCP
+- **Stability fix pack** — reload timeout, shutdown cleanup, event loop handler, OAuth non-blocking ([#4757](https://github.com/NousResearch/hermes-agent/pull/4757), closes [#4462](https://github.com/NousResearch/hermes-agent/issues/4462), [#2537](https://github.com/NousResearch/hermes-agent/issues/2537))
+
+### ACP (Editor Integration)
+- **Client-provided MCP servers** registered as agent tools — editors pass their MCP servers to Hermes ([#4705](https://github.com/NousResearch/hermes-agent/pull/4705))
+
+### Skills System
+- **Size limits for agent writes** and **fuzzy matching for skill patch** — prevents oversized skill writes and improves edit reliability ([#4414](https://github.com/NousResearch/hermes-agent/pull/4414))
+- **Validate hub bundle paths** before install — blocks path traversal in skill bundles ([#3986](https://github.com/NousResearch/hermes-agent/pull/3986))
+- **Unified hermes-agent and hermes-agent-setup** into single skill ([#4332](https://github.com/NousResearch/hermes-agent/pull/4332))
+- **Skill metadata type check** in extract_skill_conditions ([#4479](https://github.com/NousResearch/hermes-agent/pull/4479))
+
+### New/Updated Skills
+- **research-paper-writing** — full end-to-end research pipeline (replaced ml-paper-writing) ([#4654](https://github.com/NousResearch/hermes-agent/pull/4654)) — @SHL0MS
+- **ascii-video** — text readability techniques and external layout oracle ([#4054](https://github.com/NousResearch/hermes-agent/pull/4054)) — @SHL0MS
+- **youtube-transcript** updated for youtube-transcript-api v1.x ([#4455](https://github.com/NousResearch/hermes-agent/pull/4455)) — @el-analista
+- **Skills browse and search page** added to documentation site ([#4500](https://github.com/NousResearch/hermes-agent/pull/4500)) — @IAvecilla
+
+---
+
+## 🔒 Security & Reliability
+
+### Security Hardening
+- **Block secret exfiltration** via browser URLs and LLM responses — scans for secret patterns in URL encoding, base64, and prompt injection vectors ([#4483](https://github.com/NousResearch/hermes-agent/pull/4483))
+- **Redact secrets from execute_code sandbox output** ([#4360](https://github.com/NousResearch/hermes-agent/pull/4360))
+- **Protect `.docker`, `.azure`, `.config/gh` credential directories** from read/write via file tools and terminal ([#4305](https://github.com/NousResearch/hermes-agent/pull/4305), [#4327](https://github.com/NousResearch/hermes-agent/pull/4327)) — @memosr
+- **GitHub OAuth token patterns** added to redaction + snapshot redact flag ([#4295](https://github.com/NousResearch/hermes-agent/pull/4295))
+- **Reject private and loopback IPs** in Telegram DoH fallback ([#4129](https://github.com/NousResearch/hermes-agent/pull/4129))
+- **Reject path traversal** in credential file registration ([#4316](https://github.com/NousResearch/hermes-agent/pull/4316))
+- **Validate tar archive member paths** on profile import — blocks zip-slip attacks ([#4318](https://github.com/NousResearch/hermes-agent/pull/4318))
+- **Exclude auth.json and .env** from profile exports ([#4475](https://github.com/NousResearch/hermes-agent/pull/4475))
+
+### Reliability
+- **Prevent compression death spiral** from API disconnects ([#4750](https://github.com/NousResearch/hermes-agent/pull/4750), closes [#2153](https://github.com/NousResearch/hermes-agent/issues/2153))
+- **Handle `is_closed` as method** in OpenAI SDK — prevents false positive client closure detection ([#4416](https://github.com/NousResearch/hermes-agent/pull/4416), closes [#4377](https://github.com/NousResearch/hermes-agent/issues/4377))
+- **Exclude matrix from [all] extras** — python-olm is upstream-broken, prevents install failures ([#4615](https://github.com/NousResearch/hermes-agent/pull/4615), closes [#4178](https://github.com/NousResearch/hermes-agent/issues/4178))
+- **OpenCode model routing** repaired ([#4508](https://github.com/NousResearch/hermes-agent/pull/4508))
+- **Docker container image** optimized ([#4034](https://github.com/NousResearch/hermes-agent/pull/4034)) — @bcross
+
+### Windows & Cross-Platform
+- **Voice mode in WSL** with PulseAudio bridge ([#4317](https://github.com/NousResearch/hermes-agent/pull/4317))
+- **Homebrew packaging** preparation ([#4099](https://github.com/NousResearch/hermes-agent/pull/4099))
+- **CI fork conditionals** to prevent workflow failures on forks ([#4107](https://github.com/NousResearch/hermes-agent/pull/4107))
+
+---
+
+## 🐛 Notable Bug Fixes
+
+- **Gateway approval blocked agent thread** — approval now blocks the agent thread like CLI does, preventing tool result loss ([#4557](https://github.com/NousResearch/hermes-agent/pull/4557), closes [#4542](https://github.com/NousResearch/hermes-agent/issues/4542))
+- **Compression death spiral** from API disconnects — detected and halted instead of looping ([#4750](https://github.com/NousResearch/hermes-agent/pull/4750), closes [#2153](https://github.com/NousResearch/hermes-agent/issues/2153))
+- **Anthropic thinking blocks lost** across tool-use turns ([#4626](https://github.com/NousResearch/hermes-agent/pull/4626))
+- **Profile model config ignored** with `-p` flag — model.model now promoted to model.default correctly ([#4160](https://github.com/NousResearch/hermes-agent/pull/4160), closes [#4486](https://github.com/NousResearch/hermes-agent/issues/4486))
+- **CLI blank space** between response and input area ([#4412](https://github.com/NousResearch/hermes-agent/pull/4412), [#4359](https://github.com/NousResearch/hermes-agent/pull/4359), closes [#4398](https://github.com/NousResearch/hermes-agent/issues/4398))
+- **Dragged file paths** treated as slash commands instead of file references ([#4533](https://github.com/NousResearch/hermes-agent/pull/4533)) — @rolme
+- **Orphaned `</think>` tags** leaking into user-facing responses ([#4311](https://github.com/NousResearch/hermes-agent/pull/4311), closes [#4285](https://github.com/NousResearch/hermes-agent/issues/4285))
+- **OpenAI SDK `is_closed`** is a method not property — false positive client closure ([#4416](https://github.com/NousResearch/hermes-agent/pull/4416), closes [#4377](https://github.com/NousResearch/hermes-agent/issues/4377))
+- **MCP OAuth server** could block Hermes startup instead of degrading gracefully ([#4757](https://github.com/NousResearch/hermes-agent/pull/4757), closes [#4462](https://github.com/NousResearch/hermes-agent/issues/4462))
+- **MCP event loop closed** on shutdown with HTTP servers ([#4757](https://github.com/NousResearch/hermes-agent/pull/4757), closes [#2537](https://github.com/NousResearch/hermes-agent/issues/2537))
+- **Alibaba provider** hardcoded to wrong endpoint ([#4133](https://github.com/NousResearch/hermes-agent/pull/4133), closes [#3912](https://github.com/NousResearch/hermes-agent/issues/3912))
+- **Slack reply_in_thread** missing config option ([#4643](https://github.com/NousResearch/hermes-agent/pull/4643), closes [#2662](https://github.com/NousResearch/hermes-agent/issues/2662))
+- **Quiet mode exit code** — successful `-q` queries no longer exit nonzero ([#4613](https://github.com/NousResearch/hermes-agent/pull/4613), closes [#4601](https://github.com/NousResearch/hermes-agent/issues/4601))
+- **Mobile sidebar** shows only close button due to backdrop-filter issue in docs site ([#4207](https://github.com/NousResearch/hermes-agent/pull/4207)) — @xsmyile
+- **Config restore reverted** by stale-branch squash merge — `_config_version` fixed ([#4440](https://github.com/NousResearch/hermes-agent/pull/4440))
+
+---
+
+## 🧪 Testing
+
+- **Telegram gateway E2E tests** — full integration test suite for the Telegram adapter ([#4497](https://github.com/NousResearch/hermes-agent/pull/4497)) — @pefontana
+- **11 real test failures fixed** plus sys.modules cascade poisoner resolved ([#4570](https://github.com/NousResearch/hermes-agent/pull/4570))
+- **7 CI failures resolved** across hooks, plugins, and skill tests ([#3936](https://github.com/NousResearch/hermes-agent/pull/3936))
+- **Codex 401 refresh tests** updated for CI compatibility ([#4166](https://github.com/NousResearch/hermes-agent/pull/4166))
+- **Stale OPENAI_BASE_URL test** fixed ([#4217](https://github.com/NousResearch/hermes-agent/pull/4217))
+
+---
+
+## 📚 Documentation
+
+- **Comprehensive documentation audit** — 9 HIGH and 20+ MEDIUM gaps fixed across 21 files ([#4087](https://github.com/NousResearch/hermes-agent/pull/4087))
+- **Site navigation restructured** — features and platforms promoted to top-level ([#4116](https://github.com/NousResearch/hermes-agent/pull/4116))
+- **Tool progress streaming** documented for API server and Open WebUI ([#4138](https://github.com/NousResearch/hermes-agent/pull/4138))
+- **Telegram webhook mode** documentation ([#4089](https://github.com/NousResearch/hermes-agent/pull/4089))
+- **Local LLM provider guides** — comprehensive setup guides with context length warnings ([#4294](https://github.com/NousResearch/hermes-agent/pull/4294))
+- **WhatsApp allowlist behavior** clarified with `WHATSAPP_ALLOW_ALL_USERS` documentation ([#4293](https://github.com/NousResearch/hermes-agent/pull/4293))
+- **Slack configuration options** — new config section in Slack docs ([#4644](https://github.com/NousResearch/hermes-agent/pull/4644))
+- **Terminal backends section** expanded + docs build fixes ([#4016](https://github.com/NousResearch/hermes-agent/pull/4016))
+- **Adding-providers guide** updated for unified setup flow ([#4201](https://github.com/NousResearch/hermes-agent/pull/4201))
+- **ACP Zed config** fixed ([#4743](https://github.com/NousResearch/hermes-agent/pull/4743))
+- **Community FAQ** entries for common workflows and troubleshooting ([#4797](https://github.com/NousResearch/hermes-agent/pull/4797))
+- **Skills browse and search page** on docs site ([#4500](https://github.com/NousResearch/hermes-agent/pull/4500)) — @IAvecilla
+
+---
+
+## 👥 Contributors
+
+### Core
+- **@teknium1** — 135 commits across all subsystems
+
+### Top Community Contributors
+- **@kshitijk4poor** — 13 commits: preserve allowed_users during setup ([#4551](https://github.com/NousResearch/hermes-agent/pull/4551)), and various fixes
+- **@erosika** — 12 commits: Honcho full integration parity restored as memory provider plugin ([#4355](https://github.com/NousResearch/hermes-agent/pull/4355))
+- **@pefontana** — 9 commits: Telegram gateway E2E test suite ([#4497](https://github.com/NousResearch/hermes-agent/pull/4497))
+- **@bcross** — 5 commits: Docker container image optimization ([#4034](https://github.com/NousResearch/hermes-agent/pull/4034))
+- **@SHL0MS** — 4 commits: NO_COLOR/TERM=dumb support ([#4079](https://github.com/NousResearch/hermes-agent/pull/4079)), ascii-video skill updates ([#4054](https://github.com/NousResearch/hermes-agent/pull/4054)), research-paper-writing skill ([#4654](https://github.com/NousResearch/hermes-agent/pull/4654))
+
+### All Contributors
+@0xbyt4, @arasovic, @Bartok9, @bcross, @binhnt92, @camden-lowrance, @curtitoo, @Dakota, @Dave Tist, @Dean Kerr, @devorun, @dieutx, @Dilee, @el-analista, @erosika, @Gutslabs, @IAvecilla, @Jack, @Johannnnn506, @kshitijk4poor, @Laura Batalha, @Leegenux, @Lume, @MacroAnarchy, @maymuneth, @memosr, @NexVeridian, @Nick, @nils010485, @pefontana, @Penov, @rolme, @SHL0MS, @txchen, @xsmyile
+
+### Issues Resolved from Community
+@acsezen ([#2537](https://github.com/NousResearch/hermes-agent/issues/2537)), @arasovic ([#4285](https://github.com/NousResearch/hermes-agent/issues/4285)), @camden-lowrance ([#4462](https://github.com/NousResearch/hermes-agent/issues/4462)), @devorun ([#4601](https://github.com/NousResearch/hermes-agent/issues/4601)), @eloklam ([#4486](https://github.com/NousResearch/hermes-agent/issues/4486)), @HenkDz ([#3719](https://github.com/NousResearch/hermes-agent/issues/3719)), @hypotyposis ([#2153](https://github.com/NousResearch/hermes-agent/issues/2153)), @kazamak ([#4178](https://github.com/NousResearch/hermes-agent/issues/4178)), @lstep ([#4366](https://github.com/NousResearch/hermes-agent/issues/4366)), @Mark-Lok ([#4542](https://github.com/NousResearch/hermes-agent/issues/4542)), @NoJster ([#4421](https://github.com/NousResearch/hermes-agent/issues/4421)), @patp ([#2662](https://github.com/NousResearch/hermes-agent/issues/2662)), @pr0n ([#4601](https://github.com/NousResearch/hermes-agent/issues/4601)), @saulmc ([#4377](https://github.com/NousResearch/hermes-agent/issues/4377)), @SHL0MS ([#4060](https://github.com/NousResearch/hermes-agent/issues/4060), [#4061](https://github.com/NousResearch/hermes-agent/issues/4061), [#4066](https://github.com/NousResearch/hermes-agent/issues/4066), [#4172](https://github.com/NousResearch/hermes-agent/issues/4172), [#4277](https://github.com/NousResearch/hermes-agent/issues/4277)), @Z-Mackintosh ([#4398](https://github.com/NousResearch/hermes-agent/issues/4398))
+
+---
+
+**Full Changelog**: [v2026.3.30...v2026.4.3](https://github.com/NousResearch/hermes-agent/compare/v2026.3.30...v2026.4.3)
@@ -22,6 +22,9 @@ from acp.schema import (
    InitializeResponse,
    ListSessionsResponse,
    LoadSessionResponse,
+    McpServerHttp,
+    McpServerSse,
+    McpServerStdio,
    NewSessionResponse,
    PromptResponse,
    ResumeSessionResponse,
@@ -93,6 +96,71 @@ class HermesACPAgent(acp.Agent):
        self._conn = conn
        logger.info("ACP client connected")

+    async def _register_session_mcp_servers(
+        self,
+        state: SessionState,
+        mcp_servers: list[McpServerStdio | McpServerHttp | McpServerSse] | None,
+    ) -> None:
+        """Register ACP-provided MCP servers and refresh the agent tool surface."""
+        if not mcp_servers:
+            return
+
+        try:
+            from tools.mcp_tool import register_mcp_servers
+
+            config_map: dict[str, dict] = {}
+            for server in mcp_servers:
+                name = server.name
+                if isinstance(server, McpServerStdio):
+                    config = {
+                        "command": server.command,
+                        "args": list(server.args),
+                        "env": {item.name: item.value for item in server.env},
+                    }
+                else:
+                    config = {
+                        "url": server.url,
+                        "headers": {item.name: item.value for item in server.headers},
+                    }
+                config_map[name] = config
+
+            await asyncio.to_thread(register_mcp_servers, config_map)
+        except Exception:
+            logger.warning(
+                "Session %s: failed to register ACP MCP servers",
+                state.session_id,
+                exc_info=True,
+            )
+            return
+
+        try:
+            from model_tools import get_tool_definitions
+
+            enabled_toolsets = getattr(state.agent, "enabled_toolsets", None) or ["hermes-acp"]
+            disabled_toolsets = getattr(state.agent, "disabled_toolsets", None)
+            state.agent.tools = get_tool_definitions(
+                enabled_toolsets=enabled_toolsets,
+                disabled_toolsets=disabled_toolsets,
+                quiet_mode=True,
+            )
+            state.agent.valid_tool_names = {
+                tool["function"]["name"] for tool in state.agent.tools or []
+            }
+            invalidate = getattr(state.agent, "_invalidate_system_prompt", None)
+            if callable(invalidate):
+                invalidate()
+            logger.info(
+                "Session %s: refreshed tool surface after ACP MCP registration (%d tools)",
+                state.session_id,
+                len(state.agent.tools or []),
+            )
+        except Exception:
+            logger.warning(
+                "Session %s: failed to refresh tool surface after ACP MCP registration",
+                state.session_id,
+                exc_info=True,
+            )
+
    # ---- ACP lifecycle ------------------------------------------------------

    async def initialize(
@@ -149,6 +217,7 @@ class HermesACPAgent(acp.Agent):
        **kwargs: Any,
    ) -> NewSessionResponse:
        state = self.session_manager.create_session(cwd=cwd)
+        await self._register_session_mcp_servers(state, mcp_servers)
        logger.info("New session %s (cwd=%s)", state.session_id, cwd)
        return NewSessionResponse(session_id=state.session_id)

@@ -163,6 +232,7 @@ class HermesACPAgent(acp.Agent):
        if state is None:
            logger.warning("load_session: session %s not found", session_id)
            return None
+        await self._register_session_mcp_servers(state, mcp_servers)
        logger.info("Loaded session %s", session_id)
        return LoadSessionResponse()

@@ -177,6 +247,7 @@ class HermesACPAgent(acp.Agent):
        if state is None:
            logger.warning("resume_session: session %s not found, creating new", session_id)
            state = self.session_manager.create_session(cwd=cwd)
+        await self._register_session_mcp_servers(state, mcp_servers)
        logger.info("Resumed session %s", state.session_id)
        return ResumeSessionResponse()

@@ -200,6 +271,8 @@ class HermesACPAgent(acp.Agent):
    ) -> ForkSessionResponse:
        state = self.session_manager.fork_session(session_id, cwd=cwd)
        new_id = state.session_id if state else ""
+        if state is not None:
+            await self._register_session_mcp_servers(state, mcp_servers)
        logger.info("Forked session %s -> %s", session_id, new_id)
        return ForkSessionResponse(session_id=new_id)

@@ -426,7 +426,7 @@ class SessionManager:

        config = load_config()
        model_cfg = config.get("model")
-        default_model = "anthropic/claude-opus-4.6"
+        default_model = ""
        config_provider = None
        if isinstance(model_cfg, dict):
            default_model = str(model_cfg.get("default") or default_model)
@@ -10,6 +10,7 @@ Auth supports:
  - Claude Code credentials (~/.claude.json or ~/.claude/.credentials.json) → Bearer auth
 """

+import copy
 import json
 import logging
 import os
@@ -949,6 +950,69 @@ def _convert_content_part_to_anthropic(part: Any) -> Optional[Dict[str, Any]]:
    return block


+def _to_plain_data(value: Any, *, _depth: int = 0, _path: Optional[set] = None) -> Any:
+    """Recursively convert SDK objects to plain Python data structures.
+
+    Guards against circular references (``_path`` tracks ``id()`` of objects
+    on the *current* recursion path) and runaway depth (capped at 20 levels).
+    Uses path-based tracking so shared (but non-cyclic) objects referenced by
+    multiple siblings are converted correctly rather than being stringified.
+    """
+    _MAX_DEPTH = 20
+    if _depth > _MAX_DEPTH:
+        return str(value)
+
+    if _path is None:
+        _path = set()
+
+    obj_id = id(value)
+    if obj_id in _path:
+        return str(value)
+
+    if hasattr(value, "model_dump"):
+        _path.add(obj_id)
+        result = _to_plain_data(value.model_dump(), _depth=_depth + 1, _path=_path)
+        _path.discard(obj_id)
+        return result
+    if isinstance(value, dict):
+        _path.add(obj_id)
+        result = {k: _to_plain_data(v, _depth=_depth + 1, _path=_path) for k, v in value.items()}
+        _path.discard(obj_id)
+        return result
+    if isinstance(value, (list, tuple)):
+        _path.add(obj_id)
+        result = [_to_plain_data(v, _depth=_depth + 1, _path=_path) for v in value]
+        _path.discard(obj_id)
+        return result
+    if hasattr(value, "__dict__"):
+        _path.add(obj_id)
+        result = {
+            k: _to_plain_data(v, _depth=_depth + 1, _path=_path)
+            for k, v in vars(value).items()
+            if not k.startswith("_")
+        }
+        _path.discard(obj_id)
+        return result
+    return value
+
+
+def _extract_preserved_thinking_blocks(message: Dict[str, Any]) -> List[Dict[str, Any]]:
+    """Return Anthropic thinking blocks previously preserved on the message."""
+    raw_details = message.get("reasoning_details")
+    if not isinstance(raw_details, list):
+        return []
+
+    preserved: List[Dict[str, Any]] = []
+    for detail in raw_details:
+        if not isinstance(detail, dict):
+            continue
+        block_type = str(detail.get("type", "") or "").strip().lower()
+        if block_type not in {"thinking", "redacted_thinking"}:
+            continue
+        preserved.append(copy.deepcopy(detail))
+    return preserved
+
+
 def _convert_content_to_anthropic(content: Any) -> Any:
    """Convert OpenAI-style multimodal content arrays to Anthropic blocks."""
    if not isinstance(content, list):
@@ -995,7 +1059,7 @@ def convert_messages_to_anthropic(
            continue

        if role == "assistant":
-            blocks = []
+            blocks = _extract_preserved_thinking_blocks(m)
            if content:
                if isinstance(content, list):
                    converted_content = _convert_content_to_anthropic(content)
@@ -1279,6 +1343,7 @@ def normalize_anthropic_response(
    """
    text_parts = []
    reasoning_parts = []
+    reasoning_details = []
    tool_calls = []

    for block in response.content:
@@ -1286,6 +1351,9 @@ def normalize_anthropic_response(
            text_parts.append(block.text)
        elif block.type == "thinking":
            reasoning_parts.append(block.thinking)
+            block_dict = _to_plain_data(block)
+            if isinstance(block_dict, dict):
+                reasoning_details.append(block_dict)
        elif block.type == "tool_use":
            name = block.name
            if strip_tool_prefix and name.startswith(_MCP_TOOL_PREFIX):
@@ -1316,7 +1384,7 @@ def normalize_anthropic_response(
            tool_calls=tool_calls or None,
            reasoning="\n\n".join(reasoning_parts) if reasoning_parts else None,
            reasoning_content=None,
-            reasoning_details=None,
+            reasoning_details=reasoning_details or None,
        ),
        finish_reason,
    )
@@ -697,6 +697,25 @@ def _read_main_model() -> str:
    return ""


+def _read_main_provider() -> str:
+    """Read the user's configured main provider from config.yaml.
+
+    Returns the lowercase provider id (e.g. "alibaba", "openrouter") or ""
+    if not configured.
+    """
+    try:
+        from hermes_cli.config import load_config
+        cfg = load_config()
+        model_cfg = cfg.get("model", {})
+        if isinstance(model_cfg, dict):
+            provider = model_cfg.get("provider", "")
+            if isinstance(provider, str) and provider.strip():
+                return provider.strip().lower()
+    except Exception:
+        pass
+    return ""
+
+
 def _resolve_custom_runtime() -> Tuple[Optional[str], Optional[str]]:
    """Resolve the active custom/main endpoint the same way the main CLI does.

@@ -855,10 +874,35 @@ _AUTO_PROVIDER_LABELS = {
 }


+_AGGREGATOR_PROVIDERS = frozenset({"openrouter", "nous"})
+
+
 def _resolve_auto() -> Tuple[Optional[OpenAI], Optional[str]]:
-    """Full auto-detection chain: OpenRouter → Nous → custom → Codex → API-key → None."""
+    """Full auto-detection chain.
+
+    Priority:
+      1. If the user's main provider is NOT an aggregator (OpenRouter / Nous),
+         use their main provider + main model directly.  This ensures users on
+         Alibaba, DeepSeek, ZAI, etc. get auxiliary tasks handled by the same
+         provider they already have credentials for — no OpenRouter key needed.
+      2. OpenRouter → Nous → custom → Codex → API-key providers (original chain).
+    """
    global auxiliary_is_nous
    auxiliary_is_nous = False  # Reset — _try_nous() will set True if it wins
+
+    # ── Step 1: non-aggregator main provider → use main model directly ──
+    main_provider = _read_main_provider()
+    main_model = _read_main_model()
+    if (main_provider and main_model
+            and main_provider not in _AGGREGATOR_PROVIDERS
+            and main_provider not in ("auto", "custom", "")):
+        client, resolved = resolve_provider_client(main_provider, main_model)
+        if client is not None:
+            logger.info("Auxiliary auto-detect: using main provider %s (%s)",
+                        main_provider, resolved or main_model)
+            return client, resolved or main_model
+
+    # ── Step 2: aggregator / fallback chain ──────────────────────────────
    tried = []
    for try_fn in (_try_openrouter, _try_nous, _try_custom_endpoint,
                   _try_codex, _resolve_api_key_provider):
@@ -1078,9 +1122,9 @@ def resolve_provider_client(
            tried_sources = list(pconfig.api_key_env_vars)
            if provider == "copilot":
                tried_sources.append("gh auth token")
-            logger.warning("resolve_provider_client: provider %s has no API "
-                           "key configured (tried: %s)",
-                           provider, ", ".join(tried_sources))
+            logger.debug("resolve_provider_client: provider %s has no API "
+                         "key configured (tried: %s)",
+                         provider, ", ".join(tried_sources))
            return None, None

        base_url = str(creds.get("base_url", "")).strip().rstrip("/") or pconfig.inference_base_url
@@ -0,0 +1,113 @@
+"""BuiltinMemoryProvider — wraps MEMORY.md / USER.md as a MemoryProvider.
+
+Always registered as the first provider. Cannot be disabled or removed.
+This is the existing Hermes memory system exposed through the provider
+interface for compatibility with the MemoryManager.
+
+The actual storage logic lives in tools/memory_tool.py (MemoryStore).
+This provider is a thin adapter that delegates to MemoryStore and
+exposes the memory tool schema.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from typing import Any, Dict, List, Optional
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+
+class BuiltinMemoryProvider(MemoryProvider):
+    """Built-in file-backed memory (MEMORY.md + USER.md).
+
+    Always active, never disabled by other providers. The `memory` tool
+    is handled by run_agent.py's agent-level tool interception (not through
+    the normal registry), so get_tool_schemas() returns an empty list —
+    the memory tool is already wired separately.
+    """
+
+    def __init__(
+        self,
+        memory_store=None,
+        memory_enabled: bool = False,
+        user_profile_enabled: bool = False,
+    ):
+        self._store = memory_store
+        self._memory_enabled = memory_enabled
+        self._user_profile_enabled = user_profile_enabled
+
+    @property
+    def name(self) -> str:
+        return "builtin"
+
+    def is_available(self) -> bool:
+        """Built-in memory is always available."""
+        return True
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        """Load memory from disk if not already loaded."""
+        if self._store is not None:
+            self._store.load_from_disk()
+
+    def system_prompt_block(self) -> str:
+        """Return MEMORY.md and USER.md content for the system prompt.
+
+        Uses the frozen snapshot captured at load time. This ensures the
+        system prompt stays stable throughout a session (preserving the
+        prompt cache), even though the live entries may change via tool calls.
+        """
+        if not self._store:
+            return ""
+
+        parts = []
+        if self._memory_enabled:
+            mem_block = self._store.format_for_system_prompt("memory")
+            if mem_block:
+                parts.append(mem_block)
+        if self._user_profile_enabled:
+            user_block = self._store.format_for_system_prompt("user")
+            if user_block:
+                parts.append(user_block)
+
+        return "\n\n".join(parts)
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        """Built-in memory doesn't do query-based recall — it's injected via system_prompt_block."""
+        return ""
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Built-in memory doesn't auto-sync turns — writes happen via the memory tool."""
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        """Return empty list.
+
+        The `memory` tool is an agent-level intercepted tool, handled
+        specially in run_agent.py before normal tool dispatch. It's not
+        part of the standard tool registry. We don't duplicate it here.
+        """
+        return []
+
+    def handle_tool_call(self, tool_name: str, args: Dict[str, Any], **kwargs) -> str:
+        """Not used — the memory tool is intercepted in run_agent.py."""
+        return json.dumps({"error": "Built-in memory tool is handled by the agent loop"})
+
+    def shutdown(self) -> None:
+        """No cleanup needed — files are saved on every write."""
+
+    # -- Property access for backward compatibility --------------------------
+
+    @property
+    def store(self):
+        """Access the underlying MemoryStore for legacy code paths."""
+        return self._store
+
+    @property
+    def memory_enabled(self) -> bool:
+        return self._memory_enabled
+
+    @property
+    def user_profile_enabled(self) -> bool:
+        return self._user_profile_enabled
@@ -303,6 +303,43 @@ class CredentialPool:
        self._persist()
        return updated

+    def _sync_anthropic_entry_from_credentials_file(self, entry: PooledCredential) -> PooledCredential:
+        """Sync a claude_code pool entry from ~/.claude/.credentials.json if tokens differ.
+
+        OAuth refresh tokens are single-use. When something external (e.g.
+        Claude Code CLI, or another profile's pool) refreshes the token, it
+        writes the new pair to ~/.claude/.credentials.json. The pool entry's
+        refresh token becomes stale. This method detects that and syncs.
+        """
+        if self.provider != "anthropic" or entry.source != "claude_code":
+            return entry
+        try:
+            from agent.anthropic_adapter import read_claude_code_credentials
+            creds = read_claude_code_credentials()
+            if not creds:
+                return entry
+            file_refresh = creds.get("refreshToken", "")
+            file_access = creds.get("accessToken", "")
+            file_expires = creds.get("expiresAt", 0)
+            # If the credentials file has a different token pair, sync it
+            if file_refresh and file_refresh != entry.refresh_token:
+                logger.debug("Pool entry %s: syncing tokens from credentials file (refresh token changed)", entry.id)
+                updated = replace(
+                    entry,
+                    access_token=file_access,
+                    refresh_token=file_refresh,
+                    expires_at_ms=file_expires,
+                    last_status=None,
+                    last_status_at=None,
+                    last_error_code=None,
+                )
+                self._replace_entry(entry, updated)
+                self._persist()
+                return updated
+        except Exception as exc:
+            logger.debug("Failed to sync from credentials file: %s", exc)
+        return entry
+
    def _refresh_entry(self, entry: PooledCredential, *, force: bool) -> Optional[PooledCredential]:
        if entry.auth_type != AUTH_TYPE_OAUTH or not entry.refresh_token:
            if force:
@@ -323,6 +360,19 @@ class CredentialPool:
                    refresh_token=refreshed["refresh_token"],
                    expires_at_ms=refreshed["expires_at_ms"],
                )
+                # Keep ~/.claude/.credentials.json in sync so that the
+                # fallback path (resolve_anthropic_token) and other profiles
+                # see the latest tokens.
+                if entry.source == "claude_code":
+                    try:
+                        from agent.anthropic_adapter import _write_claude_code_credentials
+                        _write_claude_code_credentials(
+                            refreshed["access_token"],
+                            refreshed["refresh_token"],
+                            refreshed["expires_at_ms"],
+                        )
+                    except Exception as wexc:
+                        logger.debug("Failed to write refreshed token to credentials file: %s", wexc)
            elif self.provider == "openai-codex":
                refreshed = auth_mod.refresh_codex_oauth_pure(
                    entry.access_token,
@@ -369,6 +419,46 @@ class CredentialPool:
                return entry
        except Exception as exc:
            logger.debug("Credential refresh failed for %s/%s: %s", self.provider, entry.id, exc)
+            # For anthropic claude_code entries: the refresh token may have been
+            # consumed by another process. Check if ~/.claude/.credentials.json
+            # has a newer token pair and retry once.
+            if self.provider == "anthropic" and entry.source == "claude_code":
+                synced = self._sync_anthropic_entry_from_credentials_file(entry)
+                if synced.refresh_token != entry.refresh_token:
+                    logger.debug("Retrying refresh with synced token from credentials file")
+                    try:
+                        from agent.anthropic_adapter import refresh_anthropic_oauth_pure
+                        refreshed = refresh_anthropic_oauth_pure(
+                            synced.refresh_token,
+                            use_json=synced.source.endswith("hermes_pkce"),
+                        )
+                        updated = replace(
+                            synced,
+                            access_token=refreshed["access_token"],
+                            refresh_token=refreshed["refresh_token"],
+                            expires_at_ms=refreshed["expires_at_ms"],
+                            last_status=STATUS_OK,
+                            last_status_at=None,
+                            last_error_code=None,
+                        )
+                        self._replace_entry(synced, updated)
+                        self._persist()
+                        try:
+                            from agent.anthropic_adapter import _write_claude_code_credentials
+                            _write_claude_code_credentials(
+                                refreshed["access_token"],
+                                refreshed["refresh_token"],
+                                refreshed["expires_at_ms"],
+                            )
+                        except Exception as wexc:
+                            logger.debug("Failed to write refreshed token to credentials file (retry path): %s", wexc)
+                        return updated
+                    except Exception as retry_exc:
+                        logger.debug("Retry refresh also failed: %s", retry_exc)
+                elif not self._entry_needs_refresh(synced):
+                    # Credentials file had a valid (non-expired) token — use it directly
+                    logger.debug("Credentials file has valid token, using without refresh")
+                    return synced
            self._mark_exhausted(entry, None)
            return None

@@ -422,6 +512,15 @@ class CredentialPool:
        cleared_any = False
        available: List[PooledCredential] = []
        for entry in self._entries:
+            # For anthropic claude_code entries, sync from the credentials file
+            # before any status/refresh checks. This picks up tokens refreshed
+            # by other processes (Claude Code CLI, other Hermes profiles).
+            if (self.provider == "anthropic" and entry.source == "claude_code"
+                    and entry.last_status == STATUS_EXHAUSTED):
+                synced = self._sync_anthropic_entry_from_credentials_file(entry)
+                if synced is not entry:
+                    entry = synced
+                    cleared_any = True
            if entry.last_status == STATUS_EXHAUSTED:
                ttl = _exhausted_ttl(entry.last_error_code)
                if entry.last_status_at and now - entry.last_status_at < ttl:
@@ -0,0 +1,335 @@
+"""MemoryManager — orchestrates the built-in memory provider plus at most
+ONE external plugin memory provider.
+
+Single integration point in run_agent.py. Replaces scattered per-backend
+code with one manager that delegates to registered providers.
+
+The BuiltinMemoryProvider is always registered first and cannot be removed.
+Only ONE external (non-builtin) provider is allowed at a time — attempting
+to register a second external provider is rejected with a warning.  This
+prevents tool schema bloat and conflicting memory backends.
+
+Usage in run_agent.py:
+    self._memory_manager = MemoryManager()
+    self._memory_manager.add_provider(BuiltinMemoryProvider(...))
+    # Only ONE of these:
+    self._memory_manager.add_provider(plugin_provider)
+
+    # System prompt
+    prompt_parts.append(self._memory_manager.build_system_prompt())
+
+    # Pre-turn
+    context = self._memory_manager.prefetch_all(user_message)
+
+    # Post-turn
+    self._memory_manager.sync_all(user_msg, assistant_response)
+    self._memory_manager.queue_prefetch_all(user_msg)
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from typing import Any, Dict, List, Optional
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+
+class MemoryManager:
+    """Orchestrates the built-in provider plus at most one external provider.
+
+    The builtin provider is always first. Only one non-builtin (external)
+    provider is allowed.  Failures in one provider never block the other.
+    """
+
+    def __init__(self) -> None:
+        self._providers: List[MemoryProvider] = []
+        self._tool_to_provider: Dict[str, MemoryProvider] = {}
+        self._has_external: bool = False  # True once a non-builtin provider is added
+
+    # -- Registration --------------------------------------------------------
+
+    def add_provider(self, provider: MemoryProvider) -> None:
+        """Register a memory provider.
+
+        Built-in provider (name ``"builtin"``) is always accepted.
+        Only **one** external (non-builtin) provider is allowed — a second
+        attempt is rejected with a warning.
+        """
+        is_builtin = provider.name == "builtin"
+
+        if not is_builtin:
+            if self._has_external:
+                existing = next(
+                    (p.name for p in self._providers if p.name != "builtin"), "unknown"
+                )
+                logger.warning(
+                    "Rejected memory provider '%s' — external provider '%s' is "
+                    "already registered. Only one external memory provider is "
+                    "allowed at a time. Configure which one via memory.provider "
+                    "in config.yaml.",
+                    provider.name, existing,
+                )
+                return
+            self._has_external = True
+
+        self._providers.append(provider)
+
+        # Index tool names → provider for routing
+        for schema in provider.get_tool_schemas():
+            tool_name = schema.get("name", "")
+            if tool_name and tool_name not in self._tool_to_provider:
+                self._tool_to_provider[tool_name] = provider
+            elif tool_name in self._tool_to_provider:
+                logger.warning(
+                    "Memory tool name conflict: '%s' already registered by %s, "
+                    "ignoring from %s",
+                    tool_name,
+                    self._tool_to_provider[tool_name].name,
+                    provider.name,
+                )
+
+        logger.info(
+            "Memory provider '%s' registered (%d tools)",
+            provider.name,
+            len(provider.get_tool_schemas()),
+        )
+
+    @property
+    def providers(self) -> List[MemoryProvider]:
+        """All registered providers in order."""
+        return list(self._providers)
+
+    @property
+    def provider_names(self) -> List[str]:
+        """Names of all registered providers."""
+        return [p.name for p in self._providers]
+
+    def get_provider(self, name: str) -> Optional[MemoryProvider]:
+        """Get a provider by name, or None if not registered."""
+        for p in self._providers:
+            if p.name == name:
+                return p
+        return None
+
+    # -- System prompt -------------------------------------------------------
+
+    def build_system_prompt(self) -> str:
+        """Collect system prompt blocks from all providers.
+
+        Returns combined text, or empty string if no providers contribute.
+        Each non-empty block is labeled with the provider name.
+        """
+        blocks = []
+        for provider in self._providers:
+            try:
+                block = provider.system_prompt_block()
+                if block and block.strip():
+                    blocks.append(block)
+            except Exception as e:
+                logger.warning(
+                    "Memory provider '%s' system_prompt_block() failed: %s",
+                    provider.name, e,
+                )
+        return "\n\n".join(blocks)
+
+    # -- Prefetch / recall ---------------------------------------------------
+
+    def prefetch_all(self, query: str, *, session_id: str = "") -> str:
+        """Collect prefetch context from all providers.
+
+        Returns merged context text labeled by provider. Empty providers
+        are skipped. Failures in one provider don't block others.
+        """
+        parts = []
+        for provider in self._providers:
+            try:
+                result = provider.prefetch(query, session_id=session_id)
+                if result and result.strip():
+                    parts.append(result)
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' prefetch failed (non-fatal): %s",
+                    provider.name, e,
+                )
+        return "\n\n".join(parts)
+
+    def queue_prefetch_all(self, query: str, *, session_id: str = "") -> None:
+        """Queue background prefetch on all providers for the next turn."""
+        for provider in self._providers:
+            try:
+                provider.queue_prefetch(query, session_id=session_id)
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' queue_prefetch failed (non-fatal): %s",
+                    provider.name, e,
+                )
+
+    # -- Sync ----------------------------------------------------------------
+
+    def sync_all(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Sync a completed turn to all providers."""
+        for provider in self._providers:
+            try:
+                provider.sync_turn(user_content, assistant_content, session_id=session_id)
+            except Exception as e:
+                logger.warning(
+                    "Memory provider '%s' sync_turn failed: %s",
+                    provider.name, e,
+                )
+
+    # -- Tools ---------------------------------------------------------------
+
+    def get_all_tool_schemas(self) -> List[Dict[str, Any]]:
+        """Collect tool schemas from all providers."""
+        schemas = []
+        seen = set()
+        for provider in self._providers:
+            try:
+                for schema in provider.get_tool_schemas():
+                    name = schema.get("name", "")
+                    if name and name not in seen:
+                        schemas.append(schema)
+                        seen.add(name)
+            except Exception as e:
+                logger.warning(
+                    "Memory provider '%s' get_tool_schemas() failed: %s",
+                    provider.name, e,
+                )
+        return schemas
+
+    def get_all_tool_names(self) -> set:
+        """Return set of all tool names across all providers."""
+        return set(self._tool_to_provider.keys())
+
+    def has_tool(self, tool_name: str) -> bool:
+        """Check if any provider handles this tool."""
+        return tool_name in self._tool_to_provider
+
+    def handle_tool_call(
+        self, tool_name: str, args: Dict[str, Any], **kwargs
+    ) -> str:
+        """Route a tool call to the correct provider.
+
+        Returns JSON string result. Raises ValueError if no provider
+        handles the tool.
+        """
+        provider = self._tool_to_provider.get(tool_name)
+        if provider is None:
+            return json.dumps({"error": f"No memory provider handles tool '{tool_name}'"})
+        try:
+            return provider.handle_tool_call(tool_name, args, **kwargs)
+        except Exception as e:
+            logger.error(
+                "Memory provider '%s' handle_tool_call(%s) failed: %s",
+                provider.name, tool_name, e,
+            )
+            return json.dumps({"error": f"Memory tool '{tool_name}' failed: {e}"})
+
+    # -- Lifecycle hooks -----------------------------------------------------
+
+    def on_turn_start(self, turn_number: int, message: str, **kwargs) -> None:
+        """Notify all providers of a new turn.
+
+        kwargs may include: remaining_tokens, model, platform, tool_count.
+        """
+        for provider in self._providers:
+            try:
+                provider.on_turn_start(turn_number, message, **kwargs)
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' on_turn_start failed: %s",
+                    provider.name, e,
+                )
+
+    def on_session_end(self, messages: List[Dict[str, Any]]) -> None:
+        """Notify all providers of session end."""
+        for provider in self._providers:
+            try:
+                provider.on_session_end(messages)
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' on_session_end failed: %s",
+                    provider.name, e,
+                )
+
+    def on_pre_compress(self, messages: List[Dict[str, Any]]) -> str:
+        """Notify all providers before context compression.
+
+        Returns combined text from providers to include in the compression
+        summary prompt. Empty string if no provider contributes.
+        """
+        parts = []
+        for provider in self._providers:
+            try:
+                result = provider.on_pre_compress(messages)
+                if result and result.strip():
+                    parts.append(result)
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' on_pre_compress failed: %s",
+                    provider.name, e,
+                )
+        return "\n\n".join(parts)
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        """Notify external providers when the built-in memory tool writes.
+
+        Skips the builtin provider itself (it's the source of the write).
+        """
+        for provider in self._providers:
+            if provider.name == "builtin":
+                continue
+            try:
+                provider.on_memory_write(action, target, content)
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' on_memory_write failed: %s",
+                    provider.name, e,
+                )
+
+    def on_delegation(self, task: str, result: str, *,
+                      child_session_id: str = "", **kwargs) -> None:
+        """Notify all providers that a subagent completed."""
+        for provider in self._providers:
+            try:
+                provider.on_delegation(
+                    task, result, child_session_id=child_session_id, **kwargs
+                )
+            except Exception as e:
+                logger.debug(
+                    "Memory provider '%s' on_delegation failed: %s",
+                    provider.name, e,
+                )
+
+    def shutdown_all(self) -> None:
+        """Shut down all providers (reverse order for clean teardown)."""
+        for provider in reversed(self._providers):
+            try:
+                provider.shutdown()
+            except Exception as e:
+                logger.warning(
+                    "Memory provider '%s' shutdown failed: %s",
+                    provider.name, e,
+                )
+
+    def initialize_all(self, session_id: str, **kwargs) -> None:
+        """Initialize all providers.
+
+        Automatically injects ``hermes_home`` into *kwargs* so that every
+        provider can resolve profile-scoped storage paths without importing
+        ``get_hermes_home()`` themselves.
+        """
+        if "hermes_home" not in kwargs:
+            from hermes_constants import get_hermes_home
+            kwargs["hermes_home"] = str(get_hermes_home())
+        for provider in self._providers:
+            try:
+                provider.initialize(session_id=session_id, **kwargs)
+            except Exception as e:
+                logger.warning(
+                    "Memory provider '%s' initialize failed: %s",
+                    provider.name, e,
+                )
@@ -0,0 +1,231 @@
+"""Abstract base class for pluggable memory providers.
+
+Memory providers give the agent persistent recall across sessions. One
+external provider is active at a time alongside the always-on built-in
+memory (MEMORY.md / USER.md). The MemoryManager enforces this limit.
+
+Built-in memory is always active as the first provider and cannot be removed.
+External providers (Honcho, Hindsight, Mem0, etc.) are additive — they never
+disable the built-in store. Only one external provider runs at a time to
+prevent tool schema bloat and conflicting memory backends.
+
+Registration:
+  1. Built-in: BuiltinMemoryProvider — always present, not removable.
+  2. Plugins: Ship in plugins/memory/<name>/, activated by memory.provider config.
+
+Lifecycle (called by MemoryManager, wired in run_agent.py):
+  initialize()          — connect, create resources, warm up
+  system_prompt_block()  — static text for the system prompt
+  prefetch(query)        — background recall before each turn
+  sync_turn(user, asst)  — async write after each turn
+  get_tool_schemas()     — tool schemas to expose to the model
+  handle_tool_call()     — dispatch a tool call
+  shutdown()             — clean exit
+
+Optional hooks (override to opt in):
+  on_turn_start(turn, message, **kwargs) — per-turn tick with runtime context
+  on_session_end(messages)               — end-of-session extraction
+  on_pre_compress(messages) -> str       — extract before context compression
+  on_memory_write(action, target, content) — mirror built-in memory writes
+  on_delegation(task, result, **kwargs)  — parent-side observation of subagent work
+"""
+
+from __future__ import annotations
+
+import logging
+from abc import ABC, abstractmethod
+from typing import Any, Dict, List, Optional
+
+logger = logging.getLogger(__name__)
+
+
+class MemoryProvider(ABC):
+    """Abstract base class for memory providers."""
+
+    @property
+    @abstractmethod
+    def name(self) -> str:
+        """Short identifier for this provider (e.g. 'builtin', 'honcho', 'hindsight')."""
+
+    # -- Core lifecycle (implement these) ------------------------------------
+
+    @abstractmethod
+    def is_available(self) -> bool:
+        """Return True if this provider is configured, has credentials, and is ready.
+
+        Called during agent init to decide whether to activate the provider.
+        Should not make network calls — just check config and installed deps.
+        """
+
+    @abstractmethod
+    def initialize(self, session_id: str, **kwargs) -> None:
+        """Initialize for a session.
+
+        Called once at agent startup. May create resources (banks, tables),
+        establish connections, start background threads, etc.
+
+        kwargs always include:
+          - hermes_home (str): The active HERMES_HOME directory path. Use this
+            for profile-scoped storage instead of hardcoding ``~/.hermes``.
+          - platform (str): "cli", "telegram", "discord", "cron", etc.
+
+        kwargs may also include:
+          - agent_context (str): "primary", "subagent", "cron", or "flush".
+            Providers should skip writes for non-primary contexts (cron system
+            prompts would corrupt user representations).
+          - agent_identity (str): Profile name (e.g. "coder"). Use for
+            per-profile provider identity scoping.
+          - agent_workspace (str): Shared workspace name (e.g. "hermes").
+          - parent_session_id (str): For subagents, the parent's session_id.
+          - user_id (str): Platform user identifier (gateway sessions).
+        """
+
+    def system_prompt_block(self) -> str:
+        """Return text to include in the system prompt.
+
+        Called during system prompt assembly. Return empty string to skip.
+        This is for STATIC provider info (instructions, status). Prefetched
+        recall context is injected separately via prefetch().
+        """
+        return ""
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        """Recall relevant context for the upcoming turn.
+
+        Called before each API call. Return formatted text to inject as
+        context, or empty string if nothing relevant. Implementations
+        should be fast — use background threads for the actual recall
+        and return cached results here.
+
+        session_id is provided for providers serving concurrent sessions
+        (gateway group chats, cached agents). Providers that don't need
+        per-session scoping can ignore it.
+        """
+        return ""
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        """Queue a background recall for the NEXT turn.
+
+        Called after each turn completes. The result will be consumed
+        by prefetch() on the next turn. Default is no-op — providers
+        that do background prefetching should override this.
+        """
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Persist a completed turn to the backend.
+
+        Called after each turn. Should be non-blocking — queue for
+        background processing if the backend has latency.
+        """
+
+    @abstractmethod
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        """Return tool schemas this provider exposes.
+
+        Each schema follows the OpenAI function calling format:
+        {"name": "...", "description": "...", "parameters": {...}}
+
+        Return empty list if this provider has no tools (context-only).
+        """
+
+    def handle_tool_call(self, tool_name: str, args: Dict[str, Any], **kwargs) -> str:
+        """Handle a tool call for one of this provider's tools.
+
+        Must return a JSON string (the tool result).
+        Only called for tool names returned by get_tool_schemas().
+        """
+        raise NotImplementedError(f"Provider {self.name} does not handle tool {tool_name}")
+
+    def shutdown(self) -> None:
+        """Clean shutdown — flush queues, close connections."""
+
+    # -- Optional hooks (override to opt in) ---------------------------------
+
+    def on_turn_start(self, turn_number: int, message: str, **kwargs) -> None:
+        """Called at the start of each turn with the user message.
+
+        Use for turn-counting, scope management, periodic maintenance.
+
+        kwargs may include: remaining_tokens, model, platform, tool_count.
+        Providers use what they need; extras are ignored.
+        """
+
+    def on_session_end(self, messages: List[Dict[str, Any]]) -> None:
+        """Called when a session ends (explicit exit or timeout).
+
+        Use for end-of-session fact extraction, summarization, etc.
+        messages is the full conversation history.
+
+        NOT called after every turn — only at actual session boundaries
+        (CLI exit, /reset, gateway session expiry).
+        """
+
+    def on_pre_compress(self, messages: List[Dict[str, Any]]) -> str:
+        """Called before context compression discards old messages.
+
+        Use to extract insights from messages about to be compressed.
+        messages is the list that will be summarized/discarded.
+
+        Return text to include in the compression summary prompt so the
+        compressor preserves provider-extracted insights. Return empty
+        string for no contribution (backwards-compatible default).
+        """
+        return ""
+
+    def on_delegation(self, task: str, result: str, *,
+                      child_session_id: str = "", **kwargs) -> None:
+        """Called on the PARENT agent when a subagent completes.
+
+        The parent's memory provider gets the task+result pair as an
+        observation of what was delegated and what came back. The subagent
+        itself has no provider session (skip_memory=True).
+
+        task: the delegation prompt
+        result: the subagent's final response
+        child_session_id: the subagent's session_id
+        """
+
+    def get_config_schema(self) -> List[Dict[str, Any]]:
+        """Return config fields this provider needs for setup.
+
+        Used by 'hermes memory setup' to walk the user through configuration.
+        Each field is a dict with:
+          key:         config key name (e.g. 'api_key', 'mode')
+          description: human-readable description
+          secret:      True if this should go to .env (default: False)
+          required:    True if required (default: False)
+          default:     default value (optional)
+          choices:     list of valid values (optional)
+          url:         URL where user can get this credential (optional)
+          env_var:     explicit env var name for secrets (default: auto-generated)
+
+        Return empty list if no config needed (e.g. local-only providers).
+        """
+        return []
+
+    def save_config(self, values: Dict[str, Any], hermes_home: str) -> None:
+        """Write non-secret config to the provider's native location.
+
+        Called by 'hermes memory setup' after collecting user inputs.
+        ``values`` contains only non-secret fields (secrets go to .env).
+        ``hermes_home`` is the active HERMES_HOME directory path.
+
+        Providers with native config files (JSON, YAML) should override
+        this to write to their expected location. Providers that use only
+        env vars can leave the default (no-op).
+
+        All new memory provider plugins MUST implement either:
+        - save_config() for native config file formats, OR
+        - use only env vars (in which case get_config_schema() fields
+          should all have ``env_var`` set and this method stays no-op).
+        """
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        """Called when the built-in memory tool writes an entry.
+
+        action: 'add', 'replace', or 'remove'
+        target: 'memory' or 'user'
+        content: the entry content
+
+        Use to mirror built-in memory writes to your backend.
+        """
@@ -113,6 +113,8 @@ DEFAULT_CONTEXT_LENGTHS = {
    "glm": 202752,
    # Kimi
    "kimi": 262144,
+    # Arcee
+    "trinity": 262144,
    # Hugging Face Inference Providers — model IDs use org/name format
    "Qwen/Qwen3.5-397B-A17B": 131072,
    "Qwen/Qwen3.5-35B-A3B": 131072,
@@ -121,6 +123,8 @@ DEFAULT_CONTEXT_LENGTHS = {
    "moonshotai/Kimi-K2-Thinking": 262144,
    "MiniMaxAI/MiniMax-M2.5": 204800,
    "XiaomiMiMo/MiMo-V2-Flash": 32768,
+    "mimo-v2-pro": 1048576,
+    "mimo-v2-omni": 1048576,
    "zai-org/GLM-5": 202752,
 }

@@ -187,7 +187,36 @@ TOOL_USE_ENFORCEMENT_GUIDANCE = (

 # Model name substrings that trigger tool-use enforcement guidance.
 # Add new patterns here when a model family needs explicit steering.
-TOOL_USE_ENFORCEMENT_MODELS = ("gpt", "codex")
+TOOL_USE_ENFORCEMENT_MODELS = ("gpt", "codex", "gemini", "gemma")
+
+# Gemini/Gemma-specific operational guidance, adapted from OpenCode's gemini.txt.
+# Injected alongside TOOL_USE_ENFORCEMENT_GUIDANCE when the model is Gemini or Gemma.
+GOOGLE_MODEL_OPERATIONAL_GUIDANCE = (
+    "# Google model operational directives\n"
+    "Follow these operational rules strictly:\n"
+    "- **Absolute paths:** Always construct and use absolute file paths for all "
+    "file system operations. Combine the project root with relative paths.\n"
+    "- **Verify first:** Use read_file/search_files to check file contents and "
+    "project structure before making changes. Never guess at file contents.\n"
+    "- **Dependency checks:** Never assume a library is available. Check "
+    "package.json, requirements.txt, Cargo.toml, etc. before importing.\n"
+    "- **Conciseness:** Keep explanatory text brief — a few sentences, not "
+    "paragraphs. Focus on actions and results over narration.\n"
+    "- **Parallel tool calls:** When you need to perform multiple independent "
+    "operations (e.g. reading several files), make all the tool calls in a "
+    "single response rather than sequentially.\n"
+    "- **Non-interactive commands:** Use flags like -y, --yes, --non-interactive "
+    "to prevent CLI tools from hanging on prompts.\n"
+    "- **Keep going:** Work autonomously until the task is fully resolved. "
+    "Don't stop with a plan — execute it.\n"
+)
+
+# Model name substrings that should use the 'developer' role instead of
+# 'system' for the system prompt.  OpenAI's newer models (GPT-5, Codex)
+# give stronger instruction-following weight to the 'developer' role.
+# The swap happens at the API boundary in _build_api_kwargs() so internal
+# message representation stays consistent ("system" everywhere).
+DEVELOPER_ROLE_MODELS = ("gpt-5", "codex")

 PLATFORM_HINTS = {
    "whatsapp": (
@@ -459,11 +488,19 @@ def build_skills_system_prompt(
        return ""

    # ── Layer 1: in-process LRU cache ─────────────────────────────────
+    # Include the resolved platform so per-platform disabled-skill lists
+    # produce distinct cache entries (gateway serves multiple platforms).
+    _platform_hint = (
+        os.environ.get("HERMES_PLATFORM")
+        or os.environ.get("HERMES_SESSION_PLATFORM")
+        or ""
+    )
    cache_key = (
        str(skills_dir.resolve()),
        tuple(str(d) for d in external_dirs),
        tuple(sorted(str(t) for t in (available_tools or set()))),
        tuple(sorted(str(ts) for ts in (available_toolsets or set()))),
+        _platform_hint,
    )
    with _SKILLS_PROMPT_CACHE_LOCK:
        cached = _SKILLS_PROMPT_CACHE.get(cache_key)
@@ -645,6 +682,73 @@ def build_skills_system_prompt(
    return result


+def build_nous_subscription_prompt(valid_tool_names: "set[str] | None" = None) -> str:
+    """Build a compact Nous subscription capability block for the system prompt."""
+    try:
+        from hermes_cli.nous_subscription import get_nous_subscription_features
+        from tools.tool_backend_helpers import managed_nous_tools_enabled
+    except Exception as exc:
+        logger.debug("Failed to import Nous subscription helper: %s", exc)
+        return ""
+
+    if not managed_nous_tools_enabled():
+        return ""
+
+    valid_names = set(valid_tool_names or set())
+    relevant_tool_names = {
+        "web_search",
+        "web_extract",
+        "browser_navigate",
+        "browser_snapshot",
+        "browser_click",
+        "browser_type",
+        "browser_scroll",
+        "browser_console",
+        "browser_close",
+        "browser_press",
+        "browser_get_images",
+        "browser_vision",
+        "image_generate",
+        "text_to_speech",
+        "terminal",
+        "process",
+        "execute_code",
+    }
+
+    if valid_names and not (valid_names & relevant_tool_names):
+        return ""
+
+    features = get_nous_subscription_features()
+
+    def _status_line(feature) -> str:
+        if feature.managed_by_nous:
+            return f"- {feature.label}: active via Nous subscription"
+        if feature.active:
+            current = feature.current_provider or "configured provider"
+            return f"- {feature.label}: currently using {current}"
+        if feature.included_by_default and features.nous_auth_present:
+            return f"- {feature.label}: included with Nous subscription, not currently selected"
+        if feature.key == "modal" and features.nous_auth_present:
+            return f"- {feature.label}: optional via Nous subscription"
+        return f"- {feature.label}: not currently available"
+
+    lines = [
+        "# Nous Subscription",
+        "Nous subscription includes managed web tools (Firecrawl), image generation (FAL), OpenAI TTS, and browser automation (Browserbase) by default. Modal execution is optional.",
+        "Current capability status:",
+    ]
+    lines.extend(_status_line(feature) for feature in features.items())
+    lines.extend(
+        [
+            "When a Nous-managed feature is active, do not ask the user for Firecrawl, FAL, OpenAI TTS, or Browserbase API keys.",
+            "If the user is not subscribed and asks for a capability that Nous subscription would unlock or simplify, suggest Nous subscription as one option alongside direct setup or local alternatives.",
+            "Do not mention subscription unless the user asks about it or it directly solves the current missing capability.",
+            "Useful commands: hermes setup, hermes setup tools, hermes setup terminal, hermes status.",
+        ]
+    )
+    return "\n".join(lines)
+
+
 # =========================================================================
 # Context files (SOUL.md, AGENTS.md, .cursorrules)
 # =========================================================================
@@ -53,8 +53,7 @@ _PREFIX_PATTERNS = [
 # ENV assignment patterns: KEY=value where KEY contains a secret-like name
 _SECRET_ENV_NAMES = r"(?:API_?KEY|TOKEN|SECRET|PASSWORD|PASSWD|CREDENTIAL|AUTH)"
 _ENV_ASSIGN_RE = re.compile(
-    rf"([A-Z_]*{_SECRET_ENV_NAMES}[A-Z_]*)\s*=\s*(['\"]?)(\S+)\2",
-    re.IGNORECASE,
+    rf"([A-Z0-9_]{{0,50}}{_SECRET_ENV_NAMES}[A-Z0-9_]{{0,50}})\s*=\s*(['\"]?)(\S+)\2",
 )

 # JSON field patterns: "apiKey": "value", "token": "value", etc.
@@ -118,12 +118,17 @@ def skill_matches_platform(frontmatter: Dict[str, Any]) -> bool:
 # ── Disabled skills ───────────────────────────────────────────────────────


-def get_disabled_skill_names() -> Set[str]:
+def get_disabled_skill_names(platform: str | None = None) -> Set[str]:
    """Read disabled skill names from config.yaml.

-    Resolves platform from ``HERMES_PLATFORM`` env var, falls back to
-    the global disabled list.  Reads the config file directly (no CLI
-    config imports) to stay lightweight.
+    Args:
+        platform: Explicit platform name (e.g. ``"telegram"``).  When
+            *None*, resolves from ``HERMES_PLATFORM`` or
+            ``HERMES_SESSION_PLATFORM`` env vars.  Falls back to the
+            global disabled list when no platform is determined.
+
+    Reads the config file directly (no CLI config imports) to stay
+    lightweight.
    """
    config_path = get_hermes_home() / "config.yaml"
    if not config_path.exists():
@@ -140,7 +145,11 @@ def get_disabled_skill_names() -> Set[str]:
    if not isinstance(skills_cfg, dict):
        return set()

-    resolved_platform = os.getenv("HERMES_PLATFORM")
+    resolved_platform = (
+        platform
+        or os.getenv("HERMES_PLATFORM")
+        or os.getenv("HERMES_SESSION_PLATFORM")
+    )
    if resolved_platform:
        platform_disabled = (skills_cfg.get("platform_disabled") or {}).get(
            resolved_platform
@@ -230,7 +239,13 @@ def get_all_skills_dirs() -> List[Path]:

 def extract_skill_conditions(frontmatter: Dict[str, Any]) -> Dict[str, List]:
    """Extract conditional activation fields from parsed frontmatter."""
-    hermes = (frontmatter.get("metadata") or {}).get("hermes") or {}
+    metadata = frontmatter.get("metadata")
+    # Handle cases where metadata is not a dict (e.g., a string from malformed YAML)
+    if not isinstance(metadata, dict):
+        metadata = {}
+    hermes = metadata.get("hermes") or {}
+    if not isinstance(hermes, dict):
+        hermes = {}
    return {
        "fallback_for_toolsets": hermes.get("fallback_for_toolsets", []),
        "requires_toolsets": hermes.get("requires_toolsets", []),
@@ -6,6 +6,8 @@ import os
 import re
 from typing import Any, Dict, Optional

+from utils import is_truthy_value
+
 _COMPLEX_KEYWORDS = {
    "debug",
    "debugging",
@@ -47,13 +49,7 @@ _URL_RE = re.compile(r"https?://|www\.", re.IGNORECASE)


 def _coerce_bool(value: Any, default: bool = False) -> bool:
-    if value is None:
-        return default
-    if isinstance(value, bool):
-        return value
-    if isinstance(value, str):
-        return value.strip().lower() in {"1", "true", "yes", "on"}
-    return bool(value)
+    return is_truthy_value(value, default=default)


 def _coerce_int(value: Any, default: int) -> int:
@@ -539,7 +539,7 @@ platform_toolsets:
 #   skills_hub   - skill_hub (search/install/manage from online registries — user-driven only)
 #   moa          - mixture_of_agents  (requires OPENROUTER_API_KEY)
 #   todo         - todo (in-memory task planning, no deps)
-#   tts          - text_to_speech  (Edge TTS free, or ELEVENLABS/OPENAI key)
+#   tts          - text_to_speech  (Edge TTS free, or ELEVENLABS/OPENAI/MINIMAX key)
 #   cronjob      - cronjob (create/list/update/pause/resume/run/remove scheduled tasks)
 #   rl           - rl_list_environments, rl_start_training, etc. (requires TINKER_API_KEY)
 #
@@ -568,7 +568,7 @@ platform_toolsets:
 #   todo         - Task planning and tracking for multi-step work
 #   memory       - Persistent memory across sessions (personal notes + user profile)
 #   session_search - Search and recall past conversations (FTS5 + Gemini Flash summarization)
-#   tts          - Text-to-speech (Edge TTS free, ElevenLabs, OpenAI)
+#   tts          - Text-to-speech (Edge TTS free, ElevenLabs, OpenAI, MiniMax)
 #   cronjob      - Schedule and manage automated tasks (CLI-only)
 #   rl           - RL training tools (Tinker-Atropos)
 #
@@ -144,8 +144,8 @@ def load_cli_config() -> Dict[str, Any]:
    # Default configuration
    defaults = {
        "model": {
-            "default": "anthropic/claude-opus-4.6",
-            "base_url": OPENROUTER_BASE_URL,
+            "default": "",
+            "base_url": "",
            "provider": "auto",
        },
        "terminal": {
@@ -262,6 +262,14 @@ def load_cli_config() -> Dict[str, Any]:
                elif isinstance(file_config["model"], dict):
                    # Old format: model is a dict with default/base_url
                    defaults["model"].update(file_config["model"])
+                    # If the user config sets model.model but not model.default,
+                    # promote model.model to model.default so the user's explicit
+                    # choice isn't shadowed by the hardcoded default.  Without this,
+                    # profile configs that only set "model:" (not "default:") silently
+                    # fall back to claude-opus because the merge preserves the
+                    # hardcoded default and HermesCLI.__init__ checks "default" first.
+                    if "model" in file_config["model"] and "default" not in file_config["model"]:
+                        defaults["model"]["default"] = file_config["model"]["model"]

            # Legacy root-level provider/base_url fallback.
            # Some users (or old code) put provider: / base_url: at the
@@ -500,6 +508,8 @@ from tools.browser_tool import _emergency_cleanup_all_sessions as _cleanup_all_b

 # Guard to prevent cleanup from running multiple times on exit
 _cleanup_done = False
+# Weak reference to the active AIAgent for memory provider shutdown at exit
+_active_agent_ref = None

 def _run_cleanup():
    """Run resource cleanup exactly once."""
@@ -528,6 +538,15 @@ def _run_cleanup():
        shutdown_cached_clients()
    except Exception:
        pass
+    # Shut down memory provider (on_session_end + shutdown_all) at actual
+    # session boundary — NOT per-turn inside run_conversation().
+    try:
+        if _active_agent_ref and hasattr(_active_agent_ref, 'shutdown_memory_provider'):
+            _active_agent_ref.shutdown_memory_provider(
+                getattr(_active_agent_ref, 'conversation_history', None) or []
+            )
+    except Exception:
+        pass


 # =============================================================================
@@ -822,6 +841,63 @@ def _cprint(text: str):
    _pt_print(_PT_ANSI(text))


+# ---------------------------------------------------------------------------
+# File-drop detection — extracted as a pure function for testability.
+# ---------------------------------------------------------------------------
+
+_IMAGE_EXTENSIONS = frozenset({
+    '.png', '.jpg', '.jpeg', '.gif', '.webp',
+    '.bmp', '.tiff', '.tif', '.svg', '.ico',
+})
+
+
+def _detect_file_drop(user_input: str) -> "dict | None":
+    """Detect if *user_input* is a dragged/pasted file path, not a slash command.
+
+    When a user drags a file into the terminal, macOS pastes the absolute path
+    (e.g. ``/Users/roland/Desktop/file.png``) which starts with ``/`` and would
+    otherwise be mistaken for a slash command.
+
+    Returns a dict on match::
+
+        {
+            "path": Path,          # resolved file path
+            "is_image": bool,      # True when suffix is a known image type
+            "remainder": str,      # any text after the path
+        }
+
+    Returns ``None`` when the input is not a real file path.
+    """
+    if not isinstance(user_input, str) or not user_input.startswith("/"):
+        return None
+
+    # Walk the string absorbing backslash-escaped spaces ("\ ").
+    raw = user_input
+    pos = 0
+    while pos < len(raw):
+        ch = raw[pos]
+        if ch == '\\' and pos + 1 < len(raw) and raw[pos + 1] == ' ':
+            pos += 2  # skip escaped space
+        elif ch == ' ':
+            break
+        else:
+            pos += 1
+
+    first_token_raw = raw[:pos]
+    first_token = first_token_raw.replace('\\ ', ' ')
+    drop_path = Path(first_token)
+
+    if not drop_path.exists() or not drop_path.is_file():
+        return None
+
+    remainder = raw[pos:].strip()
+    return {
+        "path": drop_path,
+        "is_image": drop_path.suffix.lower() in _IMAGE_EXTENSIONS,
+        "remainder": remainder,
+    }
+
+
 class ChatConsole:
    """Rich Console adapter for prompt_toolkit's patch_stdout context.

@@ -907,6 +983,28 @@ def _build_compact_banner() -> str:



+# ============================================================================
+# Slash-command detection helper
+# ============================================================================
+
+def _looks_like_slash_command(text: str) -> bool:
+    """Return True if *text* looks like a slash command, not a file path.
+
+    Slash commands are ``/help``, ``/model gpt-4``, ``/q``, etc.
+    File paths like ``/Users/ironin/file.md:45-46 can you fix this?``
+    also start with ``/`` but contain additional ``/`` characters in
+    the first whitespace-delimited word.  This helper distinguishes
+    the two so that pasted paths are sent to the agent instead of
+    triggering "Unknown command".
+    """
+    if not text or not text.startswith("/"):
+        return False
+    first_word = text.split()[0]
+    # After stripping the leading /, a command name has no slashes.
+    # A path like /Users/foo/bar.md always does.
+    return "/" not in first_word[1:]
+
+
 # ============================================================================
 # Skill Slash Commands — dynamic commands generated from installed skills
 # ============================================================================
@@ -1095,7 +1193,7 @@ class HermesCLI:
        # env vars would stomp each other.
        _model_config = CLI_CONFIG.get("model", {})
        _config_model = (_model_config.get("default") or _model_config.get("model") or "") if isinstance(_model_config, dict) else (_model_config or "")
-        _DEFAULT_CONFIG_MODEL = "anthropic/claude-opus-4.6"
+        _DEFAULT_CONFIG_MODEL = ""
        self.model = model or _config_model or _DEFAULT_CONFIG_MODEL
        # Auto-detect model from local server if still on default
        if self.model == _DEFAULT_CONFIG_MODEL:
@@ -1537,6 +1635,28 @@ class HermesCLI:
                pass
            return changed

+        if resolved_provider in {"opencode-zen", "opencode-go"}:
+            try:
+                from hermes_cli.models import normalize_opencode_model_id, opencode_model_api_mode
+
+                canonical = normalize_opencode_model_id(resolved_provider, current_model)
+                if canonical and canonical != current_model:
+                    if not self._model_is_default:
+                        self.console.print(
+                            f"[yellow]⚠️  Stripped provider prefix from '{current_model}'; using '{canonical}' for {resolved_provider}.[/]"
+                        )
+                    self.model = canonical
+                    current_model = canonical
+                    changed = True
+
+                resolved_mode = opencode_model_api_mode(resolved_provider, current_model)
+                if resolved_mode != self.api_mode:
+                    self.api_mode = resolved_mode
+                    changed = True
+            except Exception:
+                pass
+            return changed
+
        if resolved_provider != "openai-codex":
            return False

@@ -1979,10 +2099,12 @@ class HermesCLI:
                    base_url, _source,
                )
            else:
-                self.console.print("[bold red]Provider resolver returned an empty API key.[/]")
+                print("\n⚠️  Provider resolver returned an empty API key. "
+                      "Set OPENROUTER_API_KEY or run: hermes setup")
                return False
        if not isinstance(base_url, str) or not base_url:
-            self.console.print("[bold red]Provider resolver returned an empty base URL.[/]")
+            print("\n⚠️  Provider resolver returned an empty base URL. "
+                  "Check your provider config or run: hermes setup")
            return False

        credentials_changed = api_key != self.api_key or base_url != self.base_url
@@ -2066,6 +2188,7 @@ class HermesCLI:
                return False
            restored = self._session_db.get_messages_as_conversation(self.session_id)
            if restored:
+                restored = [m for m in restored if m.get("role") != "session_meta"]
                self.conversation_history = restored
                msg_count = len([m for m in restored if m.get("role") == "user"])
                title_part = ""
@@ -2129,7 +2252,7 @@ class HermesCLI:
                session_db=self._session_db,
                clarify_callback=self._clarify_callback,
                reasoning_callback=self._current_reasoning_callback(),
-                honcho_session_key=None,  # resolved by run_agent via config sessions map / title
+
                fallback_model=self._fallback_model,
                thinking_callback=self._on_thinking,
                checkpoints_enabled=self.checkpoints_enabled,
@@ -2141,6 +2264,9 @@ class HermesCLI:
                stream_delta_callback=self._stream_delta if self.streaming_enabled else None,
                tool_gen_callback=self._on_tool_gen_start if self.streaming_enabled else None,
            )
+            # Store reference for atexit memory provider shutdown
+            global _active_agent_ref
+            _active_agent_ref = self.agent
            # Route agent status output through prompt_toolkit so ANSI escape
            # sequences aren't garbled by patch_stdout's StdoutProxy (#2262).
            self.agent._print_fn = _cprint
@@ -2258,6 +2384,7 @@ class HermesCLI:

        restored = self._session_db.get_messages_as_conversation(self.session_id)
        if restored:
+            restored = [m for m in restored if m.get("role") != "session_meta"]
            self.conversation_history = restored
            msg_count = len([m for m in restored if m.get("role") == "user"])
            title_part = ""
@@ -2949,10 +3076,54 @@ class HermesCLI:
        print(f"  Config File: {config_path} {config_status}")
        print()
    
+    def _list_recent_sessions(self, limit: int = 10) -> list[dict[str, Any]]:
+        """Return recent CLI sessions for in-chat browsing/resume affordances."""
+        if not self._session_db:
+            return []
+        try:
+            sessions = self._session_db.list_sessions_rich(
+                source="cli",
+                exclude_sources=["tool"],
+                limit=limit,
+            )
+        except Exception:
+            return []
+        return [s for s in sessions if s.get("id") != self.session_id]
+
+    def _show_recent_sessions(self, *, reason: str = "history", limit: int = 10) -> bool:
+        """Render recent sessions inline from the active chat TUI.
+
+        Returns True when something was shown, False if no session list was available.
+        """
+        sessions = self._list_recent_sessions(limit=limit)
+        if not sessions:
+            return False
+
+        from hermes_cli.main import _relative_time
+
+        print()
+        if reason == "history":
+            print("(._.) No messages in the current chat yet — here are recent sessions you can resume:")
+        else:
+            print("  Recent sessions:")
+        print()
+        print(f"  {'Title':<32} {'Preview':<40} {'Last Active':<13} {'ID'}")
+        print(f"  {'─' * 32} {'─' * 40} {'─' * 13} {'─' * 24}")
+        for session in sessions:
+            title = (session.get("title") or "—")[:30]
+            preview = (session.get("preview") or "")[:38]
+            last_active = _relative_time(session.get("last_active"))
+            print(f"  {title:<32} {preview:<40} {last_active:<13} {session['id']}")
+        print()
+        print("  Use /resume <session id or title> to continue where you left off.")
+        print()
+        return True
+
    def show_history(self):
        """Display conversation history."""
        if not self.conversation_history:
-            print("(._.) No conversation history yet.")
+            if not self._show_recent_sessions(reason="history"):
+                print("(._.) No conversation history yet.")
            return

        preview_limit = 400
@@ -3077,6 +3248,8 @@ class HermesCLI:

        if not target:
            _cprint("  Usage: /resume <session_id_or_title>")
+            if self._show_recent_sessions(reason="resume"):
+                return
            _cprint("  Tip:   Use /history or `hermes sessions list` to find sessions.")
            return

@@ -3110,9 +3283,10 @@ class HermesCLI:
        self._resumed = True
        self._pending_title = None

-        # Load conversation history
+        # Load conversation history (strip transcript-only metadata entries)
        restored = self._session_db.get_messages_as_conversation(target_id)
-        self.conversation_history = restored or []
+        restored = [m for m in (restored or []) if m.get("role") != "session_meta"]
+        self.conversation_history = restored

        # Re-open the target session so it's not marked as ended
        try:
@@ -3146,8 +3320,122 @@ class HermesCLI:
        else:
            _cprint(f"  ↻ Resumed session {target_id}{title_part} — no messages, starting fresh.")

+    def _handle_branch_command(self, cmd_original: str) -> None:
+        """Handle /branch [name] — fork the current session into a new independent copy.
+
+        Copies the full conversation history to a new session so the user can
+        explore a different approach without losing the original session state.
+        Inspired by Claude Code's /branch command.
+        """
+        if not self.conversation_history:
+            _cprint("  No conversation to branch — send a message first.")
+            return
+
+        if not self._session_db:
+            _cprint("  Session database not available.")
+            return
+
+        parts = cmd_original.split(None, 1)
+        branch_name = parts[1].strip() if len(parts) > 1 else ""
+
+        # Generate the new session ID
+        now = datetime.now()
+        timestamp_str = now.strftime("%Y%m%d_%H%M%S")
+        short_uuid = uuid.uuid4().hex[:6]
+        new_session_id = f"{timestamp_str}_{short_uuid}"
+
+        # Determine branch title
+        if branch_name:
+            branch_title = branch_name
+        else:
+            # Auto-generate from the current session title
+            current_title = None
+            if self._session_db:
+                current_title = self._session_db.get_session_title(self.session_id)
+            base = current_title or "branch"
+            branch_title = self._session_db.get_next_title_in_lineage(base)
+
+        # Save the current session's state before branching
+        parent_session_id = self.session_id
+
+        # End the old session
+        try:
+            self._session_db.end_session(self.session_id, "branched")
+        except Exception:
+            pass
+
+        # Create the new session with parent link
+        try:
+            self._session_db.create_session(
+                session_id=new_session_id,
+                source=os.environ.get("HERMES_SESSION_SOURCE", "cli"),
+                model=self.model,
+                model_config={
+                    "max_iterations": self.max_turns,
+                    "reasoning_config": self.reasoning_config,
+                },
+                parent_session_id=parent_session_id,
+            )
+        except Exception as e:
+            _cprint(f"  Failed to create branch session: {e}")
+            return
+
+        # Copy conversation history to the new session
+        for msg in self.conversation_history:
+            try:
+                self._session_db.append_message(
+                    session_id=new_session_id,
+                    role=msg.get("role", "user"),
+                    content=msg.get("content"),
+                    tool_name=msg.get("tool_name") or msg.get("name"),
+                    tool_calls=msg.get("tool_calls"),
+                    tool_call_id=msg.get("tool_call_id"),
+                    reasoning=msg.get("reasoning"),
+                )
+            except Exception:
+                pass  # Best-effort copy
+
+        # Set title on the branch
+        try:
+            self._session_db.set_session_title(new_session_id, branch_title)
+        except Exception:
+            pass
+
+        # Switch to the new session
+        self.session_id = new_session_id
+        self.session_start = now
+        self._pending_title = None
+        self._resumed = True  # Prevents auto-title generation
+
+        # Sync the agent
+        if self.agent:
+            self.agent.session_id = new_session_id
+            self.agent.session_start = now
+            self.agent.reset_session_state()
+            if hasattr(self.agent, "_last_flushed_db_idx"):
+                self.agent._last_flushed_db_idx = len(self.conversation_history)
+            if hasattr(self.agent, "_todo_store"):
+                try:
+                    from tools.todo_tool import TodoStore
+                    self.agent._todo_store = TodoStore()
+                except Exception:
+                    pass
+            if hasattr(self.agent, "_invalidate_system_prompt"):
+                self.agent._invalidate_system_prompt()
+
+        msg_count = len([m for m in self.conversation_history if m.get("role") == "user"])
+        _cprint(
+            f"  ⑂ Branched session \"{branch_title}\""
+            f" ({msg_count} user message{'s' if msg_count != 1 else ''})"
+        )
+        _cprint(f"  Original session: {parent_session_id}")
+        _cprint(f"  Branch session:   {new_session_id}")
+
    def reset_conversation(self):
        """Reset the conversation by starting a new session."""
+        # Shut down memory provider before resetting — actual session boundary
+        if hasattr(self, 'agent') and self.agent:
+            self.agent.shutdown_memory_provider(self.conversation_history)
        self.new_session()
    
    def save_conversation(self):
@@ -3812,28 +4100,6 @@ class HermesCLI:
                            try:
                                if self._session_db.set_session_title(self.session_id, new_title):
                                    _cprint(f"  Session title set: {new_title}")
-                                    # Re-map Honcho session key to new title
-                                    if self.agent and getattr(self.agent, '_honcho', None):
-                                        try:
-                                            hcfg = self.agent._honcho_config
-                                            new_key = (
-                                                hcfg.resolve_session_name(
-                                                    session_title=new_title,
-                                                    session_id=self.agent.session_id,
-                                                )
-                                                if hcfg else new_title
-                                            )
-                                            if new_key and new_key != self.agent._honcho_session_key:
-                                                old_key = self.agent._honcho_session_key
-                                                self.agent._honcho.get_or_create(new_key)
-                                                self.agent._honcho_session_key = new_key
-                                                from tools.honcho_tools import set_session_context
-                                                set_session_context(self.agent._honcho, new_key)
-                                                from agent.display import honcho_session_line, write_tty
-                                                write_tty(honcho_session_line(hcfg.workspace_id, new_key) + "\n")
-                                                _cprint(f"  Honcho session: {old_key} → {new_key}")
-                                        except Exception:
-                                            pass
                                else:
                                    _cprint("  Session not found in database.")
                            except ValueError as e:
@@ -3885,6 +4151,8 @@ class HermesCLI:
                self._pending_input.put(retry_msg)
        elif canonical == "undo":
            self.undo_last()
+        elif canonical == "branch":
+            self._handle_branch_command(cmd_original)
        elif canonical == "save":
            self.save_conversation()
        elif canonical == "cron":
@@ -4298,7 +4566,6 @@ class HermesCLI:
                    user_message=btw_prompt,
                    conversation_history=history_snapshot,
                    task_id=task_id,
-                    sync_honcho=False,
                )

                response = (result.get("final_response") or "") if result else ""
@@ -4728,12 +4995,7 @@ class HermesCLI:
                f"  ✅ Compressed: {original_count} → {new_count} messages "
                f"(~{approx_tokens:,} → ~{new_tokens:,} tokens)"
            )
-            # Flush Honcho async queue so queued messages land before context resets
-            if self.agent and getattr(self.agent, '_honcho', None):
-                try:
-                    self.agent._honcho.flush_all()
-                except Exception:
-                    pass
+
        except Exception as e:
            print(f"  ❌ Compression failed: {e}")

@@ -4892,11 +5154,18 @@ class HermesCLI:
            return  # mcp_servers unchanged (some other section was edited)

        self._config_mcp_servers = new_mcp
-        # Notify user and reload
+        # Notify user and reload.  Run in a separate thread with a hard
+        # timeout so a hung MCP server cannot block the process_loop
+        # indefinitely (which would freeze the entire TUI).
        print()
        print("🔄 MCP server config changed — reloading connections...")
-        with self._busy_command(self._slow_command_status("/reload-mcp")):
-            self._reload_mcp()
+        _reload_thread = threading.Thread(
+            target=self._reload_mcp, daemon=True
+        )
+        _reload_thread.start()
+        _reload_thread.join(timeout=30)
+        if _reload_thread.is_alive():
+            print("  ⚠️  MCP reload timed out (30s). Some servers may not have reconnected.")

    def _reload_mcp(self):
        """Reload MCP servers: disconnect all, re-read config.yaml, reconnect.
@@ -6132,8 +6401,11 @@ class HermesCLI:
                ).start()


-            # Combine all interrupt messages (user may have typed multiple while waiting)
-            # and re-queue as one prompt for process_loop
+            # Re-queue the interrupt message (and any that arrived while we were
+            # processing the first) as the next prompt for process_loop.
+            # Only reached when busy_input_mode == "interrupt" (the default).
+            # In "queue" mode Enter routes directly to _pending_input so this
+            # block is never hit.
            if pending_message and hasattr(self, '_pending_input'):
                all_parts = [pending_message]
                while not self._interrupt_queue.empty():
@@ -6144,7 +6416,12 @@ class HermesCLI:
                    except queue.Empty:
                        break
                combined = "\n".join(all_parts)
-                print(f"\n📨 Queued: '{combined[:50]}{'...' if len(combined) > 50 else ''}'")
+                n = len(all_parts)
+                preview = combined[:50] + ("..." if len(combined) > 50 else "")
+                if n > 1:
+                    print(f"\n⚡ Sending {n} messages after interrupt: '{preview}'")
+                else:
+                    print(f"\n⚡ Sending after interrupt: '{preview}'")
                self._pending_input.put(combined)
            
            return response
@@ -6394,17 +6671,6 @@ class HermesCLI:
        # One-line Honcho session indicator (TTY-only, not captured by agent).
        # Only show when the user explicitly configured Honcho for Hermes
        # (not auto-enabled from a stray HONCHO_API_KEY env var).
-        try:
-            from honcho_integration.client import HonchoClientConfig
-            from agent.display import honcho_session_line, write_tty
-            hcfg = HonchoClientConfig.from_global_config()
-            if hcfg.enabled and (hcfg.api_key or hcfg.base_url) and hcfg.explicitly_configured:
-                sname = hcfg.resolve_session_name(session_id=self.session_id)
-                if sname:
-                    write_tty(honcho_session_line(hcfg.workspace_id, sname) + "\n")
-        except Exception:
-            pass
-
        # If resuming a session, load history and display it immediately
        # so the user has context before typing their first message.
        if self._resumed:
@@ -6581,7 +6847,7 @@ class HermesCLI:
                event.app.invalidate()
                # Bundle text + images as a tuple when images are present
                payload = (text, images) if images else text
-                if self._agent_running and not (text and text.startswith("/")):
+                if self._agent_running and not (text and _looks_like_slash_command(text)):
                    if self.busy_input_mode == "queue":
                        # Queue for the next turn instead of interrupting
                        self._pending_input.put(payload)
@@ -6890,6 +7156,9 @@ class HermesCLI:
            buffer.
            """
            pasted_text = event.data or ""
+            # Normalise line endings — Windows \r\n and old Mac \r both become \n
+            # so the 5-line collapse threshold and display are consistent.
+            pasted_text = pasted_text.replace('\r\n', '\n').replace('\r', '\n')
            if self._try_attach_clipboard_image():
                event.app.invalidate()
            if pasted_text:
@@ -7503,6 +7772,49 @@ class HermesCLI:
        )
        self._app = app  # Store reference for clarify_callback

+        # ── Fix ghost status-bar lines on terminal resize ──────────────
+        # When the terminal shrinks (e.g. un-maximize), the emulator reflows
+        # the previously-rendered full-width rows (status bar, input rules)
+        # into multiple narrower rows.  prompt_toolkit's _on_resize handler
+        # only cursor_up()s by the stored layout height, missing the extra
+        # rows created by reflow — leaving ghost duplicates visible.
+        #
+        # Fix: before the standard erase, inflate _cursor_pos.y so the
+        # cursor moves up far enough to cover the reflowed ghost content.
+        _original_on_resize = app._on_resize
+
+        def _resize_clear_ghosts():
+            from prompt_toolkit.data_structures import Point as _Pt
+            renderer = app.renderer
+            try:
+                old_size = renderer._last_size
+                new_size = renderer.output.get_size()
+                if (
+                    old_size
+                    and new_size.columns < old_size.columns
+                    and new_size.columns > 0
+                ):
+                    reflow_factor = (
+                        (old_size.columns + new_size.columns - 1)
+                        // new_size.columns
+                    )
+                    last_h = (
+                        renderer._last_screen.height
+                        if renderer._last_screen
+                        else 0
+                    )
+                    extra = last_h * (reflow_factor - 1)
+                    if extra > 0:
+                        renderer._cursor_pos = _Pt(
+                            x=renderer._cursor_pos.x,
+                            y=renderer._cursor_pos.y + extra,
+                        )
+            except Exception:
+                pass  # never break resize handling
+            _original_on_resize()
+
+        app._on_resize = _resize_clear_ghosts
+
        def spinner_loop():
            import time as _time

@@ -7545,8 +7857,24 @@ class HermesCLI:
                    if isinstance(user_input, tuple):
                        user_input, submit_images = user_input
                    
-                    # Check for commands
-                    if isinstance(user_input, str) and user_input.startswith("/"):
+                    # Check for commands — but detect dragged/pasted file paths first.
+                    # See _detect_file_drop() for details.
+                    _file_drop = _detect_file_drop(user_input) if isinstance(user_input, str) else None
+                    if _file_drop:
+                        _drop_path = _file_drop["path"]
+                        _remainder = _file_drop["remainder"]
+                        if _file_drop["is_image"]:
+                            submit_images.append(_drop_path)
+                            user_input = _remainder or f"[User attached image: {_drop_path.name}]"
+                            _cprint(f"  📎 Auto-attached image: {_drop_path.name}")
+                        else:
+                            _cprint(f"  📄 Detected file: {_drop_path.name}")
+                            user_input = (
+                                f"[User attached file: {_drop_path}]"
+                                + (f"\n{_remainder}" if _remainder else "")
+                            )
+
+                    if not _file_drop and isinstance(user_input, str) and _looks_like_slash_command(user_input):
                        _cprint(f"\n⚙️  {user_input}")
                        if not self.process_command(user_input):
                            self._should_exit = True
@@ -7707,12 +8035,6 @@ class HermesCLI:
            set_sudo_password_callback(None)
            set_approval_callback(None)
            set_secret_capture_callback(None)
-            # Flush + shut down Honcho async writer (drains queue before exit)
-            if self.agent and getattr(self.agent, '_honcho', None):
-                try:
-                    self.agent._honcho.shutdown()
-                except (Exception, KeyboardInterrupt):
-                    pass
            # Close session in SQLite
            if hasattr(self, '_session_db') and self._session_db and self.agent:
                try:
@@ -7937,6 +8259,12 @@ def main(
                    if response:
                        print(response)
                    print(f"\nsession_id: {cli.session_id}")
+                    
+                    # Ensure proper exit code for automation wrappers
+                    sys.exit(1 if isinstance(result, dict) and result.get("failed") else 0)
+            
+            # Exit with error code if credentials or agent init fails
+            sys.exit(1)
        else:
            cli.show_banner()
            cli.console.print(f"[bold blue]Query:[/] {query}")
@@ -375,6 +375,7 @@ def create_job(
    model: Optional[str] = None,
    provider: Optional[str] = None,
    base_url: Optional[str] = None,
+    script: Optional[str] = None,
 ) -> Dict[str, Any]:
    """
    Create a new cron job.
@@ -391,6 +392,9 @@ def create_job(
        model: Optional per-job model override
        provider: Optional per-job provider override
        base_url: Optional per-job base URL override
+        script: Optional path to a Python script whose stdout is injected into the
+                prompt each run.  The script runs before the agent turn, and its output
+                is prepended as context.  Useful for data collection / change detection.

    Returns:
        The created job dict
@@ -419,6 +423,8 @@ def create_job(
    normalized_model = normalized_model or None
    normalized_provider = normalized_provider or None
    normalized_base_url = normalized_base_url or None
+    normalized_script = str(script).strip() if isinstance(script, str) else None
+    normalized_script = normalized_script or None

    label_source = (prompt or (normalized_skills[0] if normalized_skills else None)) or "cron job"
    job = {
@@ -430,6 +436,7 @@ def create_job(
        "model": normalized_model,
        "provider": normalized_provider,
        "base_url": normalized_base_url,
+        "script": normalized_script,
        "schedule": parsed_schedule,
        "schedule_display": parsed_schedule.get("display", schedule),
        "repeat": {
@@ -9,9 +9,11 @@ runs at a time if multiple processes overlap.
 """

 import asyncio
+import concurrent.futures
 import json
 import logging
 import os
+import subprocess
 import sys
 import traceback

@@ -228,11 +230,89 @@ def _deliver_result(job: dict, content: str) -> None:
        logger.info("Job '%s': delivered to %s:%s", job["id"], platform_name, chat_id)


+_SCRIPT_TIMEOUT = 120  # seconds
+
+
+def _run_job_script(script_path: str) -> tuple[bool, str]:
+    """Execute a cron job's data-collection script and capture its output.
+
+    Args:
+        script_path: Path to a Python script (resolved via HERMES_HOME/scripts/ or absolute).
+
+    Returns:
+        (success, output) — on failure *output* contains the error message so the
+        LLM can report the problem to the user.
+    """
+    from hermes_constants import get_hermes_home
+
+    path = Path(script_path).expanduser()
+    if not path.is_absolute():
+        # Resolve relative paths against HERMES_HOME/scripts/
+        path = get_hermes_home() / "scripts" / path
+
+    if not path.exists():
+        return False, f"Script not found: {path}"
+    if not path.is_file():
+        return False, f"Script path is not a file: {path}"
+
+    try:
+        result = subprocess.run(
+            [sys.executable, str(path)],
+            capture_output=True,
+            text=True,
+            timeout=_SCRIPT_TIMEOUT,
+            cwd=str(path.parent),
+        )
+        stdout = (result.stdout or "").strip()
+        stderr = (result.stderr or "").strip()
+
+        if result.returncode != 0:
+            parts = [f"Script exited with code {result.returncode}"]
+            if stderr:
+                parts.append(f"stderr:\n{stderr}")
+            if stdout:
+                parts.append(f"stdout:\n{stdout}")
+            return False, "\n".join(parts)
+
+        return True, stdout
+
+    except subprocess.TimeoutExpired:
+        return False, f"Script timed out after {_SCRIPT_TIMEOUT}s: {path}"
+    except Exception as exc:
+        return False, f"Script execution failed: {exc}"
+
+
 def _build_job_prompt(job: dict) -> str:
    """Build the effective prompt for a cron job, optionally loading one or more skills first."""
    prompt = job.get("prompt", "")
    skills = job.get("skills")

+    # Run data-collection script if configured, inject output as context.
+    script_path = job.get("script")
+    if script_path:
+        success, script_output = _run_job_script(script_path)
+        if success:
+            if script_output:
+                prompt = (
+                    "## Script Output\n"
+                    "The following data was collected by a pre-run script. "
+                    "Use it as context for your analysis.\n\n"
+                    f"```\n{script_output}\n```\n\n"
+                    f"{prompt}"
+                )
+            else:
+                prompt = (
+                    "[Script ran successfully but produced no output.]\n\n"
+                    f"{prompt}"
+                )
+        else:
+            prompt = (
+                "## Script Error\n"
+                "The data-collection script failed. Report this to the user.\n\n"
+                f"```\n{script_output}\n```\n\n"
+                f"{prompt}"
+            )
+
    # Always prepend [SILENT] guidance so the cron agent can suppress
    # delivery when it has nothing new or noteworthy to report.
    silent_hint = (
@@ -437,13 +517,36 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
            provider_sort=pr.get("sort"),
            disabled_toolsets=["cronjob", "messaging", "clarify"],
            quiet_mode=True,
+            skip_memory=True,  # Cron system prompts would corrupt user representations
            platform="cron",
            session_id=_cron_session_id,
            session_db=_session_db,
        )
        
-        result = agent.run_conversation(prompt)
-        
+        # Run the agent with a timeout so a hung API call or tool doesn't
+        # block the cron ticker thread indefinitely.  Default 10 minutes;
+        # override via env var.  Uses a separate thread because
+        # run_conversation is synchronous.
+        _cron_timeout = float(os.getenv("HERMES_CRON_TIMEOUT", 600))
+        _cron_pool = concurrent.futures.ThreadPoolExecutor(max_workers=1)
+        _cron_future = _cron_pool.submit(agent.run_conversation, prompt)
+        try:
+            result = _cron_future.result(timeout=_cron_timeout)
+        except concurrent.futures.TimeoutError:
+            logger.error(
+                "Job '%s' timed out after %.0fs — interrupting agent",
+                job_name, _cron_timeout,
+            )
+            if hasattr(agent, "interrupt"):
+                agent.interrupt("Cron job timed out")
+            _cron_pool.shutdown(wait=False, cancel_futures=True)
+            raise TimeoutError(
+                f"Cron job '{job_name}' timed out after "
+                f"{int(_cron_timeout // 60)} minutes"
+            )
+        finally:
+            _cron_pool.shutdown(wait=False)
+
        final_response = result.get("final_response", "") or ""
        # Use a separate variable for log display; keep final_response clean
        # for delivery logic (empty response = no delivery).
@@ -76,14 +76,13 @@ Open Zed settings (`Cmd+,` on macOS or `Ctrl+,` on Linux) and add to your

 ```json
 {
-  "acp": {
-    "agents": [
-      {
-        "name": "hermes-agent",
-        "registry_dir": "/path/to/hermes-agent/acp_registry"
-      }
-    ]
-  }
+  "agent_servers": {
+    "hermes-agent": {
+      "type": "custom",
+      "command": "hermes",
+      "args": ["acp"],
+    },
+  },
 }
 ```

@@ -11,11 +11,11 @@ Solution:
    _AsyncWorker thread internally, making it safe for both CLI and Atropos use.
    No monkey-patching is required.

-    This module is kept for backward compatibility — apply_patches() is now a no-op.
+    This module is kept for backward compatibility. apply_patches() is a no-op.

 Usage:
    Call apply_patches() once at import time (done automatically by hermes_base_env.py).
-    This is idempotent — calling it multiple times is safe.
+    This is idempotent and safe to call multiple times.
 """

 import logging
@@ -26,17 +26,10 @@ _patches_applied = False


 def apply_patches():
-    """Apply all monkey patches needed for Atropos compatibility.
-
-    Now a no-op — Modal async safety is built directly into ModalEnvironment.
-    Safe to call multiple times.
-    """
+    """Apply all monkey patches needed for Atropos compatibility."""
    global _patches_applied
    if _patches_applied:
        return

-    # Modal async-safety is now built into tools/environments/modal.py
-    # via the _AsyncWorker class. No monkey-patching needed.
-    logger.debug("apply_patches() called — no patches needed (async safety is built-in)")
-
+    logger.debug("apply_patches() called; no patches needed (async safety is built-in)")
    _patches_applied = True
@@ -17,6 +17,7 @@ from typing import Dict, List, Optional, Any
 from enum import Enum

 from hermes_cli.config import get_hermes_home
+from utils import is_truthy_value

 logger = logging.getLogger(__name__)

@@ -25,10 +26,6 @@ def _coerce_bool(value: Any, default: bool = True) -> bool:
    """Coerce bool-ish config values, preserving a caller-provided default."""
    if value is None:
        return default
-    if isinstance(value, bool):
-        return value
-    if isinstance(value, int):
-        return value != 0
    if isinstance(value, str):
        lowered = value.strip().lower()
        if lowered in ("true", "1", "yes", "on"):
@@ -36,7 +33,7 @@ def _coerce_bool(value: Any, default: bool = True) -> bool:
        if lowered in ("false", "0", "no", "off"):
            return False
        return default
-    return default
+    return is_truthy_value(value, default=default)


 def _normalize_unauthorized_dm_behavior(value: Any, default: str = "pair") -> str:
@@ -566,6 +563,32 @@ def load_gateway_config() -> GatewayConfig:
                    if isinstance(frc, list):
                        frc = ",".join(str(v) for v in frc)
                    os.environ["TELEGRAM_FREE_RESPONSE_CHATS"] = str(frc)
+
+            whatsapp_cfg = yaml_cfg.get("whatsapp", {})
+            if isinstance(whatsapp_cfg, dict):
+                if "require_mention" in whatsapp_cfg and not os.getenv("WHATSAPP_REQUIRE_MENTION"):
+                    os.environ["WHATSAPP_REQUIRE_MENTION"] = str(whatsapp_cfg["require_mention"]).lower()
+                if "mention_patterns" in whatsapp_cfg and not os.getenv("WHATSAPP_MENTION_PATTERNS"):
+                    os.environ["WHATSAPP_MENTION_PATTERNS"] = json.dumps(whatsapp_cfg["mention_patterns"])
+                frc = whatsapp_cfg.get("free_response_chats")
+                if frc is not None and not os.getenv("WHATSAPP_FREE_RESPONSE_CHATS"):
+                    if isinstance(frc, list):
+                        frc = ",".join(str(v) for v in frc)
+                    os.environ["WHATSAPP_FREE_RESPONSE_CHATS"] = str(frc)
+
+            # Matrix settings → env vars (env vars take precedence)
+            matrix_cfg = yaml_cfg.get("matrix", {})
+            if isinstance(matrix_cfg, dict):
+                if "require_mention" in matrix_cfg and not os.getenv("MATRIX_REQUIRE_MENTION"):
+                    os.environ["MATRIX_REQUIRE_MENTION"] = str(matrix_cfg["require_mention"]).lower()
+                frc = matrix_cfg.get("free_response_rooms")
+                if frc is not None and not os.getenv("MATRIX_FREE_RESPONSE_ROOMS"):
+                    if isinstance(frc, list):
+                        frc = ",".join(str(v) for v in frc)
+                    os.environ["MATRIX_FREE_RESPONSE_ROOMS"] = str(frc)
+                if "auto_thread" in matrix_cfg and not os.getenv("MATRIX_AUTO_THREAD"):
+                    os.environ["MATRIX_AUTO_THREAD"] = str(matrix_cfg["auto_thread"]).lower()
+
    except Exception as e:
        logger.warning(
            "Failed to process config.yaml — falling back to .env / gateway.json values. "
@@ -908,5 +931,3 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
            config.default_reset_policy.at_hour = int(reset_hour)
        except ValueError:
            pass
-
-
@@ -2,7 +2,7 @@
 OpenAI-compatible API server platform adapter.

 Exposes an HTTP server with endpoints:
- POST /v1/chat/completions        — OpenAI Chat Completions format (stateless)
+- POST /v1/chat/completions        — OpenAI Chat Completions format (stateless; opt-in session continuity via X-Hermes-Session-Id header)
 - POST /v1/responses               — OpenAI Responses API format (stateful via previous_response_id)
 - GET  /v1/responses/{response_id} — Retrieve a stored response
 - DELETE /v1/responses/{response_id} — Delete a stored response
@@ -300,6 +300,7 @@ class APIServerAdapter(BasePlatformAdapter):
        self._runner: Optional["web.AppRunner"] = None
        self._site: Optional["web.TCPSite"] = None
        self._response_store = ResponseStore()
+        self._session_db: Optional[Any] = None  # Lazy-init SessionDB for session continuity

    @staticmethod
    def _parse_cors_origins(value: Any) -> tuple[str, ...]:
@@ -371,6 +372,24 @@ class APIServerAdapter(BasePlatformAdapter):
            status=401,
        )

+    # ------------------------------------------------------------------
+    # Session DB helper
+    # ------------------------------------------------------------------
+
+    def _ensure_session_db(self):
+        """Lazily initialise and return the shared SessionDB instance.
+
+        Sessions are persisted to ``state.db`` so that ``hermes sessions list``
+        shows API-server conversations alongside CLI and gateway ones.
+        """
+        if self._session_db is None:
+            try:
+                from hermes_state import SessionDB
+                self._session_db = SessionDB()
+            except Exception as e:
+                logger.debug("SessionDB unavailable for API server: %s", e)
+        return self._session_db
+
    # ------------------------------------------------------------------
    # Agent creation helper
    # ------------------------------------------------------------------
@@ -414,6 +433,7 @@ class APIServerAdapter(BasePlatformAdapter):
            platform="api_server",
            stream_delta_callback=stream_delta_callback,
            tool_progress_callback=tool_progress_callback,
+            session_db=self._ensure_session_db(),
        )
        return agent

@@ -496,7 +516,22 @@ class APIServerAdapter(BasePlatformAdapter):
                status=400,
            )

-        session_id = str(uuid.uuid4())
+        # Allow caller to continue an existing session by passing X-Hermes-Session-Id.
+        # When provided, history is loaded from state.db instead of from the request body.
+        provided_session_id = request.headers.get("X-Hermes-Session-Id", "").strip()
+        if provided_session_id:
+            session_id = provided_session_id
+            try:
+                db = self._ensure_session_db()
+                if db is not None:
+                    history = db.get_messages_as_conversation(session_id)
+            except Exception as e:
+                logger.warning("Failed to load session history for %s: %s", session_id, e)
+                history = []
+        else:
+            session_id = str(uuid.uuid4())
+            # history already set from request body above
+
        completion_id = f"chatcmpl-{uuid.uuid4().hex[:29]}"
        model_name = body.get("model", "hermes-agent")
        created = int(time.time())
@@ -540,7 +575,7 @@ class APIServerAdapter(BasePlatformAdapter):

            return await self._write_sse_chat_completion(
                request, completion_id, model_name, created, _stream_q,
-                agent_task, agent_ref,
+                agent_task, agent_ref, session_id=session_id,
            )

        # Non-streaming: run the agent (with optional Idempotency-Key)
@@ -599,11 +634,11 @@ class APIServerAdapter(BasePlatformAdapter):
            },
        }

-        return web.json_response(response_data)
+        return web.json_response(response_data, headers={"X-Hermes-Session-Id": session_id})

    async def _write_sse_chat_completion(
        self, request: "web.Request", completion_id: str, model: str,
-        created: int, stream_q, agent_task, agent_ref=None,
+        created: int, stream_q, agent_task, agent_ref=None, session_id: str = None,
    ) -> "web.StreamResponse":
        """Write real streaming SSE from agent's stream_delta_callback queue.

@@ -620,6 +655,8 @@ class APIServerAdapter(BasePlatformAdapter):
        cors = self._cors_headers_for_origin(origin) if origin else None
        if cors:
            sse_headers.update(cors)
+        if session_id:
+            sse_headers["X-Hermes-Session-Id"] = session_id
        response = web.StreamResponse(status=200, headers=sse_headers)
        await response.prepare(request)

@@ -235,6 +235,7 @@ SUPPORTED_DOCUMENT_TYPES = {
    ".pdf": "application/pdf",
    ".md": "text/markdown",
    ".txt": "text/plain",
+    ".zip": "application/zip",
    ".docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
    ".xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
    ".pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
@@ -1021,6 +1022,32 @@ class BasePlatformAdapter(ABC):
        
        # Check if there's already an active handler for this session
        if session_key in self._active_sessions:
+            # /approve and /deny must bypass the active-session guard.
+            # The agent thread is blocked on threading.Event.wait() inside
+            # tools/approval.py — queuing these commands creates a deadlock:
+            # the agent waits for approval, approval waits for agent to finish.
+            # Dispatch directly to the message handler without touching session
+            # lifecycle (no competing background task, no session guard removal).
+            cmd = event.get_command()
+            if cmd in ("approve", "deny"):
+                logger.debug(
+                    "[%s] Approval command '/%s' bypassing active-session guard for %s",
+                    self.name, cmd, session_key,
+                )
+                try:
+                    _thread_meta = {"thread_id": event.source.thread_id} if event.source.thread_id else None
+                    response = await self._message_handler(event)
+                    if response:
+                        await self._send_with_retry(
+                            chat_id=event.source.chat_id,
+                            content=response,
+                            reply_to=event.message_id,
+                            metadata=_thread_meta,
+                        )
+                except Exception as e:
+                    logger.error("[%s] Approval dispatch failed: %s", self.name, e, exc_info=True)
+                return
+
            # Special case: photo bursts/albums frequently arrive as multiple near-
            # simultaneous messages. Queue them without interrupting the active run,
            # then process them immediately after the current task finishes.
@@ -1046,6 +1073,13 @@ class BasePlatformAdapter(ABC):
            self._active_sessions[session_key].set()
            return  # Don't process now - will be handled after current task finishes
        
+        # Mark session as active BEFORE spawning background task to close
+        # the race window where a second message arriving before the task
+        # starts would also pass the _active_sessions check and spawn a
+        # duplicate task.  (grammY sequentialize / aiogram EventIsolation
+        # pattern — set the guard synchronously, not inside the task.)
+        self._active_sessions[session_key] = asyncio.Event()
+
        # Spawn background task to process this message
        task = asyncio.create_task(self._process_message_background(event, session_key))
        try:
@@ -1092,8 +1126,10 @@ class BasePlatformAdapter(ABC):
            if getattr(result, "success", False):
                delivery_succeeded = True

-        # Create interrupt event for this session
-        interrupt_event = asyncio.Event()
+        # Reuse the interrupt event set by handle_message() (which marks
+        # the session active before spawning this task to prevent races).
+        # Fall back to a new Event only if the entry was removed externally.
+        interrupt_event = self._active_sessions.get(session_key) or asyncio.Event()
        self._active_sessions[session_key] = interrupt_event
        
        # Start continuous typing indicator (refreshes every 2 seconds)
@@ -1106,9 +1142,12 @@ class BasePlatformAdapter(ABC):
            # Call the handler (this can take a while with tool calls)
            response = await self._message_handler(event)
            
-            # Send response if any
+            # Send response if any.  A None/empty response is normal when
+            # streaming already delivered the text (already_sent=True) or
+            # when the message was queued behind an active agent.  Log at
+            # DEBUG to avoid noisy warnings for expected behavior.
            if not response:
-                logger.warning("[%s] Handler returned empty/None response for %s", self.name, event.source.chat_id)
+                logger.debug("[%s] Handler returned empty/None response for %s", self.name, event.source.chat_id)
            if response:
                # Extract MEDIA:<path> tags (from TTS tool) before other processing
                media_files, response = self.extract_media(response)
@@ -449,6 +449,11 @@ class DiscordAdapter(BasePlatformAdapter):
        self._bot_task: Optional[asyncio.Task] = None
        # Cap to prevent unbounded growth (Discord threads get archived).
        self._MAX_TRACKED_THREADS = 500
+        # Dedup cache: message_id → timestamp.  Prevents duplicate bot
+        # responses when Discord RESUME replays events after reconnects.
+        self._seen_messages: Dict[str, float] = {}
+        self._SEEN_TTL = 300   # 5 minutes
+        self._SEEN_MAX = 2000  # prune threshold

    async def connect(self) -> bool:
        """Connect to Discord and start receiving events."""
@@ -539,6 +544,19 @@ class DiscordAdapter(BasePlatformAdapter):

            @self._client.event
            async def on_message(message: DiscordMessage):
+                # Dedup: Discord RESUME replays events after reconnects (#4777)
+                msg_id = str(message.id)
+                now = time.time()
+                if msg_id in adapter_self._seen_messages:
+                    return
+                adapter_self._seen_messages[msg_id] = now
+                if len(adapter_self._seen_messages) > adapter_self._SEEN_MAX:
+                    cutoff = now - adapter_self._SEEN_TTL
+                    adapter_self._seen_messages = {
+                        k: v for k, v in adapter_self._seen_messages.items()
+                        if v > cutoff
+                    }
+
                # Always ignore our own messages
                if message.author == self._client.user:
                    return
@@ -1617,6 +1635,16 @@ class DiscordAdapter(BasePlatformAdapter):
        async def slash_update(interaction: discord.Interaction):
            await self._run_simple_slash(interaction, "/update", "Update initiated~")

+        @tree.command(name="approve", description="Approve a pending dangerous command")
+        @discord.app_commands.describe(scope="Optional: 'all', 'session', 'always', 'all session', 'all always'")
+        async def slash_approve(interaction: discord.Interaction, scope: str = ""):
+            await self._run_simple_slash(interaction, f"/approve {scope}".strip())
+
+        @tree.command(name="deny", description="Deny a pending dangerous command")
+        @discord.app_commands.describe(scope="Optional: 'all' to deny all pending commands")
+        async def slash_deny(interaction: discord.Interaction, scope: str = ""):
+            await self._run_simple_slash(interaction, f"/deny {scope}".strip())
+
        @tree.command(name="thread", description="Create a new thread and start a Hermes session in it")
        @discord.app_commands.describe(
            name="Thread name",
@@ -1860,33 +1888,41 @@ class DiscordAdapter(BasePlatformAdapter):
            return None

    async def send_exec_approval(
-        self, chat_id: str, command: str, approval_id: str
+        self, chat_id: str, command: str, session_key: str,
+        description: str = "dangerous command",
+        metadata: Optional[dict] = None,
    ) -> SendResult:
        """
        Send a button-based exec approval prompt for a dangerous command.

-        Returns SendResult. The approval is resolved when a user clicks a button.
+        The buttons call ``resolve_gateway_approval()`` to unblock the waiting
+        agent thread — this replaces the text-based ``/approve`` flow on Discord.
        """
        if not self._client or not DISCORD_AVAILABLE:
            return SendResult(success=False, error="Not connected")

        try:
-            channel = self._client.get_channel(int(chat_id))
+            # Resolve channel — use thread_id from metadata if present
+            target_id = chat_id
+            if metadata and metadata.get("thread_id"):
+                target_id = metadata["thread_id"]
+
+            channel = self._client.get_channel(int(target_id))
            if not channel:
-                channel = await self._client.fetch_channel(int(chat_id))
+                channel = await self._client.fetch_channel(int(target_id))

            # Discord embed description limit is 4096; show full command up to that
            max_desc = 4088
            cmd_display = command if len(command) <= max_desc else command[: max_desc - 3] + "..."
            embed = discord.Embed(
-                title="Command Approval Required",
+                title="⚠️ Command Approval Required",
                description=f"```\n{cmd_display}\n```",
                color=discord.Color.orange(),
            )
-            embed.set_footer(text=f"Approval ID: {approval_id}")
+            embed.add_field(name="Reason", value=description, inline=False)

            view = ExecApprovalView(
-                approval_id=approval_id,
+                session_key=session_key,
                allowed_user_ids=self._allowed_user_ids,
            )

@@ -2219,13 +2255,15 @@ if DISCORD_AVAILABLE:
        """
        Interactive button view for exec approval of dangerous commands.

-        Shows three buttons: Allow Once (green), Always Allow (blue), Deny (red).
-        Only users in the allowed list can click. The view times out after 5 minutes.
+        Shows four buttons: Allow Once, Allow Session, Always Allow, Deny.
+        Clicking a button calls ``resolve_gateway_approval()`` to unblock the
+        waiting agent thread — the same mechanism as the text ``/approve`` flow.
+        Only users in the allowed list can click.  Times out after 5 minutes.
        """

-        def __init__(self, approval_id: str, allowed_user_ids: set):
+        def __init__(self, session_key: str, allowed_user_ids: set):
            super().__init__(timeout=300)  # 5-minute timeout
-            self.approval_id = approval_id
+            self.session_key = session_key
            self.allowed_user_ids = allowed_user_ids
            self.resolved = False

@@ -2236,9 +2274,10 @@ if DISCORD_AVAILABLE:
            return str(interaction.user.id) in self.allowed_user_ids

        async def _resolve(
-            self, interaction: discord.Interaction, action: str, color: discord.Color
+            self, interaction: discord.Interaction, choice: str,
+            color: discord.Color, label: str,
        ):
-            """Resolve the approval and update the message."""
+            """Resolve the approval via the gateway approval queue and update the embed."""
            if self.resolved:
                await interaction.response.send_message(
                    "This approval has already been resolved~", ephemeral=True
@@ -2257,7 +2296,7 @@ if DISCORD_AVAILABLE:
            embed = interaction.message.embeds[0] if interaction.message.embeds else None
            if embed:
                embed.color = color
-                embed.set_footer(text=f"{action} by {interaction.user.display_name}")
+                embed.set_footer(text=f"{label} by {interaction.user.display_name}")

            # Disable all buttons
            for child in self.children:
@@ -2265,33 +2304,40 @@ if DISCORD_AVAILABLE:

            await interaction.response.edit_message(embed=embed, view=self)

-            # Store the approval decision
+            # Unblock the waiting agent thread via the gateway approval queue
            try:
-                from tools.approval import approve_permanent
-                if action == "allow_once":
-                    pass  # One-time approval handled by gateway
-                elif action == "allow_always":
-                    approve_permanent(self.approval_id)
-            except ImportError:
-                pass
+                from tools.approval import resolve_gateway_approval
+                count = resolve_gateway_approval(self.session_key, choice)
+                logger.info(
+                    "Discord button resolved %d approval(s) for session %s (choice=%s, user=%s)",
+                    count, self.session_key, choice, interaction.user.display_name,
+                )
+            except Exception as exc:
+                logger.error("Failed to resolve gateway approval from button: %s", exc)

        @discord.ui.button(label="Allow Once", style=discord.ButtonStyle.green)
        async def allow_once(
            self, interaction: discord.Interaction, button: discord.ui.Button
        ):
-            await self._resolve(interaction, "allow_once", discord.Color.green())
+            await self._resolve(interaction, "once", discord.Color.green(), "Approved once")
+
+        @discord.ui.button(label="Allow Session", style=discord.ButtonStyle.grey)
+        async def allow_session(
+            self, interaction: discord.Interaction, button: discord.ui.Button
+        ):
+            await self._resolve(interaction, "session", discord.Color.blue(), "Approved for session")

        @discord.ui.button(label="Always Allow", style=discord.ButtonStyle.blurple)
        async def allow_always(
            self, interaction: discord.Interaction, button: discord.ui.Button
        ):
-            await self._resolve(interaction, "allow_always", discord.Color.blue())
+            await self._resolve(interaction, "always", discord.Color.purple(), "Approved permanently")

        @discord.ui.button(label="Deny", style=discord.ButtonStyle.red)
        async def deny(
            self, interaction: discord.Interaction, button: discord.ui.Button
        ):
-            await self._resolve(interaction, "deny", discord.Color.red())
+            await self._resolve(interaction, "deny", discord.Color.red(), "Denied")

        async def on_timeout(self):
            """Handle view timeout -- disable buttons and mark as expired."""
@@ -5,13 +5,16 @@ matrix-nio Python SDK.  Supports optional end-to-end encryption (E2EE)
 when installed with ``pip install "matrix-nio[e2e]"``.

 Environment variables:
-    MATRIX_HOMESERVER       Homeserver URL (e.g. https://matrix.example.org)
-    MATRIX_ACCESS_TOKEN     Access token (preferred auth method)
-    MATRIX_USER_ID          Full user ID (@bot:server) — required for password login
-    MATRIX_PASSWORD         Password (alternative to access token)
-    MATRIX_ENCRYPTION       Set "true" to enable E2EE
-    MATRIX_ALLOWED_USERS    Comma-separated Matrix user IDs (@user:server)
-    MATRIX_HOME_ROOM        Room ID for cron/notification delivery
+    MATRIX_HOMESERVER           Homeserver URL (e.g. https://matrix.example.org)
+    MATRIX_ACCESS_TOKEN         Access token (preferred auth method)
+    MATRIX_USER_ID              Full user ID (@bot:server) — required for password login
+    MATRIX_PASSWORD             Password (alternative to access token)
+    MATRIX_ENCRYPTION           Set "true" to enable E2EE
+    MATRIX_ALLOWED_USERS        Comma-separated Matrix user IDs (@user:server)
+    MATRIX_HOME_ROOM            Room ID for cron/notification delivery
+    MATRIX_REQUIRE_MENTION      Require @mention in rooms (default: true)
+    MATRIX_FREE_RESPONSE_ROOMS  Comma-separated room IDs exempt from mention requirement
+    MATRIX_AUTO_THREAD          Auto-create threads for room messages (default: true)
 """

 from __future__ import annotations
@@ -123,6 +126,10 @@ class MatrixAdapter(BasePlatformAdapter):
        # Each entry: (room, event, timestamp)
        self._pending_megolm: list = []

+        # Thread participation tracking (for require_mention bypass)
+        self._bot_participated_threads: set = self._load_participated_threads()
+        self._MAX_TRACKED_THREADS = 500
+
    def _is_duplicate_event(self, event_id) -> bool:
        """Return True if this event was already processed. Tracks the ID otherwise."""
        if not event_id:
@@ -902,6 +909,30 @@ class MatrixAdapter(BasePlatformAdapter):
        if relates_to.get("rel_type") == "m.thread":
            thread_id = relates_to.get("event_id")

+        # Require-mention gating.
+        if not is_dm:
+            free_rooms_raw = os.getenv("MATRIX_FREE_RESPONSE_ROOMS", "")
+            free_rooms = {r.strip() for r in free_rooms_raw.split(",") if r.strip()}
+            require_mention = os.getenv("MATRIX_REQUIRE_MENTION", "true").lower() not in ("false", "0", "no")
+            is_free_room = room.room_id in free_rooms
+            in_bot_thread = bool(thread_id and thread_id in self._bot_participated_threads)
+
+            formatted_body = source_content.get("formatted_body")
+            if require_mention and not is_free_room and not in_bot_thread:
+                if not self._is_bot_mentioned(body, formatted_body):
+                    return
+
+        # Strip mention from body when present (including in DMs).
+        if self._is_bot_mentioned(body, source_content.get("formatted_body")):
+            body = self._strip_mention(body)
+
+        # Auto-thread: create a thread for non-DM, non-threaded messages.
+        if not is_dm and not thread_id:
+            auto_thread = os.getenv("MATRIX_AUTO_THREAD", "true").lower() in ("true", "1", "yes")
+            if auto_thread:
+                thread_id = event.event_id
+                self._track_thread(thread_id)
+
        # Reply-to detection.
        reply_to = None
        in_reply_to = relates_to.get("m.in_reply_to", {})
@@ -946,6 +977,9 @@ class MatrixAdapter(BasePlatformAdapter):
            reply_to_message_id=reply_to,
        )

+        if thread_id:
+            self._track_thread(thread_id)
+
        await self.handle_message(msg_event)

    async def _on_room_message_media(self, room: Any, event: Any) -> None:
@@ -1031,6 +1065,30 @@ class MatrixAdapter(BasePlatformAdapter):
        if relates_to.get("rel_type") == "m.thread":
            thread_id = relates_to.get("event_id")

+        # Require-mention gating (media messages).
+        if not is_dm:
+            free_rooms_raw = os.getenv("MATRIX_FREE_RESPONSE_ROOMS", "")
+            free_rooms = {r.strip() for r in free_rooms_raw.split(",") if r.strip()}
+            require_mention = os.getenv("MATRIX_REQUIRE_MENTION", "true").lower() not in ("false", "0", "no")
+            is_free_room = room.room_id in free_rooms
+            in_bot_thread = bool(thread_id and thread_id in self._bot_participated_threads)
+
+            if require_mention and not is_free_room and not in_bot_thread:
+                formatted_body = source_content.get("formatted_body")
+                if not self._is_bot_mentioned(body, formatted_body):
+                    return
+
+        # Strip mention from body when present (including in DMs).
+        if self._is_bot_mentioned(body, source_content.get("formatted_body")):
+            body = self._strip_mention(body)
+
+        # Auto-thread: create a thread for non-DM, non-threaded messages.
+        if not is_dm and not thread_id:
+            auto_thread = os.getenv("MATRIX_AUTO_THREAD", "true").lower() in ("true", "1", "yes")
+            if auto_thread:
+                thread_id = event.event_id
+                self._track_thread(thread_id)
+
        # For voice messages, cache audio locally for transcription tools.
        # Use the authenticated nio client to download (Matrix requires auth for media).
        media_urls = [http_url] if http_url else None
@@ -1079,6 +1137,9 @@ class MatrixAdapter(BasePlatformAdapter):
            media_types=media_types,
        )

+        if thread_id:
+            self._track_thread(thread_id)
+
        await self.handle_message(msg_event)

    async def _on_invite(self, room: Any, event: Any) -> None:
@@ -1166,6 +1227,82 @@ class MatrixAdapter(BasePlatformAdapter):
            for rid in self._joined_rooms
        }

+    # ------------------------------------------------------------------
+    # Thread participation tracking
+    # ------------------------------------------------------------------
+
+    @staticmethod
+    def _thread_state_path() -> Path:
+        """Path to the persisted thread participation set."""
+        from hermes_cli.config import get_hermes_home
+        return get_hermes_home() / "matrix_threads.json"
+
+    @classmethod
+    def _load_participated_threads(cls) -> set:
+        """Load persisted thread IDs from disk."""
+        path = cls._thread_state_path()
+        try:
+            if path.exists():
+                data = json.loads(path.read_text(encoding="utf-8"))
+                if isinstance(data, list):
+                    return set(data)
+        except Exception as e:
+            logger.debug("Could not load matrix thread state: %s", e)
+        return set()
+
+    def _save_participated_threads(self) -> None:
+        """Persist the current thread set to disk (best-effort)."""
+        path = self._thread_state_path()
+        try:
+            thread_list = list(self._bot_participated_threads)
+            if len(thread_list) > self._MAX_TRACKED_THREADS:
+                thread_list = thread_list[-self._MAX_TRACKED_THREADS:]
+                self._bot_participated_threads = set(thread_list)
+            path.parent.mkdir(parents=True, exist_ok=True)
+            path.write_text(json.dumps(thread_list), encoding="utf-8")
+        except Exception as e:
+            logger.debug("Could not save matrix thread state: %s", e)
+
+    def _track_thread(self, thread_id: str) -> None:
+        """Add a thread to the participation set and persist."""
+        if thread_id not in self._bot_participated_threads:
+            self._bot_participated_threads.add(thread_id)
+            self._save_participated_threads()
+
+    # ------------------------------------------------------------------
+    # Mention detection helpers
+    # ------------------------------------------------------------------
+
+    def _is_bot_mentioned(self, body: str, formatted_body: Optional[str] = None) -> bool:
+        """Return True if the bot is mentioned in the message."""
+        if not body and not formatted_body:
+            return False
+        # Check for full @user:server in body
+        if self._user_id and self._user_id in body:
+            return True
+        # Check for localpart with word boundaries (case-insensitive)
+        if self._user_id and ":" in self._user_id:
+            localpart = self._user_id.split(":")[0].lstrip("@")
+            if localpart and re.search(r'\b' + re.escape(localpart) + r'\b', body, re.IGNORECASE):
+                return True
+        # Check formatted_body for Matrix pill
+        if formatted_body and self._user_id:
+            if f"matrix.to/#/{self._user_id}" in formatted_body:
+                return True
+        return False
+
+    def _strip_mention(self, body: str) -> str:
+        """Remove bot mention from message body."""
+        # Remove full @user:server
+        if self._user_id:
+            body = body.replace(self._user_id, "")
+        # If still contains localpart mention, remove it
+        if self._user_id and ":" in self._user_id:
+            localpart = self._user_id.split(":")[0].lstrip("@")
+            if localpart:
+                body = re.sub(r'\b' + re.escape(localpart) + r'\b', '', body, flags=re.IGNORECASE)
+        return body.strip()
+
    def _get_display_name(self, room: Any, user_id: str) -> str:
        """Get a user's display name in a room, falling back to user_id."""
        if room and hasattr(room, "users"):
@@ -13,6 +13,7 @@ import json
 import logging
 import os
 import re
+import time
 from typing import Dict, Optional, Any

 try:
@@ -78,6 +79,11 @@ class SlackAdapter(BasePlatformAdapter):
        self._team_clients: Dict[str, AsyncWebClient] = {}   # team_id → WebClient
        self._team_bot_user_ids: Dict[str, str] = {}          # team_id → bot_user_id
        self._channel_team: Dict[str, str] = {}                # channel_id → team_id
+        # Dedup cache: event_ts → timestamp.  Prevents duplicate bot
+        # responses when Socket Mode reconnects redeliver events.
+        self._seen_messages: Dict[str, float] = {}
+        self._SEEN_TTL = 300   # 5 minutes
+        self._SEEN_MAX = 2000  # prune threshold

    async def connect(self) -> bool:
        """Connect to Slack via Socket Mode."""
@@ -323,7 +329,18 @@ class SlackAdapter(BasePlatformAdapter):

        Prefers metadata thread_id (the thread parent's ts, set by the
        gateway) over reply_to (which may be a child message's ts).
+
+        When ``reply_in_thread`` is ``false`` in the platform extra config,
+        top-level channel messages receive direct channel replies instead of
+        thread replies.  Messages that originate inside an existing thread are
+        always replied to in-thread to preserve conversation context.
        """
+        # When reply_in_thread is disabled (default: True for backward compat),
+        # only thread messages that are already part of an existing thread.
+        if not self.config.extra.get("reply_in_thread", True):
+            existing_thread = (metadata or {}).get("thread_id") or (metadata or {}).get("thread_ts")
+            return existing_thread or None
+
        if metadata:
            if metadata.get("thread_id"):
                return metadata["thread_id"]
@@ -699,6 +716,20 @@ class SlackAdapter(BasePlatformAdapter):

    async def _handle_slack_message(self, event: dict) -> None:
        """Handle an incoming Slack message event."""
+        # Dedup: Slack Socket Mode can redeliver events after reconnects (#4777)
+        event_ts = event.get("ts", "")
+        if event_ts:
+            now = time.time()
+            if event_ts in self._seen_messages:
+                return
+            self._seen_messages[event_ts] = now
+            if len(self._seen_messages) > self._SEEN_MAX:
+                cutoff = now - self._SEEN_TTL
+                self._seen_messages = {
+                    k: v for k, v in self._seen_messages.items()
+                    if v > cutoff
+                }
+
        # Ignore bot messages (including our own)
        if event.get("bot_id") or event.get("subtype") == "bot_message":
            return
@@ -900,7 +900,9 @@ class TelegramAdapter(BasePlatformAdapter):
                except Exception:
                    pass  # best-effort truncation
                return SendResult(success=True, message_id=message_id)
-            # Flood control / RetryAfter — back off and retry once
+            # Flood control / RetryAfter — short waits are retried inline,
+            # long waits return a failure immediately so streaming can fall back
+            # to a normal final send instead of leaving a truncated partial.
            retry_after = getattr(e, "retry_after", None)
            if retry_after is not None or "retry after" in err_str:
                wait = retry_after if retry_after else 1.0
@@ -908,6 +910,8 @@ class TelegramAdapter(BasePlatformAdapter):
                    "[%s] Telegram flood control, waiting %.1fs",
                    self.name, wait,
                )
+                if wait > 5.0:
+                    return SendResult(success=False, error=f"flood_control:{wait}")
                await asyncio.sleep(wait)
                try:
                    await self._bot.edit_message_text(
@@ -2097,6 +2101,19 @@ class TelegramAdapter(BasePlatformAdapter):
                    if not chat_topic:
                        chat_topic = created_name

+        elif chat_type == "group" and thread_id_str:
+            # Group/supergroup forum topic skill binding via config.extra['group_topics']
+            group_topics_config: list = self.config.extra.get("group_topics", [])
+            for chat_entry in group_topics_config:
+                if str(chat_entry.get("chat_id", "")) == str(chat.id):
+                    for topic in chat_entry.get("topics", []):
+                        tid = topic.get("thread_id")
+                        if tid is not None and str(tid) == thread_id_str:
+                            chat_topic = topic.get("name")
+                            topic_skill = topic.get("skill")
+                            break
+                    break
+
        # Build source
        source = self.build_source(
            chat_id=str(chat.id),
@@ -16,9 +16,11 @@ with different backends via a bridge pattern.
 """

 import asyncio
+import json
 import logging
 import os
 import platform
+import re
 import subprocess

 _IS_WINDOWS = platform.system() == "Windows"
@@ -138,12 +140,137 @@ class WhatsAppAdapter(BasePlatformAdapter):
            get_hermes_dir("platforms/whatsapp/session", "whatsapp/session")
        ))
        self._reply_prefix: Optional[str] = config.extra.get("reply_prefix")
+        self._mention_patterns = self._compile_mention_patterns()
        self._message_queue: asyncio.Queue = asyncio.Queue()
        self._bridge_log_fh = None
        self._bridge_log: Optional[Path] = None
        self._poll_task: Optional[asyncio.Task] = None
        self._http_session: Optional["aiohttp.ClientSession"] = None
        self._session_lock_identity: Optional[str] = None
+
+    def _whatsapp_require_mention(self) -> bool:
+        configured = self.config.extra.get("require_mention")
+        if configured is not None:
+            if isinstance(configured, str):
+                return configured.lower() in ("true", "1", "yes", "on")
+            return bool(configured)
+        return os.getenv("WHATSAPP_REQUIRE_MENTION", "false").lower() in ("true", "1", "yes", "on")
+
+    def _whatsapp_free_response_chats(self) -> set[str]:
+        raw = self.config.extra.get("free_response_chats")
+        if raw is None:
+            raw = os.getenv("WHATSAPP_FREE_RESPONSE_CHATS", "")
+        if isinstance(raw, list):
+            return {str(part).strip() for part in raw if str(part).strip()}
+        return {part.strip() for part in str(raw).split(",") if part.strip()}
+
+    def _compile_mention_patterns(self):
+        patterns = self.config.extra.get("mention_patterns")
+        if patterns is None:
+            raw = os.getenv("WHATSAPP_MENTION_PATTERNS", "").strip()
+            if raw:
+                try:
+                    patterns = json.loads(raw)
+                except Exception:
+                    patterns = [part.strip() for part in raw.splitlines() if part.strip()]
+                    if not patterns:
+                        patterns = [part.strip() for part in raw.split(",") if part.strip()]
+        if patterns is None:
+            return []
+        if isinstance(patterns, str):
+            patterns = [patterns]
+        if not isinstance(patterns, list):
+            logger.warning("[%s] whatsapp mention_patterns must be a list or string; got %s", self.name, type(patterns).__name__)
+            return []
+
+        compiled = []
+        for pattern in patterns:
+            if not isinstance(pattern, str) or not pattern.strip():
+                continue
+            try:
+                compiled.append(re.compile(pattern, re.IGNORECASE))
+            except re.error as exc:
+                logger.warning("[%s] Invalid WhatsApp mention pattern %r: %s", self.name, pattern, exc)
+        if compiled:
+            logger.info("[%s] Loaded %d WhatsApp mention pattern(s)", self.name, len(compiled))
+        return compiled
+
+    @staticmethod
+    def _normalize_whatsapp_id(value: Optional[str]) -> str:
+        if not value:
+            return ""
+        normalized = str(value).strip()
+        if ":" in normalized and "@" in normalized:
+            normalized = normalized.replace(":", "@", 1)
+        return normalized
+
+    def _bot_ids_from_message(self, data: Dict[str, Any]) -> set[str]:
+        bot_ids = set()
+        for candidate in data.get("botIds") or []:
+            normalized = self._normalize_whatsapp_id(candidate)
+            if normalized:
+                bot_ids.add(normalized)
+        return bot_ids
+
+    def _message_is_reply_to_bot(self, data: Dict[str, Any]) -> bool:
+        quoted_participant = self._normalize_whatsapp_id(data.get("quotedParticipant"))
+        if not quoted_participant:
+            return False
+        return quoted_participant in self._bot_ids_from_message(data)
+
+    def _message_mentions_bot(self, data: Dict[str, Any]) -> bool:
+        bot_ids = self._bot_ids_from_message(data)
+        if not bot_ids:
+            return False
+        mentioned_ids = {
+            nid
+            for candidate in (data.get("mentionedIds") or [])
+            if (nid := self._normalize_whatsapp_id(candidate))
+        }
+        if mentioned_ids & bot_ids:
+            return True
+
+        body = str(data.get("body") or "")
+        lower_body = body.lower()
+        for bot_id in bot_ids:
+            bare_id = bot_id.split("@", 1)[0].lower()
+            if bare_id and (f"@{bare_id}" in lower_body or bare_id in lower_body):
+                return True
+        return False
+
+    def _message_matches_mention_patterns(self, data: Dict[str, Any]) -> bool:
+        if not self._mention_patterns:
+            return False
+        body = str(data.get("body") or "")
+        return any(pattern.search(body) for pattern in self._mention_patterns)
+
+    def _clean_bot_mention_text(self, text: str, data: Dict[str, Any]) -> str:
+        if not text:
+            return text
+        bot_ids = self._bot_ids_from_message(data)
+        cleaned = text
+        for bot_id in bot_ids:
+            bare_id = bot_id.split("@", 1)[0]
+            if bare_id:
+                cleaned = re.sub(rf"@{re.escape(bare_id)}\b[,:\-]*\s*", "", cleaned)
+        return cleaned.strip() or text
+
+    def _should_process_message(self, data: Dict[str, Any]) -> bool:
+        if not data.get("isGroup"):
+            return True
+        chat_id = str(data.get("chatId") or "")
+        if chat_id in self._whatsapp_free_response_chats():
+            return True
+        if not self._whatsapp_require_mention():
+            return True
+        body = str(data.get("body") or "").strip()
+        if body.startswith("/"):
+            return True
+        if self._message_is_reply_to_bot(data):
+            return True
+        if self._message_mentions_bot(data):
+            return True
+        return self._message_matches_mention_patterns(data)
    
    async def connect(self) -> bool:
        """
@@ -687,6 +814,9 @@ class WhatsAppAdapter(BasePlatformAdapter):
    async def _build_message_event(self, data: Dict[str, Any]) -> Optional[MessageEvent]:
        """Build a MessageEvent from bridge message data, downloading images to cache."""
        try:
+            if not self._should_process_message(data):
+                return None
+
            # Determine message type
            msg_type = MessageType.TEXT
            if data.get("hasMedia"):
@@ -768,6 +898,8 @@ class WhatsAppAdapter(BasePlatformAdapter):
            # the message text so the agent can read it inline.
            # Cap at 100KB to match Telegram/Discord/Slack behaviour.
            body = data.get("body", "")
+            if data.get("isGroup"):
+                body = self._clean_bot_mention_text(body, data)
            MAX_TEXT_INJECT_BYTES = 100 * 1024
            if msg_type == MessageType.DOCUMENT and cached_urls:
                for doc_path in cached_urls:
@@ -364,6 +364,12 @@ class SessionEntry:
    auto_reset_reason: Optional[str] = None  # "idle" or "daily"
    reset_had_activity: bool = False  # whether the expired session had any messages
    
+    # Set by the background expiry watcher after it successfully flushes
+    # memories for this session.  Persisted to sessions.json so the flag
+    # survives gateway restarts (the old in-memory _pre_flushed_sessions
+    # set was lost on restart, causing redundant re-flushes).
+    memory_flushed: bool = False
+    
    def to_dict(self) -> Dict[str, Any]:
        result = {
            "session_key": self.session_key,
@@ -381,6 +387,7 @@ class SessionEntry:
            "last_prompt_tokens": self.last_prompt_tokens,
            "estimated_cost_usd": self.estimated_cost_usd,
            "cost_status": self.cost_status,
+            "memory_flushed": self.memory_flushed,
        }
        if self.origin:
            result["origin"] = self.origin.to_dict()
@@ -416,6 +423,7 @@ class SessionEntry:
            last_prompt_tokens=data.get("last_prompt_tokens", 0),
            estimated_cost_usd=data.get("estimated_cost_usd", 0.0),
            cost_status=data.get("cost_status", "unknown"),
+            memory_flushed=data.get("memory_flushed", False),
        )


@@ -479,9 +487,6 @@ class SessionStore:
        self._loaded = False
        self._lock = threading.Lock()
        self._has_active_processes_fn = has_active_processes_fn
-        # on_auto_reset is deprecated — memory flush now runs proactively
-        # via the background session expiry watcher in GatewayRunner.
-        self._pre_flushed_sessions: set = set()  # session_ids already flushed by watcher
        
        # Initialize SQLite session database
        self._db = None
@@ -684,15 +689,12 @@ class SessionStore:
                    self._save()
                    return entry
                else:
-                    # Session is being auto-reset.  The background expiry watcher
-                    # should have already flushed memories proactively; discard
-                    # the marker so it doesn't accumulate.
+                    # Session is being auto-reset.
                    was_auto_reset = True
                    auto_reset_reason = reset_reason
                    # Track whether the expired session had any real conversation
                    reset_had_activity = entry.total_tokens > 0
                    db_end_session_id = entry.session_id
-                    self._pre_flushed_sessions.discard(entry.session_id)
            else:
                was_auto_reset = False
                auto_reset_reason = None
@@ -736,71 +738,58 @@ class SessionStore:
            except Exception as e:
                print(f"[gateway] Warning: Failed to create SQLite session: {e}")

+        # Seed new DM thread sessions with parent DM session history.
+        # When a bot reply creates a Slack thread and the user responds in it,
+        # the thread gets a new session (keyed by thread_ts).  Without seeding,
+        # the thread session starts with zero context — the user's original
+        # question and the bot's answer are invisible.  Fix: copy the parent
+        # DM session's transcript into the new thread session so context carries
+        # over while still keeping threads isolated from each other.
+        if (
+            source.chat_type == "dm"
+            and source.thread_id
+            and entry.created_at == entry.updated_at  # brand-new session
+            and not was_auto_reset
+        ):
+            parent_source = SessionSource(
+                platform=source.platform,
+                chat_id=source.chat_id,
+                chat_type="dm",
+                user_id=source.user_id,
+                # no thread_id — this is the parent DM session
+            )
+            parent_key = self._generate_session_key(parent_source)
+            with self._lock:
+                parent_entry = self._entries.get(parent_key)
+            if parent_entry and parent_entry.session_id != entry.session_id:
+                try:
+                    parent_history = self.load_transcript(parent_entry.session_id)
+                    if parent_history:
+                        self.rewrite_transcript(entry.session_id, parent_history)
+                        logger.info(
+                            "[Session] Seeded DM thread session %s with %d messages from parent %s",
+                            entry.session_id, len(parent_history), parent_entry.session_id,
+                        )
+                except Exception as e:
+                    logger.warning("[Session] Failed to seed thread session: %s", e)
+
        return entry

    def update_session(
        self,
        session_key: str,
-        input_tokens: int = 0,
-        output_tokens: int = 0,
-        cache_read_tokens: int = 0,
-        cache_write_tokens: int = 0,
        last_prompt_tokens: int = None,
-        model: str = None,
-        estimated_cost_usd: Optional[float] = None,
-        cost_status: Optional[str] = None,
-        cost_source: Optional[str] = None,
-        provider: Optional[str] = None,
-        base_url: Optional[str] = None,
    ) -> None:
-        """Update a session's metadata after an interaction."""
-        db_session_id = None
-
+        """Update lightweight session metadata after an interaction."""
        with self._lock:
            self._ensure_loaded_locked()

            if session_key in self._entries:
                entry = self._entries[session_key]
                entry.updated_at = _now()
-                # Direct assignment — the gateway receives cumulative totals
-                # from the cached agent, not per-call deltas.
-                entry.input_tokens = input_tokens
-                entry.output_tokens = output_tokens
-                entry.cache_read_tokens = cache_read_tokens
-                entry.cache_write_tokens = cache_write_tokens
                if last_prompt_tokens is not None:
                    entry.last_prompt_tokens = last_prompt_tokens
-                if estimated_cost_usd is not None:
-                    entry.estimated_cost_usd = estimated_cost_usd
-                if cost_status:
-                    entry.cost_status = cost_status
-                entry.total_tokens = (
-                    entry.input_tokens
-                    + entry.output_tokens
-                    + entry.cache_read_tokens
-                    + entry.cache_write_tokens
-                )
                self._save()
-                db_session_id = entry.session_id
-
-        if self._db and db_session_id:
-            try:
-                self._db.set_token_counts(
-                    db_session_id,
-                    input_tokens=input_tokens,
-                    output_tokens=output_tokens,
-                    cache_read_tokens=cache_read_tokens,
-                    cache_write_tokens=cache_write_tokens,
-                    estimated_cost_usd=estimated_cost_usd,
-                    cost_status=cost_status,
-                    cost_source=cost_source,
-                    billing_provider=provider,
-                    billing_base_url=base_url,
-                    model=model,
-                    absolute=True,
-                )
-            except Exception as e:
-                logger.debug("Session DB operation failed: %s", e)

    def reset_session(self, session_key: str) -> Optional[SessionEntry]:
        """Force reset a session, creating a new session ID."""
@@ -174,12 +174,12 @@ class GatewayStreamConsumer:
                        self._already_sent = True
                        self._last_sent_text = text
                    else:
-                        # Edit not supported by this adapter — stop streaming,
-                        # let the normal send path handle the final response.
-                        # Without this guard, adapters like Signal/Email would
-                        # flood the chat with a new message every edit_interval.
+                        # If an edit fails mid-stream (especially Telegram flood control),
+                        # stop progressive edits and let the normal final send path deliver
+                        # the complete answer instead of leaving the user with a partial.
                        logger.debug("Edit failed, disabling streaming for this adapter")
                        self._edit_supported = False
+                        self._already_sent = False
                else:
                    # Editing not supported — skip intermediate updates.
                    # The final response will be sent by the normal path.
@@ -11,5 +11,5 @@ Provides subcommands for:
 - hermes cron          - Manage cron jobs
 """

-__version__ = "0.6.0"
-__release_date__ = "2026.3.30"
+__version__ = "0.7.0"
+__release_date__ = "2026.4.3"
@@ -200,6 +200,10 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
        id="opencode-go",
        name="OpenCode Go",
        auth_type="api_key",
+        # OpenCode Go mixes API surfaces by model:
+        # - GLM / Kimi use OpenAI-compatible chat completions under /v1
+        # - MiniMax models use Anthropic Messages under /v1/messages
+        # Keep the provider base at /v1 and select api_mode per-model.
        inference_base_url="https://opencode.ai/zen/go/v1",
        api_key_env_vars=("OPENCODE_GO_API_KEY",),
        base_url_env_var="OPENCODE_GO_BASE_URL",
@@ -1377,6 +1381,89 @@ def _agent_key_is_usable(state: Dict[str, Any], min_ttl_seconds: int) -> bool:
    return not _is_expiring(state.get("agent_key_expires_at"), min_ttl_seconds)


+def resolve_nous_access_token(
+    *,
+    timeout_seconds: float = 15.0,
+    insecure: Optional[bool] = None,
+    ca_bundle: Optional[str] = None,
+    refresh_skew_seconds: int = ACCESS_TOKEN_REFRESH_SKEW_SECONDS,
+) -> str:
+    """Resolve a refresh-aware Nous Portal access token for managed tool gateways."""
+    with _auth_store_lock():
+        auth_store = _load_auth_store()
+        state = _load_provider_state(auth_store, "nous")
+
+        if not state:
+            raise AuthError(
+                "Hermes is not logged into Nous Portal.",
+                provider="nous",
+                relogin_required=True,
+            )
+
+        portal_base_url = (
+            _optional_base_url(state.get("portal_base_url"))
+            or os.getenv("HERMES_PORTAL_BASE_URL")
+            or os.getenv("NOUS_PORTAL_BASE_URL")
+            or DEFAULT_NOUS_PORTAL_URL
+        ).rstrip("/")
+        client_id = str(state.get("client_id") or DEFAULT_NOUS_CLIENT_ID)
+        verify = _resolve_verify(insecure=insecure, ca_bundle=ca_bundle, auth_state=state)
+
+        access_token = state.get("access_token")
+        refresh_token = state.get("refresh_token")
+        if not isinstance(access_token, str) or not access_token:
+            raise AuthError(
+                "No access token found for Nous Portal login.",
+                provider="nous",
+                relogin_required=True,
+            )
+
+        if not _is_expiring(state.get("expires_at"), refresh_skew_seconds):
+            return access_token
+
+        if not isinstance(refresh_token, str) or not refresh_token:
+            raise AuthError(
+                "Session expired and no refresh token is available.",
+                provider="nous",
+                relogin_required=True,
+            )
+
+        timeout = httpx.Timeout(timeout_seconds if timeout_seconds else 15.0)
+        with httpx.Client(
+            timeout=timeout,
+            headers={"Accept": "application/json"},
+            verify=verify,
+        ) as client:
+            refreshed = _refresh_access_token(
+                client=client,
+                portal_base_url=portal_base_url,
+                client_id=client_id,
+                refresh_token=refresh_token,
+            )
+
+        now = datetime.now(timezone.utc)
+        access_ttl = _coerce_ttl_seconds(refreshed.get("expires_in"))
+        state["access_token"] = refreshed["access_token"]
+        state["refresh_token"] = refreshed.get("refresh_token") or refresh_token
+        state["token_type"] = refreshed.get("token_type") or state.get("token_type") or "Bearer"
+        state["scope"] = refreshed.get("scope") or state.get("scope")
+        state["obtained_at"] = now.isoformat()
+        state["expires_in"] = access_ttl
+        state["expires_at"] = datetime.fromtimestamp(
+            now.timestamp() + access_ttl,
+            tz=timezone.utc,
+        ).isoformat()
+        state["portal_base_url"] = portal_base_url
+        state["client_id"] = client_id
+        state["tls"] = {
+            "insecure": verify is False,
+            "ca_bundle": verify if isinstance(verify, str) else None,
+        }
+        _save_provider_state(auth_store, "nous", state)
+        _save_auth_store(auth_store)
+        return state["access_token"]
+
+
 def refresh_nous_oauth_pure(
    access_token: str,
    refresh_token: str,
@@ -57,6 +57,8 @@ COMMAND_REGISTRY: list[CommandDef] = [
    CommandDef("undo", "Remove the last user/assistant exchange", "Session"),
    CommandDef("title", "Set a title for the current session", "Session",
               args_hint="[name]"),
+    CommandDef("branch", "Branch the current session (explore a different path)", "Session",
+               aliases=("fork",), args_hint="[name]"),
    CommandDef("compress", "Manually compress conversation context", "Session"),
    CommandDef("rollback", "List or restore filesystem checkpoints", "Session",
               args_hint="[number]"),
@@ -414,6 +416,8 @@ def telegram_menu_commands(max_commands: int = 100) -> tuple[list[tuple[str, str

    Skills are the only tier that gets trimmed when the cap is hit.
    User-installed hub skills are excluded — accessible via /skills.
+    Skills disabled for the ``"telegram"`` platform (via ``hermes skills
+    config``) are excluded from the menu entirely.

    Returns:
        (menu_commands, hidden_count) where hidden_count is the number of
@@ -444,6 +448,17 @@ def telegram_menu_commands(max_commands: int = 100) -> tuple[list[tuple[str, str
    reserved_names.update(n for n, _ in plugin_entries)
    all_commands.extend(plugin_entries)

+    # Load per-platform disabled skills so they don't consume menu slots.
+    # get_skill_commands() already filters the *global* disabled list, but
+    # per-platform overrides (skills.platform_disabled.telegram) were never
+    # applied here — that's what this block fixes.
+    _platform_disabled: set[str] = set()
+    try:
+        from agent.skill_utils import get_disabled_skill_names
+        _platform_disabled = get_disabled_skill_names(platform="telegram")
+    except Exception:
+        pass
+
    # Remaining slots go to built-in skill commands (not hub-installed).
    skill_entries: list[tuple[str, str]] = []
    try:
@@ -459,6 +474,10 @@ def telegram_menu_commands(max_commands: int = 100) -> tuple[list[tuple[str, str
                continue
            if skill_path.startswith(_hub_dir):
                continue
+            # Skip skills disabled for telegram
+            skill_name = info.get("name", "")
+            if skill_name in _platform_disabled:
+                continue
            name = cmd_key.lstrip("/").replace("-", "_")
            desc = info.get("description", "")
            # Keep descriptions short — setMyCommands has an undocumented
@@ -22,6 +22,8 @@ import tempfile
 from pathlib import Path
 from typing import Dict, Any, Optional, List, Tuple

+from tools.tool_backend_helpers import managed_nous_tools_enabled as _managed_nous_tools_enabled
+
 _IS_WINDOWS = platform.system() == "Windows"
 _ENV_VAR_NAME_RE = re.compile(r"^[A-Za-z_][A-Za-z0-9_]*$")
 # Env var names written to .env that aren't in OPTIONAL_ENV_VARS
@@ -40,8 +42,8 @@ _EXTRA_ENV_KEYS = frozenset({
    "WHATSAPP_MODE", "WHATSAPP_ENABLED",
    "MATTERMOST_HOME_CHANNEL", "MATTERMOST_REPLY_MODE",
    "MATRIX_PASSWORD", "MATRIX_ENCRYPTION", "MATRIX_HOME_ROOM",
+    "MATRIX_REQUIRE_MENTION", "MATRIX_FREE_RESPONSE_ROOMS", "MATRIX_AUTO_THREAD",
 })
-
 import yaml

 from hermes_cli.colors import Colors, color
@@ -196,7 +198,7 @@ def ensure_hermes_home():
 # =============================================================================

 DEFAULT_CONFIG = {
-    "model": "anthropic/claude-opus-4.6",
+    "model": "",
    "fallback_providers": [],
    "credential_pool_strategies": {},
    "toolsets": ["hermes-cli"],
@@ -212,6 +214,7 @@ DEFAULT_CONFIG = {
    
    "terminal": {
        "backend": "local",
+        "modal_mode": "auto",
        "cwd": ".",  # Use current directory
        "timeout": 180,
        # Environment variables to pass through to sandboxed execution
@@ -220,6 +223,12 @@ DEFAULT_CONFIG = {
        "env_passthrough": [],
        "docker_image": "nikolaik/python-nodejs:python3.11-nodejs20",
        "docker_forward_env": [],
+        # Explicit environment variables to set inside Docker containers.
+        # Unlike docker_forward_env (which reads values from the host process),
+        # docker_env lets you specify exact key-value pairs — useful when Hermes
+        # runs as a systemd service without access to the user's shell environment.
+        # Example: {"SSH_AUTH_SOCK": "/run/user/1000/ssh-agent.sock"}
+        "docker_env": {},
        "singularity_image": "docker://nikolaik/python-nodejs:python3.11-nodejs20",
        "modal_image": "nikolaik/python-nodejs:python3.11-nodejs20",
        "daytona_image": "nikolaik/python-nodejs:python3.11-nodejs20",
@@ -247,6 +256,13 @@ DEFAULT_CONFIG = {
        "command_timeout": 30,  # Timeout for browser commands in seconds (screenshot, navigate, etc.)
        "record_sessions": False,  # Auto-record browser sessions as WebM videos
        "allow_private_urls": False,  # Allow navigating to private/internal IPs (localhost, 192.168.x.x, etc.)
+        "camofox": {
+            # When true, Hermes sends a stable profile-scoped userId to Camofox
+            # so the server can map it to a persistent browser profile directory.
+            # Requires Camofox server to be configured with CAMOFOX_PROFILE_DIR.
+            # When false (default), each session gets a random userId (ephemeral).
+            "managed_persistence": False,
+        },
    },

    # Filesystem checkpoints — automatic snapshots before destructive file ops.
@@ -419,6 +435,11 @@ DEFAULT_CONFIG = {
        "user_profile_enabled": True,
        "memory_char_limit": 2200,   # ~800 tokens at 2.75 chars/token
        "user_char_limit": 1375,     # ~500 tokens at 2.75 chars/token
+        # External memory provider plugin (empty = built-in only).
+        # Set to a provider name to activate: "openviking", "mem0",
+        # "hindsight", "holographic", "retaindb", "byterover".
+        # Only ONE external provider is allowed at a time.
+        "provider": "",
    },

    # Subagent delegation — override the provider:model used by delegate_task
@@ -525,6 +546,7 @@ ENV_VARS_BY_VERSION: Dict[int, List[str]] = {
    5: ["WHATSAPP_ENABLED", "WHATSAPP_MODE", "WHATSAPP_ALLOWED_USERS",
        "SLACK_BOT_TOKEN", "SLACK_APP_TOKEN", "SLACK_ALLOWED_USERS"],
    10: ["TAVILY_API_KEY"],
+    11: ["TERMINAL_MODAL_MODE"],
 }

 # Required environment variables with metadata for migration prompts.
@@ -743,6 +765,38 @@ OPTIONAL_ENV_VARS = {
        "category": "tool",
        "advanced": True,
    },
+    "FIRECRAWL_GATEWAY_URL": {
+        "description": "Exact Firecrawl tool-gateway origin override for Nous Subscribers only (optional)",
+        "prompt": "Firecrawl gateway URL (leave empty to derive from domain)",
+        "url": None,
+        "password": False,
+        "category": "tool",
+        "advanced": True,
+    },
+    "TOOL_GATEWAY_DOMAIN": {
+        "description": "Shared tool-gateway domain suffix for Nous Subscribers only, used to derive vendor hosts, e.g. nousresearch.com -> firecrawl-gateway.nousresearch.com",
+        "prompt": "Tool-gateway domain suffix",
+        "url": None,
+        "password": False,
+        "category": "tool",
+        "advanced": True,
+    },
+    "TOOL_GATEWAY_SCHEME": {
+        "description": "Shared tool-gateway URL scheme for Nous Subscribers only, used to derive vendor hosts (`https` by default, set `http` for local gateway testing)",
+        "prompt": "Tool-gateway URL scheme",
+        "url": None,
+        "password": False,
+        "category": "tool",
+        "advanced": True,
+    },
+    "TOOL_GATEWAY_USER_TOKEN": {
+        "description": "Explicit Nous Subscriber access token for tool-gateway requests (optional; otherwise read from the Hermes auth store)",
+        "prompt": "Tool-gateway user token",
+        "url": None,
+        "password": True,
+        "category": "tool",
+        "advanced": True,
+    },
    "TAVILY_API_KEY": {
        "description": "Tavily API key for AI-native web search, extract, and crawl",
        "prompt": "Tavily API key",
@@ -955,6 +1009,30 @@ OPTIONAL_ENV_VARS = {
        "password": False,
        "category": "messaging",
    },
+    "MATRIX_REQUIRE_MENTION": {
+        "description": "Require @mention in Matrix rooms (default: true). Set to false to respond to all messages.",
+        "prompt": "Require @mention in rooms (true/false)",
+        "url": None,
+        "password": False,
+        "category": "messaging",
+        "advanced": True,
+    },
+    "MATRIX_FREE_RESPONSE_ROOMS": {
+        "description": "Comma-separated Matrix room IDs where bot responds without @mention",
+        "prompt": "Free-response room IDs (comma-separated)",
+        "url": None,
+        "password": False,
+        "category": "messaging",
+        "advanced": True,
+    },
+    "MATRIX_AUTO_THREAD": {
+        "description": "Auto-create threads for messages in Matrix rooms (default: true)",
+        "prompt": "Auto-create threads in rooms (true/false)",
+        "url": None,
+        "password": False,
+        "category": "messaging",
+        "advanced": True,
+    },
    "GATEWAY_ALLOW_ALL_USERS": {
        "description": "Allow all users to interact with messaging bots (true/false). Default: false.",
        "prompt": "Allow all users (true/false)",
@@ -1072,6 +1150,15 @@ OPTIONAL_ENV_VARS = {
    },
 }

+if not _managed_nous_tools_enabled():
+    for _hidden_var in (
+        "FIRECRAWL_GATEWAY_URL",
+        "TOOL_GATEWAY_DOMAIN",
+        "TOOL_GATEWAY_SCHEME",
+        "TOOL_GATEWAY_USER_TOKEN",
+    ):
+        OPTIONAL_ENV_VARS.pop(_hidden_var, None)
+

 def get_missing_env_vars(required_only: bool = False) -> List[Dict[str, Any]]:
    """
@@ -1987,7 +2074,9 @@ def set_config_value(key: str, value: str):
    # Check if it's an API key (goes to .env)
    api_keys = [
        'OPENROUTER_API_KEY', 'OPENAI_API_KEY', 'ANTHROPIC_API_KEY', 'VOICE_TOOLS_OPENAI_KEY',
-        'EXA_API_KEY', 'PARALLEL_API_KEY', 'FIRECRAWL_API_KEY', 'FIRECRAWL_API_URL', 'TAVILY_API_KEY',
+        'EXA_API_KEY', 'PARALLEL_API_KEY', 'FIRECRAWL_API_KEY', 'FIRECRAWL_API_URL',
+        'FIRECRAWL_GATEWAY_URL', 'TOOL_GATEWAY_DOMAIN', 'TOOL_GATEWAY_SCHEME',
+        'TOOL_GATEWAY_USER_TOKEN', 'TAVILY_API_KEY',
        'BROWSERBASE_API_KEY', 'BROWSERBASE_PROJECT_ID', 'BROWSER_USE_API_KEY',
        'FAL_KEY', 'TELEGRAM_BOT_TOKEN', 'DISCORD_BOT_TOKEN',
        'TERMINAL_SSH_HOST', 'TERMINAL_SSH_USER', 'TERMINAL_SSH_KEY',
@@ -2043,6 +2132,7 @@ def set_config_value(key: str, value: str):
    # config.yaml is authoritative, but terminal_tool only reads TERMINAL_ENV etc.
    _config_to_env_sync = {
        "terminal.backend": "TERMINAL_ENV",
+        "terminal.modal_mode": "TERMINAL_MODAL_MODE",
        "terminal.docker_image": "TERMINAL_DOCKER_IMAGE",
        "terminal.singularity_image": "TERMINAL_SINGULARITY_IMAGE",
        "terminal.modal_image": "TERMINAL_MODAL_IMAGE",
@@ -90,6 +90,9 @@ def cron_list(show_all: bool = False):
        print(f"    Deliver:   {deliver_str}")
        if skills:
            print(f"    Skills:    {', '.join(skills)}")
+        script = job.get("script")
+        if script:
+            print(f"    Script:    {script}")
        print()

    from hermes_cli.gateway import find_gateway_pids
@@ -149,6 +152,7 @@ def cron_create(args):
        repeat=getattr(args, "repeat", None),
        skill=getattr(args, "skill", None),
        skills=_normalize_skills(getattr(args, "skill", None), getattr(args, "skills", None)),
+        script=getattr(args, "script", None),
    )
    if not result.get("success"):
        print(color(f"Failed to create job: {result.get('error', 'unknown error')}", Colors.RED))
@@ -158,6 +162,9 @@ def cron_create(args):
    print(f"  Schedule: {result['schedule']}")
    if result.get("skills"):
        print(f"  Skills: {', '.join(result['skills'])}")
+    job_data = result.get("job", {})
+    if job_data.get("script"):
+        print(f"  Script: {job_data['script']}")
    print(f"  Next run: {result['next_run_at']}")
    return 0

@@ -195,6 +202,7 @@ def cron_edit(args):
        deliver=getattr(args, "deliver", None),
        repeat=getattr(args, "repeat", None),
        skills=final_skills,
+        script=getattr(args, "script", None),
    )
    if not result.get("success"):
        print(color(f"Failed to update job: {result.get('error', 'unknown error')}", Colors.RED))
@@ -208,6 +216,8 @@ def cron_edit(args):
        print(f"  Skills: {', '.join(updated['skills'])}")
    else:
        print("  Skills: none")
+    if updated.get("script"):
+        print(f"  Script: {updated['script']}")
    return 0


@@ -37,6 +37,7 @@ _PROVIDER_ENV_HINTS = (
    "ANTHROPIC_API_KEY",
    "ANTHROPIC_TOKEN",
    "OPENAI_BASE_URL",
+    "NOUS_API_KEY",
    "GLM_API_KEY",
    "ZAI_API_KEY",
    "Z_AI_API_KEY",
@@ -44,6 +45,12 @@ _PROVIDER_ENV_HINTS = (
    "MINIMAX_API_KEY",
    "MINIMAX_CN_API_KEY",
    "KILOCODE_API_KEY",
+    "DEEPSEEK_API_KEY",
+    "DASHSCOPE_API_KEY",
+    "HF_TOKEN",
+    "AI_GATEWAY_API_KEY",
+    "OPENCODE_ZEN_API_KEY",
+    "OPENCODE_GO_API_KEY",
 )


@@ -55,7 +62,7 @@ def _has_provider_env_config(content: str) -> bool:
 def _honcho_is_configured_for_doctor() -> bool:
    """Return True when Honcho is configured, even if this process has no active session."""
    try:
-        from honcho_integration.client import HonchoClientConfig
+        from plugins.memory.honcho.client import HonchoClientConfig

        cfg = HonchoClientConfig.from_global_config()
        return bool(cfg.enabled and (cfg.api_key or cfg.base_url))
@@ -257,7 +264,60 @@ def run_doctor(args):
                manual_issues.append(f"Create {_DHH}/config.yaml manually")
            else:
                check_warn("config.yaml not found", "(using defaults)")
-    
+
+    # Check config version and stale keys
+    config_path = HERMES_HOME / 'config.yaml'
+    if config_path.exists():
+        try:
+            from hermes_cli.config import check_config_version, migrate_config
+            current_ver, latest_ver = check_config_version()
+            if current_ver < latest_ver:
+                check_warn(
+                    f"Config version outdated (v{current_ver} → v{latest_ver})",
+                    "(new settings available)"
+                )
+                if should_fix:
+                    try:
+                        migrate_config(interactive=False, quiet=False)
+                        check_ok("Config migrated to latest version")
+                        fixed_count += 1
+                    except Exception as mig_err:
+                        check_warn(f"Auto-migration failed: {mig_err}")
+                        issues.append("Run 'hermes setup' to migrate config")
+                else:
+                    issues.append("Run 'hermes doctor --fix' or 'hermes setup' to migrate config")
+            else:
+                check_ok(f"Config version up to date (v{current_ver})")
+        except Exception:
+            pass
+
+        # Detect stale root-level model keys (known bug source — PR #4329)
+        try:
+            import yaml
+            with open(config_path) as f:
+                raw_config = yaml.safe_load(f) or {}
+            stale_root_keys = [k for k in ("provider", "base_url") if k in raw_config and isinstance(raw_config[k], str)]
+            if stale_root_keys:
+                check_warn(
+                    f"Stale root-level config keys: {', '.join(stale_root_keys)}",
+                    "(should be under 'model:' section)"
+                )
+                if should_fix:
+                    model_section = raw_config.setdefault("model", {})
+                    for k in stale_root_keys:
+                        if not model_section.get(k):
+                            model_section[k] = raw_config.pop(k)
+                        else:
+                            raw_config.pop(k)
+                    with open(config_path, "w") as f:
+                        yaml.dump(raw_config, f, default_flow_style=False)
+                    check_ok("Migrated stale root-level keys into model section")
+                    fixed_count += 1
+                else:
+                    issues.append("Stale root-level provider/base_url in config.yaml — run 'hermes doctor --fix'")
+        except Exception:
+            pass
+
    # =========================================================================
    # Check: Auth providers
    # =========================================================================
@@ -380,6 +440,31 @@ def run_doctor(args):
    else:
        check_info(f"{_DHH}/state.db not created yet (will be created on first session)")

+    # Check WAL file size (unbounded growth indicates missed checkpoints)
+    wal_path = hermes_home / "state.db-wal"
+    if wal_path.exists():
+        try:
+            wal_size = wal_path.stat().st_size
+            if wal_size > 50 * 1024 * 1024:  # 50 MB
+                check_warn(
+                    f"WAL file is large ({wal_size // (1024*1024)} MB)",
+                    "(may indicate missed checkpoints)"
+                )
+                if should_fix:
+                    import sqlite3
+                    conn = sqlite3.connect(str(state_db_path))
+                    conn.execute("PRAGMA wal_checkpoint(PASSIVE)")
+                    conn.close()
+                    new_size = wal_path.stat().st_size if wal_path.exists() else 0
+                    check_ok(f"WAL checkpoint performed ({wal_size // 1024}K → {new_size // 1024}K)")
+                    fixed_count += 1
+                else:
+                    issues.append("Large WAL file — run 'hermes doctor --fix' to checkpoint")
+            elif wal_size > 10 * 1024 * 1024:  # 10 MB
+                check_info(f"WAL file is {wal_size // (1024*1024)} MB (normal for active sessions)")
+        except Exception:
+            pass
+
    _check_gateway_service_linger(issues)
    
    # =========================================================================
@@ -566,17 +651,22 @@ def run_doctor(args):
        except Exception as e:
            print(f"\r  {color('⚠', Colors.YELLOW)} Anthropic API {color(f'({e})', Colors.DIM)}                 ")

-    # -- API-key providers (Z.AI/GLM, Kimi, MiniMax, MiniMax-CN) --
+    # -- API-key providers --
    # Tuple: (name, env_vars, default_url, base_env, supports_models_endpoint)
    # If supports_models_endpoint is False, we skip the health check and just show "configured"
    _apikey_providers = [
        ("Z.AI / GLM",      ("GLM_API_KEY", "ZAI_API_KEY", "Z_AI_API_KEY"), "https://api.z.ai/api/paas/v4/models", "GLM_BASE_URL", True),
        ("Kimi / Moonshot",  ("KIMI_API_KEY",),                              "https://api.moonshot.ai/v1/models",   "KIMI_BASE_URL", True),
+        ("DeepSeek",         ("DEEPSEEK_API_KEY",),                           "https://api.deepseek.com/v1/models",  "DEEPSEEK_BASE_URL", True),
+        ("Hugging Face",     ("HF_TOKEN",),                                   "https://router.huggingface.co/v1/models", "HF_BASE_URL", True),
+        ("Alibaba/DashScope", ("DASHSCOPE_API_KEY",),                         "https://dashscope-intl.aliyuncs.com/compatible-mode/v1/models", "DASHSCOPE_BASE_URL", True),
        # MiniMax APIs don't support /models endpoint — https://github.com/NousResearch/hermes-agent/issues/811
        ("MiniMax",          ("MINIMAX_API_KEY",),                            None,                                  "MINIMAX_BASE_URL", False),
        ("MiniMax (China)",  ("MINIMAX_CN_API_KEY",),                         None,                                  "MINIMAX_CN_BASE_URL", False),
        ("AI Gateway",       ("AI_GATEWAY_API_KEY",),                          "https://ai-gateway.vercel.sh/v1/models", "AI_GATEWAY_BASE_URL", True),
        ("Kilo Code",        ("KILOCODE_API_KEY",),                            "https://api.kilo.ai/api/gateway/models",  "KILOCODE_BASE_URL", True),
+        ("OpenCode Zen",     ("OPENCODE_ZEN_API_KEY",),                        "https://opencode.ai/zen/v1/models",  "OPENCODE_ZEN_BASE_URL", True),
+        ("OpenCode Go",      ("OPENCODE_GO_API_KEY",),                         "https://opencode.ai/zen/go/v1/models", "OPENCODE_GO_BASE_URL", True),
    ]
    for _pname, _env_vars, _default_url, _base_env, _supports_health_check in _apikey_providers:
        _key = ""
@@ -709,19 +799,19 @@ def run_doctor(args):
    print(color("◆ Honcho Memory", Colors.CYAN, Colors.BOLD))

    try:
-        from honcho_integration.client import HonchoClientConfig, resolve_config_path
+        from plugins.memory.honcho.client import HonchoClientConfig, resolve_config_path
        hcfg = HonchoClientConfig.from_global_config()
        _honcho_cfg_path = resolve_config_path()

        if not _honcho_cfg_path.exists():
-            check_warn("Honcho config not found", "run: hermes honcho setup")
+            check_warn("Honcho config not found", "run: hermes memory setup")
        elif not hcfg.enabled:
            check_info(f"Honcho disabled (set enabled: true in {_honcho_cfg_path} to activate)")
        elif not (hcfg.api_key or hcfg.base_url):
-            check_fail("Honcho API key or base URL not set", "run: hermes honcho setup")
-            issues.append("No Honcho API key — run 'hermes honcho setup'")
+            check_fail("Honcho API key or base URL not set", "run: hermes memory setup")
+            issues.append("No Honcho API key — run 'hermes memory setup'")
        else:
-            from honcho_integration.client import get_honcho_client, reset_honcho_client
+            from plugins.memory.honcho.client import get_honcho_client, reset_honcho_client
            reset_honcho_client()
            try:
                get_honcho_client(hcfg)
@@ -737,6 +827,36 @@ def run_doctor(args):
    except Exception as _e:
        check_warn("Honcho check failed", str(_e))

+    # =========================================================================
+    # Mem0 memory
+    # =========================================================================
+    print()
+    print(color("◆ Mem0 Memory", Colors.CYAN, Colors.BOLD))
+
+    try:
+        from plugins.memory.mem0 import _load_config as _load_mem0_config
+        mem0_cfg = _load_mem0_config()
+        mem0_key = mem0_cfg.get("api_key", "")
+        if mem0_key:
+            check_ok("Mem0 API key configured")
+            check_info(f"user_id={mem0_cfg.get('user_id', '?')}  agent_id={mem0_cfg.get('agent_id', '?')}")
+            # Check if mem0.json exists but is missing api_key (the bug we fixed)
+            mem0_json = HERMES_HOME / "mem0.json"
+            if mem0_json.exists():
+                try:
+                    import json as _json
+                    file_cfg = _json.loads(mem0_json.read_text())
+                    if not file_cfg.get("api_key") and mem0_key:
+                        check_info("api_key from .env (not in mem0.json) — this is fine")
+                except Exception:
+                    pass
+        else:
+            check_warn("Mem0 not configured", "(set MEM0_API_KEY in .env or run hermes memory setup)")
+    except ImportError:
+        check_warn("Mem0 plugin not loadable", "(optional)")
+    except Exception as _e:
+        check_warn("Mem0 check failed", str(_e))
+
    # =========================================================================
    # Profiles
    # =========================================================================
@@ -89,7 +89,7 @@ def find_gateway_pids() -> list:


 def kill_gateway_processes(force: bool = False) -> int:
-    """Kill any running gateway processes. Returns count killed."""
+    """Kill ALL running gateway processes (across all profiles). Returns count killed."""
    pids = find_gateway_pids()
    killed = 0
    
@@ -109,6 +109,43 @@ def kill_gateway_processes(force: bool = False) -> int:
    return killed


+def stop_profile_gateway() -> bool:
+    """Stop only the gateway for the current profile (HERMES_HOME-scoped).
+
+    Uses the PID file written by start_gateway(), so it only kills the
+    gateway belonging to this profile — not gateways from other profiles.
+    Returns True if a process was stopped, False if none was found.
+    """
+    try:
+        from gateway.status import get_running_pid, remove_pid_file
+    except ImportError:
+        return False
+
+    pid = get_running_pid()
+    if pid is None:
+        return False
+
+    try:
+        os.kill(pid, signal.SIGTERM)
+    except ProcessLookupError:
+        pass  # Already gone
+    except PermissionError:
+        print(f"⚠ Permission denied to kill PID {pid}")
+        return False
+
+    # Wait briefly for it to exit
+    import time as _time
+    for _ in range(20):
+        try:
+            os.kill(pid, 0)
+            _time.sleep(0.5)
+        except (ProcessLookupError, PermissionError):
+            break
+
+    remove_pid_file()
+    return True
+
+
 def is_linux() -> bool:
    return sys.platform.startswith('linux')

@@ -258,8 +295,11 @@ def _system_service_identity(run_as_user: str | None = None) -> tuple[str, str,
    username = (run_as_user or os.getenv("SUDO_USER") or os.getenv("USER") or os.getenv("LOGNAME") or getpass.getuser()).strip()
    if not username:
        raise ValueError("Could not determine which user the gateway service should run as")
+    if username == "root" and not run_as_user:
+        raise ValueError("Refusing to install the gateway system service as root; pass --run-as-user root to override (e.g. in LXC containers)")
    if username == "root":
-        raise ValueError("Refusing to install the gateway system service as root; pass --run-as USER")
+        print_warning("Installing gateway service to run as root.")
+        print_info("  This is fine for LXC/container environments but not recommended on bare-metal hosts.")

    try:
        user_info = pwd.getpwnam(username)
@@ -321,9 +361,9 @@ def install_linux_gateway_from_setup(force: bool = False) -> tuple[str | None, b
            while True:
                run_as_user = prompt("  Run the system gateway service as which user?", default="")
                run_as_user = (run_as_user or "").strip()
-                if run_as_user and run_as_user != "root":
+                if run_as_user:
                    break
-                print_error("  Enter a non-root username.")
+                print_error("  Enter a username.")

        systemd_install(force=force, system=True, run_as_user=run_as_user)
        return scope, True
@@ -463,6 +503,32 @@ def _build_user_local_paths(home: Path, path_entries: list[str]) -> list[str]:
    return [p for p in candidates if p not in path_entries and Path(p).exists()]


+def _hermes_home_for_target_user(target_home_dir: str) -> str:
+    """Remap the current HERMES_HOME to the equivalent under a target user's home.
+
+    When installing a system service via sudo, get_hermes_home() resolves to
+    root's home.  This translates it to the target user's equivalent path:
+      /root/.hermes                    → /home/alice/.hermes
+      /root/.hermes/profiles/coder     → /home/alice/.hermes/profiles/coder
+      /opt/custom-hermes               → /opt/custom-hermes  (kept as-is)
+    """
+    current_hermes = get_hermes_home().resolve()
+    current_default = (Path.home() / ".hermes").resolve()
+    target_default = Path(target_home_dir) / ".hermes"
+
+    # Default ~/.hermes → remap to target user's default
+    if current_hermes == current_default:
+        return str(target_default)
+
+    # Profile or subdir of ~/.hermes → preserve the relative structure
+    try:
+        relative = current_hermes.relative_to(current_default)
+        return str(target_default / relative)
+    except ValueError:
+        # Completely custom path (not under ~/.hermes) — keep as-is
+        return str(current_hermes)
+
+
 def generate_systemd_unit(system: bool = False, run_as_user: str | None = None) -> str:
    python_path = get_python_path()
    working_dir = str(PROJECT_ROOT)
@@ -478,12 +544,11 @@ def generate_systemd_unit(system: bool = False, run_as_user: str | None = None)
        if resolved_node_dir not in path_entries:
            path_entries.append(resolved_node_dir)

-    hermes_home = str(get_hermes_home().resolve())
-
    common_bin_paths = ["/usr/local/sbin", "/usr/local/bin", "/usr/sbin", "/usr/bin", "/sbin", "/bin"]

    if system:
        username, group_name, home_dir = _system_service_identity(run_as_user)
+        hermes_home = _hermes_home_for_target_user(home_dir)
        path_entries.extend(_build_user_local_paths(Path(home_dir), path_entries))
        path_entries.extend(common_bin_paths)
        sane_path = ":".join(path_entries)
@@ -518,6 +583,7 @@ StandardError=journal
 WantedBy=multi-user.target
 """

+    hermes_home = str(get_hermes_home().resolve())
    path_entries.extend(_build_user_local_paths(Path.home(), path_entries))
    path_entries.extend(common_bin_paths)
    sane_path = ":".join(path_entries)
@@ -1066,11 +1132,12 @@ def launchd_status(deep: bool = False):
 # Gateway Runner
 # =============================================================================

-def run_gateway(verbose: bool = False, replace: bool = False):
+def run_gateway(verbose: int = 0, quiet: bool = False, replace: bool = False):
    """Run the gateway in foreground.
    
    Args:
-        verbose: Enable verbose logging output.
+        verbose: Stderr log verbosity count added on top of default WARNING (0=WARNING, 1=INFO, 2+=DEBUG).
+        quiet: Suppress all stderr log output.
        replace: If True, kill any existing gateway instance before starting.
                 This prevents systemd restart loops when the old process
                 hasn't fully exited yet.
@@ -1089,7 +1156,8 @@ def run_gateway(verbose: bool = False, replace: bool = False):
    
    # Exit with code 1 if gateway fails to connect any platform,
    # so systemd Restart=on-failure will retry on transient errors
-    success = asyncio.run(start_gateway(replace=replace))
+    verbosity = None if quiet else verbose
+    success = asyncio.run(start_gateway(replace=replace, verbosity=verbosity))
    if not success:
        sys.exit(1)

@@ -1800,7 +1868,7 @@ def gateway_setup():
                    elif is_macos():
                        launchd_restart()
                    else:
-                        kill_gateway_processes()
+                        stop_profile_gateway()
                        print_info("Start manually: hermes gateway")
                except subprocess.CalledProcessError as e:
                    print_error(f"  Restart failed: {e}")
@@ -1863,9 +1931,10 @@ def gateway_command(args):
    
    # Default to run if no subcommand
    if subcmd is None or subcmd == "run":
-        verbose = getattr(args, 'verbose', False)
+        verbose = getattr(args, 'verbose', 0)
+        quiet = getattr(args, 'quiet', False)
        replace = getattr(args, 'replace', False)
-        run_gateway(verbose, replace=replace)
+        run_gateway(verbose, quiet=quiet, replace=replace)
        return

    if subcmd == "setup":
@@ -1913,31 +1982,54 @@ def gateway_command(args):
            sys.exit(1)
    
    elif subcmd == "stop":
-        # Try service first, then sweep any stray/manual gateway processes.
-        service_available = False
+        stop_all = getattr(args, 'all', False)
        system = getattr(args, 'system', False)
-        
-        if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
-            try:
-                systemd_stop(system=system)
-                service_available = True
-            except subprocess.CalledProcessError:
-                pass  # Fall through to process kill
-        elif is_macos() and get_launchd_plist_path().exists():
-            try:
-                launchd_stop()
-                service_available = True
-            except subprocess.CalledProcessError:
-                pass

-        killed = kill_gateway_processes()
-        if not service_available:
-            if killed:
-                print(f"✓ Stopped {killed} gateway process(es)")
+        if stop_all:
+            # --all: kill every gateway process on the machine
+            service_available = False
+            if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
+                try:
+                    systemd_stop(system=system)
+                    service_available = True
+                except subprocess.CalledProcessError:
+                    pass
+            elif is_macos() and get_launchd_plist_path().exists():
+                try:
+                    launchd_stop()
+                    service_available = True
+                except subprocess.CalledProcessError:
+                    pass
+            killed = kill_gateway_processes()
+            total = killed + (1 if service_available else 0)
+            if total:
+                print(f"✓ Stopped {total} gateway process(es) across all profiles")
            else:
                print("✗ No gateway processes found")
-        elif killed:
-            print(f"✓ Stopped {killed} additional manual gateway process(es)")
+        else:
+            # Default: stop only the current profile's gateway
+            service_available = False
+            if is_linux() and (get_systemd_unit_path(system=False).exists() or get_systemd_unit_path(system=True).exists()):
+                try:
+                    systemd_stop(system=system)
+                    service_available = True
+                except subprocess.CalledProcessError:
+                    pass
+            elif is_macos() and get_launchd_plist_path().exists():
+                try:
+                    launchd_stop()
+                    service_available = True
+                except subprocess.CalledProcessError:
+                    pass
+
+            if not service_available:
+                # No systemd/launchd — use profile-scoped PID file
+                if stop_profile_gateway():
+                    print("✓ Stopped gateway for this profile")
+                else:
+                    print("✗ No gateway running for this profile")
+            else:
+                print(f"✓ Stopped {get_service_name()} service")
    
    elif subcmd == "restart":
        # Try service first, fall back to killing and restarting
@@ -1984,16 +2076,15 @@ def gateway_command(args):
                print("  Fix the service, then retry: hermes gateway start")
                sys.exit(1)

-            # Manual restart: kill existing processes
-            killed = kill_gateway_processes()
-            if killed:
-                print(f"✓ Stopped {killed} gateway process(es)")
+            # Manual restart: stop only this profile's gateway
+            if stop_profile_gateway():
+                print("✓ Stopped gateway for this profile")

            _wait_for_gateway_exit(timeout=10.0, force_after=5.0)

            # Start fresh
            print("Starting gateway...")
-            run_gateway(verbose=False)
+            run_gateway(verbose=0)
    
    elif subcmd == "status":
        deep = getattr(args, 'deep', False)
@@ -0,0 +1,474 @@
+"""hermes memory setup|status — configure memory provider plugins.
+
+Auto-detects installed memory providers via the plugin system.
+Interactive curses-based UI for provider selection, then walks through
+the provider's config schema. Writes config to config.yaml + .env.
+"""
+
+from __future__ import annotations
+
+import getpass
+import os
+import sys
+from pathlib import Path
+
+
+# ---------------------------------------------------------------------------
+# Curses-based interactive picker (same pattern as hermes tools)
+# ---------------------------------------------------------------------------
+
+def _curses_select(title: str, items: list[tuple[str, str]], default: int = 0) -> int:
+    """Interactive single-select with arrow keys.
+
+    items: list of (label, description) tuples.
+    Returns selected index, or default on escape/quit.
+    """
+    try:
+        import curses
+        result = [default]
+
+        def _menu(stdscr):
+            curses.curs_set(0)
+            if curses.has_colors():
+                curses.start_color()
+                curses.use_default_colors()
+                curses.init_pair(1, curses.COLOR_GREEN, -1)
+                curses.init_pair(2, curses.COLOR_YELLOW, -1)
+                curses.init_pair(3, curses.COLOR_CYAN, -1)
+            cursor = default
+
+            while True:
+                stdscr.clear()
+                max_y, max_x = stdscr.getmaxyx()
+
+                # Title
+                try:
+                    stdscr.addnstr(0, 0, title, max_x - 1,
+                                   curses.A_BOLD | (curses.color_pair(2) if curses.has_colors() else 0))
+                    stdscr.addnstr(1, 0, "  ↑↓ navigate  ⏎ select  q quit", max_x - 1,
+                                   curses.color_pair(3) if curses.has_colors() else curses.A_DIM)
+                except curses.error:
+                    pass
+
+                for i, (label, desc) in enumerate(items):
+                    y = i + 3
+                    if y >= max_y - 1:
+                        break
+                    arrow = "→" if i == cursor else " "
+                    line = f" {arrow}  {label}"
+                    if desc:
+                        line += f"  {desc}"
+
+                    attr = curses.A_NORMAL
+                    if i == cursor:
+                        attr = curses.A_BOLD
+                        if curses.has_colors():
+                            attr |= curses.color_pair(1)
+                    try:
+                        stdscr.addnstr(y, 0, line[:max_x - 1], max_x - 1, attr)
+                    except curses.error:
+                        pass
+
+                stdscr.refresh()
+                key = stdscr.getch()
+
+                if key in (curses.KEY_UP, ord('k')):
+                    cursor = (cursor - 1) % len(items)
+                elif key in (curses.KEY_DOWN, ord('j')):
+                    cursor = (cursor + 1) % len(items)
+                elif key in (curses.KEY_ENTER, 10, 13):
+                    result[0] = cursor
+                    return
+                elif key in (27, ord('q')):
+                    return
+
+        curses.wrapper(_menu)
+        return result[0]
+
+    except Exception:
+        # Fallback: numbered input
+        print(f"\n  {title}\n")
+        for i, (label, desc) in enumerate(items):
+            marker = "→" if i == default else " "
+            d = f"  {desc}" if desc else ""
+            print(f"  {marker} {i + 1}. {label}{d}")
+        while True:
+            try:
+                val = input(f"\n  Select [1-{len(items)}] ({default + 1}): ")
+                if not val:
+                    return default
+                idx = int(val) - 1
+                if 0 <= idx < len(items):
+                    return idx
+            except (ValueError, EOFError):
+                return default
+
+
+def _prompt(label: str, default: str | None = None, secret: bool = False) -> str:
+    """Prompt for a value with optional default and secret masking."""
+    suffix = f" [{default}]" if default else ""
+    if secret:
+        sys.stdout.write(f"  {label}{suffix}: ")
+        sys.stdout.flush()
+        if sys.stdin.isatty():
+            val = getpass.getpass(prompt="")
+        else:
+            val = sys.stdin.readline().strip()
+    else:
+        sys.stdout.write(f"  {label}{suffix}: ")
+        sys.stdout.flush()
+        val = sys.stdin.readline().strip()
+    return val or (default or "")
+
+
+# ---------------------------------------------------------------------------
+# Provider discovery
+# ---------------------------------------------------------------------------
+
+def _install_dependencies(provider_name: str) -> None:
+    """Install pip dependencies declared in plugin.yaml."""
+    import subprocess
+    from pathlib import Path as _Path
+
+    plugin_dir = _Path(__file__).parent.parent / "plugins" / "memory" / provider_name
+    yaml_path = plugin_dir / "plugin.yaml"
+    if not yaml_path.exists():
+        return
+
+    try:
+        import yaml
+        with open(yaml_path) as f:
+            meta = yaml.safe_load(f) or {}
+    except Exception:
+        return
+
+    pip_deps = meta.get("pip_dependencies", [])
+    if not pip_deps:
+        return
+
+    # pip name → import name mapping for packages where they differ
+    _IMPORT_NAMES = {
+        "honcho-ai": "honcho",
+        "mem0ai": "mem0",
+        "hindsight-client": "hindsight_client",
+        "hindsight-all": "hindsight",
+    }
+
+    # Check which packages are missing
+    missing = []
+    for dep in pip_deps:
+        import_name = _IMPORT_NAMES.get(dep, dep.replace("-", "_").split("[")[0])
+        try:
+            __import__(import_name)
+        except ImportError:
+            missing.append(dep)
+
+    if not missing:
+        return
+
+    print(f"\n  Installing dependencies: {', '.join(missing)}")
+
+    import shutil
+    uv_path = shutil.which("uv")
+    if not uv_path:
+        print(f"  ⚠ uv not found — cannot install dependencies")
+        print(f"  Install uv: curl -LsSf https://astral.sh/uv/install.sh | sh")
+        print(f"  Then re-run: hermes memory setup")
+        return
+
+    try:
+        subprocess.run(
+            [uv_path, "pip", "install", "--python", sys.executable, "--quiet"] + missing,
+            check=True, timeout=120,
+            capture_output=True,
+        )
+        print(f"  ✓ Installed {', '.join(missing)}")
+    except subprocess.CalledProcessError as e:
+        print(f"  ⚠ Failed to install {', '.join(missing)}")
+        stderr = (e.stderr or b"").decode()[:200]
+        if stderr:
+            print(f"    {stderr}")
+        print(f"  Run manually: uv pip install --python {sys.executable} {' '.join(missing)}")
+    except Exception as e:
+        print(f"  ⚠ Install failed: {e}")
+        print(f"  Run manually: uv pip install --python {sys.executable} {' '.join(missing)}")
+
+    # Also show external dependencies (non-pip) if any
+    ext_deps = meta.get("external_dependencies", [])
+    for dep in ext_deps:
+        dep_name = dep.get("name", "")
+        check_cmd = dep.get("check", "")
+        install_cmd = dep.get("install", "")
+        if check_cmd:
+            try:
+                subprocess.run(
+                    check_cmd, shell=True, capture_output=True, timeout=5
+                )
+            except Exception:
+                if install_cmd:
+                    print(f"\n  ⚠ '{dep_name}' not found. Install with:")
+                    print(f"    {install_cmd}")
+
+
+def _get_available_providers() -> list:
+    """Discover memory providers from plugins/memory/.
+
+    Returns list of (name, description, provider_instance) tuples.
+    """
+    try:
+        from plugins.memory import discover_memory_providers, load_memory_provider
+        raw = discover_memory_providers()
+    except Exception:
+        raw = []
+
+    results = []
+    for name, desc, available in raw:
+        try:
+            provider = load_memory_provider(name)
+            if not provider:
+                continue
+        except Exception:
+            continue
+        # Override description with setup hint
+        schema = provider.get_config_schema() if hasattr(provider, "get_config_schema") else []
+        has_secrets = any(f.get("secret") for f in schema)
+        if has_secrets:
+            setup_hint = "requires API key"
+        elif not schema:
+            setup_hint = "no setup needed"
+        else:
+            setup_hint = "local"
+        results.append((name, setup_hint, provider))
+    return results
+
+
+# ---------------------------------------------------------------------------
+# Setup wizard
+# ---------------------------------------------------------------------------
+
+def cmd_setup(args) -> None:
+    """Interactive memory provider setup wizard."""
+    from hermes_cli.config import load_config, save_config
+
+    providers = _get_available_providers()
+
+    if not providers:
+        print("\n  No memory provider plugins detected.")
+        print("  Install a plugin to ~/.hermes/plugins/ and try again.\n")
+        return
+
+    # Build picker items
+    items = []
+    for name, desc, _ in providers:
+        items.append((name, f"— {desc}"))
+    items.append(("Built-in only", "— MEMORY.md / USER.md (default)"))
+
+    builtin_idx = len(items) - 1
+    selected = _curses_select("Memory provider setup", items, default=builtin_idx)
+
+    config = load_config()
+    if not isinstance(config.get("memory"), dict):
+        config["memory"] = {}
+
+    # Built-in only
+    if selected >= len(providers) or selected < 0:
+        config["memory"]["provider"] = ""
+        save_config(config)
+        print("\n  ✓ Memory provider: built-in only")
+        print("  Saved to config.yaml\n")
+        return
+
+    name, _, provider = providers[selected]
+
+    # Install pip dependencies if declared in plugin.yaml
+    _install_dependencies(name)
+
+    schema = provider.get_config_schema() if hasattr(provider, "get_config_schema") else []
+
+    provider_config = config["memory"].get(name, {})
+    if not isinstance(provider_config, dict):
+        provider_config = {}
+
+    env_path = Path(os.environ.get("HERMES_HOME", os.path.expanduser("~/.hermes"))) / ".env"
+    env_writes = {}
+
+    if schema:
+        print(f"\n  Configuring {name}:\n")
+
+        for field in schema:
+            key = field["key"]
+            desc = field.get("description", key)
+            default = field.get("default")
+            # Dynamic default: look up default from another field's value
+            default_from = field.get("default_from")
+            if default_from and isinstance(default_from, dict):
+                ref_field = default_from.get("field", "")
+                ref_map = default_from.get("map", {})
+                ref_value = provider_config.get(ref_field, "")
+                if ref_value and ref_value in ref_map:
+                    default = ref_map[ref_value]
+            is_secret = field.get("secret", False)
+            choices = field.get("choices")
+            env_var = field.get("env_var")
+            url = field.get("url")
+
+            # Skip fields whose "when" condition doesn't match
+            when = field.get("when")
+            if when and isinstance(when, dict):
+                if not all(provider_config.get(k) == v for k, v in when.items()):
+                    continue
+
+            if choices and not is_secret:
+                # Use curses picker for choice fields
+                choice_items = [(c, "") for c in choices]
+                current = provider_config.get(key, default)
+                current_idx = 0
+                if current and current in choices:
+                    current_idx = choices.index(current)
+                sel = _curses_select(f"  {desc}", choice_items, default=current_idx)
+                provider_config[key] = choices[sel]
+            elif is_secret:
+                # Prompt for secret
+                existing = os.environ.get(env_var, "") if env_var else ""
+                if existing:
+                    masked = f"...{existing[-4:]}" if len(existing) > 4 else "set"
+                    val = _prompt(f"{desc} (current: {masked}, blank to keep)", secret=True)
+                else:
+                    hint = f"  Get yours at {url}" if url else ""
+                    if hint:
+                        print(hint)
+                    val = _prompt(desc, secret=True)
+                if val and env_var:
+                    env_writes[env_var] = val
+            else:
+                # Regular text prompt
+                current = provider_config.get(key)
+                effective_default = current or default
+                val = _prompt(desc, default=str(effective_default) if effective_default else None)
+                if val:
+                    provider_config[key] = val
+
+    # Write activation key to config.yaml
+    config["memory"]["provider"] = name
+    save_config(config)
+
+    # Write non-secret config to provider's native location
+    hermes_home = str(Path(os.environ.get("HERMES_HOME", os.path.expanduser("~/.hermes"))))
+    if provider_config and hasattr(provider, "save_config"):
+        try:
+            provider.save_config(provider_config, hermes_home)
+        except Exception as e:
+            print(f"  ⚠ Failed to write provider config: {e}")
+
+    # Write secrets to .env
+    if env_writes:
+        _write_env_vars(env_path, env_writes)
+
+    print(f"\n  ✓ Memory provider: {name}")
+    print(f"  ✓ Activation saved to config.yaml")
+    if provider_config:
+        print(f"  ✓ Provider config saved")
+    if env_writes:
+        print(f"  ✓ API keys saved to .env")
+    print(f"\n  Start a new session to activate.\n")
+
+
+def _write_env_vars(env_path: Path, env_writes: dict) -> None:
+    """Append or update env vars in .env file."""
+    env_path.parent.mkdir(parents=True, exist_ok=True)
+
+    existing_lines = []
+    if env_path.exists():
+        existing_lines = env_path.read_text().splitlines()
+
+    updated_keys = set()
+    new_lines = []
+    for line in existing_lines:
+        key_match = line.split("=", 1)[0].strip() if "=" in line else ""
+        if key_match in env_writes:
+            new_lines.append(f"{key_match}={env_writes[key_match]}")
+            updated_keys.add(key_match)
+        else:
+            new_lines.append(line)
+
+    for key, val in env_writes.items():
+        if key not in updated_keys:
+            new_lines.append(f"{key}={val}")
+
+    env_path.write_text("\n".join(new_lines) + "\n")
+
+
+# ---------------------------------------------------------------------------
+# Status
+# ---------------------------------------------------------------------------
+
+def cmd_status(args) -> None:
+    """Show current memory provider config."""
+    from hermes_cli.config import load_config
+
+    config = load_config()
+    mem_config = config.get("memory", {})
+    provider_name = mem_config.get("provider", "")
+
+    print(f"\nMemory status\n" + "─" * 40)
+    print(f"  Built-in:  always active")
+    print(f"  Provider:  {provider_name or '(none — built-in only)'}")
+
+    if provider_name:
+        provider_config = mem_config.get(provider_name, {})
+        if provider_config:
+            print(f"\n  {provider_name} config:")
+            for key, val in provider_config.items():
+                print(f"    {key}: {val}")
+
+        providers = _get_available_providers()
+        found = any(name == provider_name for name, _, _ in providers)
+        if found:
+            print(f"\n  Plugin:    installed ✓")
+            for pname, _, p in providers:
+                if pname == provider_name:
+                    if p.is_available():
+                        print(f"  Status:    available ✓")
+                    else:
+                        print(f"  Status:    not available ✗")
+                        schema = p.get_config_schema() if hasattr(p, "get_config_schema") else []
+                        secrets = [f for f in schema if f.get("secret")]
+                        if secrets:
+                            print(f"  Missing:")
+                            for s in secrets:
+                                env_var = s.get("env_var", "")
+                                url = s.get("url", "")
+                                is_set = bool(os.environ.get(env_var))
+                                mark = "✓" if is_set else "✗"
+                                line = f"    {mark} {env_var}"
+                                if url and not is_set:
+                                    line += f"  → {url}"
+                                print(line)
+                    break
+        else:
+            print(f"\n  Plugin:    NOT installed ✗")
+            print(f"  Install the '{provider_name}' memory plugin to ~/.hermes/plugins/")
+
+    providers = _get_available_providers()
+    if providers:
+        print(f"\n  Installed plugins:")
+        for pname, desc, _ in providers:
+            active = " ← active" if pname == provider_name else ""
+            print(f"    • {pname}  ({desc}){active}")
+
+    print()
+
+
+# ---------------------------------------------------------------------------
+# Router
+# ---------------------------------------------------------------------------
+
+def memory_command(args) -> None:
+    """Route memory subcommands."""
+    sub = getattr(args, "memory_command", None)
+    if sub == "setup":
+        cmd_setup(args)
+    elif sub == "status":
+        cmd_status(args)
+    else:
+        cmd_status(args)
@@ -26,6 +26,7 @@ class ModelSwitchResult:
    provider_changed: bool = False
    api_key: str = ""
    base_url: str = ""
+    api_mode: str = ""
    persist: bool = False
    error_message: str = ""
    warning_message: str = ""
@@ -73,6 +74,7 @@ def switch_model(
        detect_provider_for_model,
        validate_requested_model,
        _PROVIDER_LABELS,
+        opencode_model_api_mode,
    )
    from hermes_cli.runtime_provider import resolve_runtime_provider

@@ -98,11 +100,13 @@ def switch_model(
    # Step 4: Resolve credentials for target provider
    api_key = current_api_key
    base_url = current_base_url
+    api_mode = ""
    if provider_changed:
        try:
            runtime = resolve_runtime_provider(requested=target_provider)
            api_key = runtime.get("api_key", "")
            base_url = runtime.get("base_url", "")
+            api_mode = runtime.get("api_mode", "")
        except Exception as e:
            provider_label = _PROVIDER_LABELS.get(target_provider, target_provider)
            if target_provider == "custom":
@@ -130,6 +134,7 @@ def switch_model(
            runtime = resolve_runtime_provider(requested=current_provider)
            api_key = runtime.get("api_key", "")
            base_url = runtime.get("base_url", "")
+            api_mode = runtime.get("api_mode", "")
        except Exception:
            pass

@@ -166,6 +171,12 @@ def switch_model(
        and ("localhost" in (base_url or "") or "127.0.0.1" in (base_url or ""))
    )

+    if target_provider in {"opencode-zen", "opencode-go"}:
+        # Recompute against the requested new model, not the currently-configured
+        # model used during runtime resolution. OpenCode mixes API surfaces by
+        # model family, so a same-provider model switch can change api_mode.
+        api_mode = opencode_model_api_mode(target_provider, new_model)
+
    return ModelSwitchResult(
        success=True,
        new_model=new_model,
@@ -173,6 +184,7 @@ def switch_model(
        provider_changed=provider_changed,
        api_key=api_key,
        base_url=base_url,
+        api_mode=api_mode,
        persist=bool(validation.get("persist")),
        warning_message=validation.get("message") or "",
        is_custom_target=is_custom_target,
@@ -28,7 +28,7 @@ GITHUB_MODELS_CATALOG_URL = COPILOT_MODELS_URL
 OPENROUTER_MODELS: list[tuple[str, str]] = [
    ("anthropic/claude-opus-4.6",       "recommended"),
    ("anthropic/claude-sonnet-4.6",     ""),
-    ("qwen/qwen3.6-plus-preview:free", "free"),
+    ("qwen/qwen3.6-plus:free", "free"),
    ("anthropic/claude-sonnet-4.5",     ""),
    ("anthropic/claude-haiku-4.5",      ""),
    ("openai/gpt-5.4",                  ""),
@@ -51,6 +51,7 @@ OPENROUTER_MODELS: list[tuple[str, str]] = [
    ("nvidia/nemotron-3-super-120b-a12b",      ""),
    ("nvidia/nemotron-3-super-120b-a12b:free", "free"),
    ("arcee-ai/trinity-large-preview:free", "free"),
+    ("arcee-ai/trinity-large-thinking",  ""),
    ("openai/gpt-5.4-pro",              ""),
    ("openai/gpt-5.4-nano",             ""),
 ]
@@ -59,7 +60,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
    "nous": [
        "anthropic/claude-opus-4.6",
        "anthropic/claude-sonnet-4.6",
-        "qwen/qwen3.6-plus-preview:free",
+        "qwen/qwen3.6-plus:free",
        "anthropic/claude-sonnet-4.5",
        "anthropic/claude-haiku-4.5",
        "openai/gpt-5.4",
@@ -82,6 +83,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
        "nvidia/nemotron-3-super-120b-a12b",
        "nvidia/nemotron-3-super-120b-a12b:free",
        "arcee-ai/trinity-large-preview:free",
+        "arcee-ai/trinity-large-thinking",
        "openai/gpt-5.4-pro",
        "openai/gpt-5.4-nano",
    ],
@@ -125,6 +127,12 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
        "kimi-k2-turbo-preview",
        "kimi-k2-0905-preview",
    ],
+    "moonshot": [
+        "kimi-k2.5",
+        "kimi-k2-thinking",
+        "kimi-k2-turbo-preview",
+        "kimi-k2-0905-preview",
+    ],
    "minimax": [
        "MiniMax-M2.7",
        "MiniMax-M2.7-highspeed",
@@ -193,7 +201,10 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
    "opencode-go": [
        "glm-5",
        "kimi-k2.5",
+        "mimo-v2-pro",
+        "mimo-v2-omni",
        "minimax-m2.7",
+        "minimax-m2.5",
    ],
    "ai-gateway": [
        "anthropic/claude-opus-4.6",
@@ -948,6 +959,53 @@ def copilot_model_api_mode(
    return "chat_completions"


+def normalize_opencode_model_id(provider_id: Optional[str], model_id: Optional[str]) -> str:
+    """Normalize OpenCode config IDs to the bare model slug used in API requests."""
+    provider = normalize_provider(provider_id)
+    current = str(model_id or "").strip()
+    if not current or provider not in {"opencode-zen", "opencode-go"}:
+        return current
+
+    prefix = f"{provider}/"
+    if current.lower().startswith(prefix):
+        return current[len(prefix):]
+    return current
+
+
+def opencode_model_api_mode(provider_id: Optional[str], model_id: Optional[str]) -> str:
+    """Determine the API mode for an OpenCode Zen / Go model.
+
+    OpenCode routes different models behind different API surfaces:
+
+    - GPT-5 / Codex models on Zen use ``/v1/responses``
+    - Claude models on Zen use ``/v1/messages``
+    - MiniMax models on Go use ``/v1/messages``
+    - GLM / Kimi on Go use ``/v1/chat/completions``
+    - Other Zen models (Gemini, GLM, Kimi, MiniMax, Qwen, etc.) use
+      ``/v1/chat/completions``
+
+    This follows the published OpenCode docs for Zen and Go endpoints.
+    """
+    provider = normalize_provider(provider_id)
+    normalized = normalize_opencode_model_id(provider_id, model_id).lower()
+    if not normalized:
+        return "chat_completions"
+
+    if provider == "opencode-go":
+        if normalized.startswith("minimax-"):
+            return "anthropic_messages"
+        return "chat_completions"
+
+    if provider == "opencode-zen":
+        if normalized.startswith("claude-"):
+            return "anthropic_messages"
+        if normalized.startswith("gpt-"):
+            return "codex_responses"
+        return "chat_completions"
+
+    return "chat_completions"
+
+
 def github_model_reasoning_efforts(
    model_id: Optional[str],
    *,
@@ -0,0 +1,517 @@
+"""Helpers for Nous subscription managed-tool capabilities."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Dict, Iterable, Optional, Set
+
+from hermes_cli.auth import get_nous_auth_status
+from hermes_cli.config import get_env_value, load_config
+from tools.managed_tool_gateway import is_managed_tool_gateway_ready
+from tools.tool_backend_helpers import (
+    has_direct_modal_credentials,
+    managed_nous_tools_enabled,
+    normalize_browser_cloud_provider,
+    normalize_modal_mode,
+    resolve_modal_backend_state,
+    resolve_openai_audio_api_key,
+)
+
+
+_DEFAULT_PLATFORM_TOOLSETS = {
+    "cli": "hermes-cli",
+}
+
+
+@dataclass(frozen=True)
+class NousFeatureState:
+    key: str
+    label: str
+    included_by_default: bool
+    available: bool
+    active: bool
+    managed_by_nous: bool
+    direct_override: bool
+    toolset_enabled: bool
+    current_provider: str = ""
+    explicit_configured: bool = False
+
+
+@dataclass(frozen=True)
+class NousSubscriptionFeatures:
+    subscribed: bool
+    nous_auth_present: bool
+    provider_is_nous: bool
+    features: Dict[str, NousFeatureState]
+
+    @property
+    def web(self) -> NousFeatureState:
+        return self.features["web"]
+
+    @property
+    def image_gen(self) -> NousFeatureState:
+        return self.features["image_gen"]
+
+    @property
+    def tts(self) -> NousFeatureState:
+        return self.features["tts"]
+
+    @property
+    def browser(self) -> NousFeatureState:
+        return self.features["browser"]
+
+    @property
+    def modal(self) -> NousFeatureState:
+        return self.features["modal"]
+
+    def items(self) -> Iterable[NousFeatureState]:
+        ordered = ("web", "image_gen", "tts", "browser", "modal")
+        for key in ordered:
+            yield self.features[key]
+
+
+def _model_config_dict(config: Dict[str, object]) -> Dict[str, object]:
+    model_cfg = config.get("model")
+    if isinstance(model_cfg, dict):
+        return dict(model_cfg)
+    if isinstance(model_cfg, str) and model_cfg.strip():
+        return {"default": model_cfg.strip()}
+    return {}
+
+
+def _toolset_enabled(config: Dict[str, object], toolset_key: str) -> bool:
+    from toolsets import resolve_toolset
+
+    platform_toolsets = config.get("platform_toolsets")
+    if not isinstance(platform_toolsets, dict) or not platform_toolsets:
+        platform_toolsets = {"cli": [_DEFAULT_PLATFORM_TOOLSETS["cli"]]}
+
+    target_tools = set(resolve_toolset(toolset_key))
+    if not target_tools:
+        return False
+
+    for platform, raw_toolsets in platform_toolsets.items():
+        if isinstance(raw_toolsets, list):
+            toolset_names = list(raw_toolsets)
+        else:
+            default_toolset = _DEFAULT_PLATFORM_TOOLSETS.get(platform)
+            toolset_names = [default_toolset] if default_toolset else []
+        if not toolset_names:
+            default_toolset = _DEFAULT_PLATFORM_TOOLSETS.get(platform)
+            if default_toolset:
+                toolset_names = [default_toolset]
+
+        available_tools: Set[str] = set()
+        for toolset_name in toolset_names:
+            if not isinstance(toolset_name, str) or not toolset_name:
+                continue
+            try:
+                available_tools.update(resolve_toolset(toolset_name))
+            except Exception:
+                continue
+
+        if target_tools and target_tools.issubset(available_tools):
+            return True
+
+    return False
+
+
+def _has_agent_browser() -> bool:
+    import shutil
+
+    agent_browser_bin = shutil.which("agent-browser")
+    local_bin = (
+        Path(__file__).parent.parent / "node_modules" / ".bin" / "agent-browser"
+    )
+    return bool(agent_browser_bin or local_bin.exists())
+
+
+def _browser_label(current_provider: str) -> str:
+    mapping = {
+        "browserbase": "Browserbase",
+        "browser-use": "Browser Use",
+        "camofox": "Camofox",
+        "local": "Local browser",
+    }
+    return mapping.get(current_provider or "local", current_provider or "Local browser")
+
+
+def _tts_label(current_provider: str) -> str:
+    mapping = {
+        "openai": "OpenAI TTS",
+        "elevenlabs": "ElevenLabs",
+        "edge": "Edge TTS",
+        "neutts": "NeuTTS",
+    }
+    return mapping.get(current_provider or "edge", current_provider or "Edge TTS")
+
+
+def _resolve_browser_feature_state(
+    *,
+    browser_tool_enabled: bool,
+    browser_provider: str,
+    browser_provider_explicit: bool,
+    browser_local_available: bool,
+    direct_camofox: bool,
+    direct_browserbase: bool,
+    direct_browser_use: bool,
+    managed_browser_available: bool,
+) -> tuple[str, bool, bool, bool]:
+    """Resolve browser availability using the same precedence as runtime."""
+    if direct_camofox:
+        return "camofox", True, bool(browser_tool_enabled), False
+
+    if browser_provider_explicit:
+        current_provider = browser_provider or "local"
+        if current_provider == "browserbase":
+            provider_available = managed_browser_available or direct_browserbase
+            available = bool(browser_local_available and provider_available)
+            managed = bool(
+                browser_tool_enabled
+                and browser_local_available
+                and managed_browser_available
+                and not direct_browserbase
+            )
+            active = bool(browser_tool_enabled and available)
+            return current_provider, available, active, managed
+        if current_provider == "browser-use":
+            available = bool(browser_local_available and direct_browser_use)
+            active = bool(browser_tool_enabled and available)
+            return current_provider, available, active, False
+        if current_provider == "camofox":
+            return current_provider, False, False, False
+
+        current_provider = "local"
+        available = bool(browser_local_available)
+        active = bool(browser_tool_enabled and available)
+        return current_provider, available, active, False
+
+    if managed_browser_available or direct_browserbase:
+        available = bool(browser_local_available)
+        managed = bool(
+            browser_tool_enabled
+            and browser_local_available
+            and managed_browser_available
+            and not direct_browserbase
+        )
+        active = bool(browser_tool_enabled and available)
+        return "browserbase", available, active, managed
+
+    available = bool(browser_local_available)
+    active = bool(browser_tool_enabled and available)
+    return "local", available, active, False
+
+
+def get_nous_subscription_features(
+    config: Optional[Dict[str, object]] = None,
+) -> NousSubscriptionFeatures:
+    if config is None:
+        config = load_config() or {}
+    config = dict(config)
+    model_cfg = _model_config_dict(config)
+    provider_is_nous = str(model_cfg.get("provider") or "").strip().lower() == "nous"
+
+    try:
+        nous_status = get_nous_auth_status()
+    except Exception:
+        nous_status = {}
+
+    managed_tools_flag = managed_nous_tools_enabled()
+    nous_auth_present = bool(nous_status.get("logged_in"))
+    subscribed = provider_is_nous or nous_auth_present
+
+    web_tool_enabled = _toolset_enabled(config, "web")
+    image_tool_enabled = _toolset_enabled(config, "image_gen")
+    tts_tool_enabled = _toolset_enabled(config, "tts")
+    browser_tool_enabled = _toolset_enabled(config, "browser")
+    modal_tool_enabled = _toolset_enabled(config, "terminal")
+
+    web_cfg = config.get("web") if isinstance(config.get("web"), dict) else {}
+    tts_cfg = config.get("tts") if isinstance(config.get("tts"), dict) else {}
+    browser_cfg = config.get("browser") if isinstance(config.get("browser"), dict) else {}
+    terminal_cfg = config.get("terminal") if isinstance(config.get("terminal"), dict) else {}
+
+    web_backend = str(web_cfg.get("backend") or "").strip().lower()
+    tts_provider = str(tts_cfg.get("provider") or "edge").strip().lower()
+    browser_provider_explicit = "cloud_provider" in browser_cfg
+    browser_provider = normalize_browser_cloud_provider(
+        browser_cfg.get("cloud_provider") if browser_provider_explicit else None
+    )
+    terminal_backend = (
+        str(terminal_cfg.get("backend") or "local").strip().lower()
+    )
+    modal_mode = normalize_modal_mode(
+        terminal_cfg.get("modal_mode")
+    )
+
+    direct_exa = bool(get_env_value("EXA_API_KEY"))
+    direct_firecrawl = bool(get_env_value("FIRECRAWL_API_KEY") or get_env_value("FIRECRAWL_API_URL"))
+    direct_parallel = bool(get_env_value("PARALLEL_API_KEY"))
+    direct_tavily = bool(get_env_value("TAVILY_API_KEY"))
+    direct_fal = bool(get_env_value("FAL_KEY"))
+    direct_openai_tts = bool(resolve_openai_audio_api_key())
+    direct_elevenlabs = bool(get_env_value("ELEVENLABS_API_KEY"))
+    direct_camofox = bool(get_env_value("CAMOFOX_URL"))
+    direct_browserbase = bool(get_env_value("BROWSERBASE_API_KEY") and get_env_value("BROWSERBASE_PROJECT_ID"))
+    direct_browser_use = bool(get_env_value("BROWSER_USE_API_KEY"))
+    direct_modal = has_direct_modal_credentials()
+
+    managed_web_available = managed_tools_flag and nous_auth_present and is_managed_tool_gateway_ready("firecrawl")
+    managed_image_available = managed_tools_flag and nous_auth_present and is_managed_tool_gateway_ready("fal-queue")
+    managed_tts_available = managed_tools_flag and nous_auth_present and is_managed_tool_gateway_ready("openai-audio")
+    managed_browser_available = managed_tools_flag and nous_auth_present and is_managed_tool_gateway_ready("browserbase")
+    managed_modal_available = managed_tools_flag and nous_auth_present and is_managed_tool_gateway_ready("modal")
+    modal_state = resolve_modal_backend_state(
+        modal_mode,
+        has_direct=direct_modal,
+        managed_ready=managed_modal_available,
+    )
+
+    web_managed = web_backend == "firecrawl" and managed_web_available and not direct_firecrawl
+    web_active = bool(
+        web_tool_enabled
+        and (
+            web_managed
+            or (web_backend == "exa" and direct_exa)
+            or (web_backend == "firecrawl" and direct_firecrawl)
+            or (web_backend == "parallel" and direct_parallel)
+            or (web_backend == "tavily" and direct_tavily)
+        )
+    )
+    web_available = bool(
+        managed_web_available or direct_exa or direct_firecrawl or direct_parallel or direct_tavily
+    )
+
+    image_managed = image_tool_enabled and managed_image_available and not direct_fal
+    image_active = bool(image_tool_enabled and (image_managed or direct_fal))
+    image_available = bool(managed_image_available or direct_fal)
+
+    tts_current_provider = tts_provider or "edge"
+    tts_managed = (
+        tts_tool_enabled
+        and tts_current_provider == "openai"
+        and managed_tts_available
+        and not direct_openai_tts
+    )
+    tts_available = bool(
+        tts_current_provider in {"edge", "neutts"}
+        or (tts_current_provider == "openai" and (managed_tts_available or direct_openai_tts))
+        or (tts_current_provider == "elevenlabs" and direct_elevenlabs)
+    )
+    tts_active = bool(tts_tool_enabled and tts_available)
+
+    browser_local_available = _has_agent_browser()
+    (
+        browser_current_provider,
+        browser_available,
+        browser_active,
+        browser_managed,
+    ) = _resolve_browser_feature_state(
+        browser_tool_enabled=browser_tool_enabled,
+        browser_provider=browser_provider,
+        browser_provider_explicit=browser_provider_explicit,
+        browser_local_available=browser_local_available,
+        direct_camofox=direct_camofox,
+        direct_browserbase=direct_browserbase,
+        direct_browser_use=direct_browser_use,
+        managed_browser_available=managed_browser_available,
+    )
+
+    if terminal_backend != "modal":
+        modal_managed = False
+        modal_available = True
+        modal_active = bool(modal_tool_enabled)
+        modal_direct_override = False
+    elif modal_state["selected_backend"] == "managed":
+        modal_managed = bool(modal_tool_enabled)
+        modal_available = True
+        modal_active = bool(modal_tool_enabled)
+        modal_direct_override = False
+    elif modal_state["selected_backend"] == "direct":
+        modal_managed = False
+        modal_available = True
+        modal_active = bool(modal_tool_enabled)
+        modal_direct_override = bool(modal_tool_enabled)
+    elif modal_mode == "managed":
+        modal_managed = False
+        modal_available = bool(managed_modal_available)
+        modal_active = False
+        modal_direct_override = False
+    elif modal_mode == "direct":
+        modal_managed = False
+        modal_available = bool(direct_modal)
+        modal_active = False
+        modal_direct_override = False
+    else:
+        modal_managed = False
+        modal_available = bool(managed_modal_available or direct_modal)
+        modal_active = False
+        modal_direct_override = False
+
+    tts_explicit_configured = False
+    raw_tts_cfg = config.get("tts")
+    if isinstance(raw_tts_cfg, dict) and "provider" in raw_tts_cfg:
+        tts_explicit_configured = tts_provider not in {"", "edge"}
+
+    features = {
+        "web": NousFeatureState(
+            key="web",
+            label="Web tools",
+            included_by_default=True,
+            available=web_available,
+            active=web_active,
+            managed_by_nous=web_managed,
+            direct_override=web_active and not web_managed,
+            toolset_enabled=web_tool_enabled,
+            current_provider=web_backend or "",
+            explicit_configured=bool(web_backend),
+        ),
+        "image_gen": NousFeatureState(
+            key="image_gen",
+            label="Image generation",
+            included_by_default=True,
+            available=image_available,
+            active=image_active,
+            managed_by_nous=image_managed,
+            direct_override=image_active and not image_managed,
+            toolset_enabled=image_tool_enabled,
+            current_provider="FAL" if direct_fal else ("Nous Subscription" if image_managed else ""),
+            explicit_configured=direct_fal,
+        ),
+        "tts": NousFeatureState(
+            key="tts",
+            label="OpenAI TTS",
+            included_by_default=True,
+            available=tts_available,
+            active=tts_active,
+            managed_by_nous=tts_managed,
+            direct_override=tts_active and not tts_managed,
+            toolset_enabled=tts_tool_enabled,
+            current_provider=_tts_label(tts_current_provider),
+            explicit_configured=tts_explicit_configured,
+        ),
+        "browser": NousFeatureState(
+            key="browser",
+            label="Browser automation",
+            included_by_default=True,
+            available=browser_available,
+            active=browser_active,
+            managed_by_nous=browser_managed,
+            direct_override=browser_active and not browser_managed,
+            toolset_enabled=browser_tool_enabled,
+            current_provider=_browser_label(browser_current_provider),
+            explicit_configured=browser_provider_explicit,
+        ),
+        "modal": NousFeatureState(
+            key="modal",
+            label="Modal execution",
+            included_by_default=False,
+            available=modal_available,
+            active=modal_active,
+            managed_by_nous=modal_managed,
+            direct_override=terminal_backend == "modal" and modal_direct_override,
+            toolset_enabled=modal_tool_enabled,
+            current_provider="Modal" if terminal_backend == "modal" else terminal_backend or "local",
+            explicit_configured=terminal_backend == "modal",
+        ),
+    }
+
+    return NousSubscriptionFeatures(
+        subscribed=subscribed,
+        nous_auth_present=nous_auth_present,
+        provider_is_nous=provider_is_nous,
+        features=features,
+    )
+
+
+def get_nous_subscription_explainer_lines() -> list[str]:
+    if not managed_nous_tools_enabled():
+        return []
+
+    return [
+        "Nous subscription enables managed web tools, image generation, OpenAI TTS, and browser automation by default.",
+        "Those managed tools bill to your Nous subscription. Modal execution is optional and can bill to your subscription too.",
+        "Change these later with: hermes setup tools, hermes setup terminal, or hermes status.",
+    ]
+
+
+def apply_nous_provider_defaults(config: Dict[str, object]) -> set[str]:
+    """Apply provider-level Nous defaults shared by `hermes setup` and `hermes model`."""
+    if not managed_nous_tools_enabled():
+        return set()
+
+    features = get_nous_subscription_features(config)
+    if not features.provider_is_nous:
+        return set()
+
+    tts_cfg = config.get("tts")
+    if not isinstance(tts_cfg, dict):
+        tts_cfg = {}
+        config["tts"] = tts_cfg
+
+    current_tts = str(tts_cfg.get("provider") or "edge").strip().lower()
+    if current_tts not in {"", "edge"}:
+        return set()
+
+    tts_cfg["provider"] = "openai"
+    return {"tts"}
+
+
+def apply_nous_managed_defaults(
+    config: Dict[str, object],
+    *,
+    enabled_toolsets: Optional[Iterable[str]] = None,
+) -> set[str]:
+    if not managed_nous_tools_enabled():
+        return set()
+
+    features = get_nous_subscription_features(config)
+    if not features.provider_is_nous:
+        return set()
+
+    selected_toolsets = set(enabled_toolsets or ())
+    changed: set[str] = set()
+
+    web_cfg = config.get("web")
+    if not isinstance(web_cfg, dict):
+        web_cfg = {}
+        config["web"] = web_cfg
+
+    tts_cfg = config.get("tts")
+    if not isinstance(tts_cfg, dict):
+        tts_cfg = {}
+        config["tts"] = tts_cfg
+
+    browser_cfg = config.get("browser")
+    if not isinstance(browser_cfg, dict):
+        browser_cfg = {}
+        config["browser"] = browser_cfg
+
+    if "web" in selected_toolsets and not features.web.explicit_configured and not (
+        get_env_value("PARALLEL_API_KEY")
+        or get_env_value("TAVILY_API_KEY")
+        or get_env_value("FIRECRAWL_API_KEY")
+        or get_env_value("FIRECRAWL_API_URL")
+    ):
+        web_cfg["backend"] = "firecrawl"
+        changed.add("web")
+
+    if "tts" in selected_toolsets and not features.tts.explicit_configured and not (
+        resolve_openai_audio_api_key()
+        or get_env_value("ELEVENLABS_API_KEY")
+    ):
+        tts_cfg["provider"] = "openai"
+        changed.add("tts")
+
+    if "browser" in selected_toolsets and not features.browser.explicit_configured and not (
+        get_env_value("BROWSERBASE_API_KEY")
+        or get_env_value("BROWSER_USE_API_KEY")
+    ):
+        browser_cfg["cloud_provider"] = "browserbase"
+        changed.add("browser")
+
+    if "image_gen" in selected_toolsets and not get_env_value("FAL_KEY"):
+        changed.add("image_gen")
+
+    return changed
@@ -38,6 +38,8 @@ from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Any, Callable, Dict, List, Optional, Set

+from utils import env_var_enabled
+
 try:
    import yaml
 except ImportError:  # pragma: no cover – yaml is optional at import time
@@ -65,7 +67,7 @@ _NS_PARENT = "hermes_plugins"

 def _env_enabled(name: str) -> bool:
    """Return True when an env var is set to a truthy opt-in value."""
-    return os.getenv(name, "").strip().lower() in {"1", "true", "yes", "on"}
+    return env_var_enabled(name)


 def _get_disabled_plugins() -> set:
@@ -51,6 +51,14 @@ _CLONE_CONFIG_FILES = [
    "SOUL.md",
 ]

+# Subdirectory files copied during --clone (path relative to profile root).
+# Memory files are part of the agent's curated identity — just as important
+# as SOUL.md for continuity when cloning a profile.
+_CLONE_SUBDIR_FILES = [
+    "memories/MEMORY.md",
+    "memories/USER.md",
+]
+
 # Runtime files stripped after --clone-all (shouldn't carry over)
 _CLONE_ALL_STRIP = [
    "gateway.pid",
@@ -74,6 +82,8 @@ _DEFAULT_EXPORT_EXCLUDE_ROOT = frozenset({
    "hermes_state.db",
    "response_store.db", "response_store.db-shm", "response_store.db-wal",
    "gateway.pid", "gateway_state.json", "processes.json",
+    "auth.json",            # API keys, OAuth tokens, credential pools
+    ".env",                 # API keys (dotenv)
    "auth.lock", "active_profile", ".update_check",
    "errors.log",
    ".hermes_history",
@@ -426,6 +436,14 @@ def create_profile(
                if src.exists():
                    shutil.copy2(src, profile_dir / filename)

+            # Clone memory and other subdirectory files
+            for relpath in _CLONE_SUBDIR_FILES:
+                src = source_dir / relpath
+                if src.exists():
+                    dst = profile_dir / relpath
+                    dst.parent.mkdir(parents=True, exist_ok=True)
+                    shutil.copy2(src, dst)
+
    return profile_dir


@@ -765,8 +783,17 @@ def export_profile(name: str, output_path: str) -> Path:
            result = shutil.make_archive(base, "gztar", tmpdir, "default")
            return Path(result)

-    result = shutil.make_archive(base, "gztar", str(profile_dir.parent), name)
-    return Path(result)
+    # Named profiles — stage a filtered copy to exclude credentials
+    with tempfile.TemporaryDirectory() as tmpdir:
+        staged = Path(tmpdir) / name
+        _CREDENTIAL_FILES = {"auth.json", ".env"}
+        shutil.copytree(
+            profile_dir,
+            staged,
+            ignore=lambda d, contents: _CREDENTIAL_FILES & set(contents),
+        )
+        result = shutil.make_archive(base, "gztar", tmpdir, name)
+        return Path(result)


 def _normalize_profile_archive_parts(member_name: str) -> List[str]:
@@ -3,6 +3,7 @@
 from __future__ import annotations

 import os
+import re
 from typing import Any, Dict, Optional

 from hermes_cli import auth as auth_mod
@@ -71,7 +72,7 @@ def _get_model_config() -> Dict[str, Any]:
        default = (cfg.get("default") or "").strip()
        base_url = (cfg.get("base_url") or "").strip()
        is_local = "localhost" in base_url or "127.0.0.1" in base_url
-        is_fallback = not default or default == "anthropic/claude-opus-4.6"
+        is_fallback = not default
        if is_local and is_fallback and base_url:
            detected = _auto_detect_local_model(base_url)
            if detected:
@@ -82,9 +83,27 @@ def _get_model_config() -> Dict[str, Any]:
    return {}


+def _provider_supports_explicit_api_mode(provider: Optional[str], configured_provider: Optional[str] = None) -> bool:
+    """Check whether a persisted api_mode should be honored for a given provider.
+
+    Prevents stale api_mode from a previous provider leaking into a
+    different one after a model/provider switch.  Only applies the
+    persisted mode when the config's provider matches the runtime
+    provider (or when no configured provider is recorded).
+    """
+    normalized_provider = (provider or "").strip().lower()
+    normalized_configured = (configured_provider or "").strip().lower()
+    if not normalized_configured:
+        return True
+    if normalized_provider == "custom":
+        return normalized_configured == "custom" or normalized_configured.startswith("custom:")
+    return normalized_configured == normalized_provider
+
+
 def _copilot_runtime_api_mode(model_cfg: Dict[str, Any], api_key: str) -> str:
+    configured_provider = str(model_cfg.get("provider") or "").strip().lower()
    configured_mode = _parse_api_mode(model_cfg.get("api_mode"))
-    if configured_mode:
+    if configured_mode and _provider_supports_explicit_api_mode("copilot", configured_provider):
        return configured_mode

    model_name = str(model_cfg.get("default") or "").strip()
@@ -133,17 +152,30 @@ def _resolve_runtime_from_pool_entry(
        if cfg_provider == "anthropic":
            cfg_base_url = str(model_cfg.get("base_url") or "").strip().rstrip("/")
        base_url = cfg_base_url or base_url or "https://api.anthropic.com"
+    elif provider == "openrouter":
+        base_url = base_url or OPENROUTER_BASE_URL
    elif provider == "nous":
        api_mode = "chat_completions"
    elif provider == "copilot":
        api_mode = _copilot_runtime_api_mode(model_cfg, getattr(entry, "runtime_api_key", ""))
    else:
+        configured_provider = str(model_cfg.get("provider") or "").strip().lower()
        configured_mode = _parse_api_mode(model_cfg.get("api_mode"))
-        if configured_mode:
+        if configured_mode and _provider_supports_explicit_api_mode(provider, configured_provider):
            api_mode = configured_mode
+        elif provider in ("opencode-zen", "opencode-go"):
+            from hermes_cli.models import opencode_model_api_mode
+            api_mode = opencode_model_api_mode(provider, model_cfg.get("default", ""))
        elif base_url.rstrip("/").endswith("/anthropic"):
            api_mode = "anthropic_messages"

+    # OpenCode base URLs end with /v1 for OpenAI-compatible models, but the
+    # Anthropic SDK prepends its own /v1/messages to the base_url.  Strip the
+    # trailing /v1 so the SDK constructs the correct path (e.g.
+    # https://opencode.ai/zen/go/v1/messages instead of .../v1/v1/messages).
+    if api_mode == "anthropic_messages" and provider in ("opencode-zen", "opencode-go"):
+        base_url = re.sub(r"/v1/?$", "", base_url)
+
    return {
        "provider": provider,
        "api_mode": api_mode,
@@ -664,14 +696,21 @@ def resolve_runtime_provider(
        if provider == "copilot":
            api_mode = _copilot_runtime_api_mode(model_cfg, creds.get("api_key", ""))
        else:
-            # Check explicit api_mode from model config first
+            configured_provider = str(model_cfg.get("provider") or "").strip().lower()
+            # Only honor persisted api_mode when it belongs to the same provider family.
            configured_mode = _parse_api_mode(model_cfg.get("api_mode"))
-            if configured_mode:
+            if configured_mode and _provider_supports_explicit_api_mode(provider, configured_provider):
                api_mode = configured_mode
+            elif provider in ("opencode-zen", "opencode-go"):
+                from hermes_cli.models import opencode_model_api_mode
+                api_mode = opencode_model_api_mode(provider, model_cfg.get("default", ""))
            # Auto-detect Anthropic-compatible endpoints by URL convention
            # (e.g. https://api.minimax.io/anthropic, https://dashscope.../anthropic)
            elif base_url.rstrip("/").endswith("/anthropic"):
                api_mode = "anthropic_messages"
+        # Strip trailing /v1 for OpenCode Anthropic models (see comment above).
+        if api_mode == "anthropic_messages" and provider in ("opencode-zen", "opencode-go"):
+            base_url = re.sub(r"/v1/?$", "", base_url)
        return {
            "provider": provider,
            "api_mode": api_mode,
@@ -18,6 +18,12 @@ import sys
 from pathlib import Path
 from typing import Optional, Dict, Any

+from hermes_cli.nous_subscription import (
+    apply_nous_provider_defaults,
+    get_nous_subscription_explainer_lines,
+    get_nous_subscription_features,
+)
+from tools.tool_backend_helpers import managed_nous_tools_enabled
 from hermes_constants import get_optional_skills_dir

 logger = logging.getLogger(__name__)
@@ -108,6 +114,8 @@ _DEFAULT_PROVIDER_MODELS = {
    "minimax-cn": ["MiniMax-M2.7", "MiniMax-M2.7-highspeed", "MiniMax-M2.5", "MiniMax-M2.5-highspeed", "MiniMax-M2.1"],
    "ai-gateway": ["anthropic/claude-opus-4.6", "anthropic/claude-sonnet-4.6", "openai/gpt-5", "google/gemini-3-flash"],
    "kilocode": ["anthropic/claude-opus-4.6", "anthropic/claude-sonnet-4.6", "openai/gpt-5.4", "google/gemini-3-pro-preview", "google/gemini-3-flash-preview"],
+    "opencode-zen": ["gpt-5.4", "gpt-5.3-codex", "claude-sonnet-4-6", "gemini-3-flash", "glm-5", "kimi-k2.5", "minimax-m2.7"],
+    "opencode-go": ["glm-5", "kimi-k2.5", "mimo-v2-pro", "mimo-v2-omni", "minimax-m2.5", "minimax-m2.7"],
    "huggingface": [
        "Qwen/Qwen3.5-397B-A17B", "Qwen/Qwen3-235B-A22B-Thinking-2507",
        "Qwen/Qwen3-Coder-480B-A35B-Instruct", "deepseek-ai/DeepSeek-R1-0528",
@@ -183,6 +191,8 @@ def _setup_provider_model_selection(config, provider_id, current_model, prompt_c
        fetch_api_models,
        fetch_github_model_catalog,
        normalize_copilot_model_id,
+        normalize_opencode_model_id,
+        opencode_model_api_mode,
    )

    pconfig = PROVIDER_REGISTRY[provider_id]
@@ -236,6 +246,11 @@ def _setup_provider_model_selection(config, provider_id, current_model, prompt_c
                f"    Use \"Custom model\" if the model you expect isn't listed."
            )

+    if provider_id in {"opencode-zen", "opencode-go"}:
+        provider_models = [normalize_opencode_model_id(provider_id, mid) for mid in provider_models]
+        current_model = normalize_opencode_model_id(provider_id, current_model)
+        provider_models = list(dict.fromkeys(mid for mid in provider_models if mid))
+
    model_choices = list(provider_models)
    model_choices.append("Custom model")
    model_choices.append(f"Keep current ({current_model})")
@@ -253,6 +268,8 @@ def _setup_provider_model_selection(config, provider_id, current_model, prompt_c
                catalog=catalog,
                api_key=api_key,
            ) or selected_model
+        elif provider_id in {"opencode-zen", "opencode-go"}:
+            selected_model = normalize_opencode_model_id(provider_id, selected_model)
        _set_default_model(config, selected_model)
    elif model_idx == len(provider_models):
        custom = prompt_fn("Enter model name")
@@ -263,6 +280,8 @@ def _setup_provider_model_selection(config, provider_id, current_model, prompt_c
                    catalog=catalog,
                    api_key=api_key,
                ) or custom
+            elif provider_id in {"opencode-zen", "opencode-go"}:
+                selected_model = normalize_opencode_model_id(provider_id, custom)
            else:
                selected_model = custom
            _set_default_model(config, selected_model)
@@ -294,6 +313,10 @@ def _setup_provider_model_selection(config, provider_id, current_model, prompt_c
            catalog=catalog,
            api_key=api_key,
        )
+    elif provider_id in {"opencode-zen", "opencode-go"} and selected_model:
+        model_cfg = _model_config_dict(config)
+        model_cfg["api_mode"] = opencode_model_api_mode(provider_id, selected_model)
+        config["model"] = model_cfg


 def _sync_model_from_disk(config: Dict[str, Any]) -> None:
@@ -594,6 +617,7 @@ def _print_setup_summary(config: dict, hermes_home):
    print_header("Tool Availability Summary")

    tool_status = []
+    subscription_features = get_nous_subscription_features(config)

    # Vision — use the same runtime resolver as the actual vision tools
    try:
@@ -615,43 +639,64 @@ def _print_setup_summary(config: dict, hermes_home):
        tool_status.append(("Mixture of Agents", False, "OPENROUTER_API_KEY"))

    # Web tools (Exa, Parallel, Firecrawl, or Tavily)
-    if get_env_value("EXA_API_KEY") or get_env_value("PARALLEL_API_KEY") or get_env_value("FIRECRAWL_API_KEY") or get_env_value("FIRECRAWL_API_URL") or get_env_value("TAVILY_API_KEY"):
-        tool_status.append(("Web Search & Extract", True, None))
+    if subscription_features.web.managed_by_nous:
+        tool_status.append(("Web Search & Extract (Nous subscription)", True, None))
+    elif subscription_features.web.available:
+        label = "Web Search & Extract"
+        if subscription_features.web.current_provider:
+            label = f"Web Search & Extract ({subscription_features.web.current_provider})"
+        tool_status.append((label, True, None))
    else:
-        tool_status.append(("Web Search & Extract", False, "EXA_API_KEY, PARALLEL_API_KEY, FIRECRAWL_API_KEY, or TAVILY_API_KEY"))
+        tool_status.append(("Web Search & Extract", False, "EXA_API_KEY, PARALLEL_API_KEY, FIRECRAWL_API_KEY/FIRECRAWL_API_URL, or TAVILY_API_KEY"))

-    # Browser tools (local Chromium or Browserbase cloud)
-    import shutil
-
-    _ab_found = (
-        shutil.which("agent-browser")
-        or (
-            Path(__file__).parent.parent / "node_modules" / ".bin" / "agent-browser"
-        ).exists()
-    )
-    if get_env_value("CAMOFOX_URL"):
-        tool_status.append(("Browser Automation (Camofox)", True, None))
-    elif get_env_value("BROWSERBASE_API_KEY"):
-        tool_status.append(("Browser Automation (Browserbase)", True, None))
-    elif _ab_found:
-        tool_status.append(("Browser Automation (local)", True, None))
+    # Browser tools (local Chromium, Camofox, Browserbase, or Browser Use)
+    browser_provider = subscription_features.browser.current_provider
+    if subscription_features.browser.managed_by_nous:
+        tool_status.append(("Browser Automation (Nous Browserbase)", True, None))
+    elif subscription_features.browser.available:
+        label = "Browser Automation"
+        if browser_provider:
+            label = f"Browser Automation ({browser_provider})"
+        tool_status.append((label, True, None))
    else:
+        missing_browser_hint = "npm install -g agent-browser, set CAMOFOX_URL, or configure Browserbase"
+        if browser_provider == "Browserbase":
+            missing_browser_hint = (
+                "npm install -g agent-browser and set "
+                "BROWSERBASE_API_KEY/BROWSERBASE_PROJECT_ID"
+            )
+        elif browser_provider == "Browser Use":
+            missing_browser_hint = (
+                "npm install -g agent-browser and set BROWSER_USE_API_KEY"
+            )
+        elif browser_provider == "Camofox":
+            missing_browser_hint = "CAMOFOX_URL"
+        elif browser_provider == "Local browser":
+            missing_browser_hint = "npm install -g agent-browser"
        tool_status.append(
-            ("Browser Automation", False, "npm install -g agent-browser or set CAMOFOX_URL")
+            ("Browser Automation", False, missing_browser_hint)
        )

    # FAL (image generation)
-    if get_env_value("FAL_KEY"):
+    if subscription_features.image_gen.managed_by_nous:
+        tool_status.append(("Image Generation (Nous subscription)", True, None))
+    elif subscription_features.image_gen.available:
        tool_status.append(("Image Generation", True, None))
    else:
        tool_status.append(("Image Generation", False, "FAL_KEY"))

    # TTS — show configured provider
    tts_provider = config.get("tts", {}).get("provider", "edge")
-    if tts_provider == "elevenlabs" and get_env_value("ELEVENLABS_API_KEY"):
+    if subscription_features.tts.managed_by_nous:
+        tool_status.append(("Text-to-Speech (OpenAI via Nous subscription)", True, None))
+    elif tts_provider == "elevenlabs" and get_env_value("ELEVENLABS_API_KEY"):
        tool_status.append(("Text-to-Speech (ElevenLabs)", True, None))
-    elif tts_provider == "openai" and get_env_value("VOICE_TOOLS_OPENAI_KEY"):
+    elif tts_provider == "openai" and (
+        get_env_value("VOICE_TOOLS_OPENAI_KEY") or get_env_value("OPENAI_API_KEY")
+    ):
        tool_status.append(("Text-to-Speech (OpenAI)", True, None))
+    elif tts_provider == "minimax" and get_env_value("MINIMAX_API_KEY"):
+        tool_status.append(("Text-to-Speech (MiniMax)", True, None))
    elif tts_provider == "neutts":
        try:
            import importlib.util
@@ -665,6 +710,16 @@ def _print_setup_summary(config: dict, hermes_home):
    else:
        tool_status.append(("Text-to-Speech (Edge TTS)", True, None))

+    if subscription_features.modal.managed_by_nous:
+        tool_status.append(("Modal Execution (Nous subscription)", True, None))
+    elif config.get("terminal", {}).get("backend") == "modal":
+        if subscription_features.modal.direct_override:
+            tool_status.append(("Modal Execution (direct Modal)", True, None))
+        else:
+            tool_status.append(("Modal Execution", False, "run 'hermes setup terminal'"))
+    elif managed_nous_tools_enabled() and subscription_features.nous_auth_present:
+        tool_status.append(("Modal Execution (optional via Nous subscription)", True, None))
+
    # Tinker + WandB (RL training)
    if get_env_value("TINKER_API_KEY") and get_env_value("WANDB_API_KEY"):
        tool_status.append(("RL Training (Tinker)", True, None))
@@ -874,6 +929,7 @@ def setup_model_provider(config: dict):
    if isinstance(_m, dict):
        selected_provider = _m.get("provider")

+    nous_subscription_selected = selected_provider == "nous"

    # ── Same-provider fallback & rotation setup ──
    if _supports_same_provider_pool_setup(selected_provider):
@@ -1039,10 +1095,20 @@ def setup_model_provider(config: dict):
            print_info("Skipped — add later with 'hermes setup' or configure AUXILIARY_VISION_* settings")


+    if selected_provider == "nous" and nous_subscription_selected:
+        changed_defaults = apply_nous_provider_defaults(config)
+        current_tts = str(config.get("tts", {}).get("provider") or "edge")
+        if "tts" in changed_defaults:
+            print_success("TTS provider set to: OpenAI TTS via your Nous subscription")
+        else:
+            print_info(f"Keeping your existing TTS provider: {current_tts}")
+
    save_config(config)

-    # Offer TTS provider selection at the end of model setup
-    _setup_tts_provider(config)
+    # Offer TTS provider selection at the end of model setup, except when
+    # Nous subscription defaults are already being applied.
+    if selected_provider != "nous":
+        _setup_tts_provider(config)


 # =============================================================================
@@ -1110,11 +1176,13 @@ def _setup_tts_provider(config: dict):
    """Interactive TTS provider selection with install flow for NeuTTS."""
    tts_config = config.get("tts", {})
    current_provider = tts_config.get("provider", "edge")
+    subscription_features = get_nous_subscription_features(config)

    provider_labels = {
        "edge": "Edge TTS",
        "elevenlabs": "ElevenLabs",
        "openai": "OpenAI TTS",
+        "minimax": "MiniMax TTS",
        "neutts": "NeuTTS",
    }
    current_label = provider_labels.get(current_provider, current_provider)
@@ -1124,20 +1192,37 @@ def _setup_tts_provider(config: dict):
    print_info(f"Current: {current_label}")
    print()

-    choices = [
-        "Edge TTS (free, cloud-based, no setup needed)",
-        "ElevenLabs (premium quality, needs API key)",
-        "OpenAI TTS (good quality, needs API key)",
-        "NeuTTS (local on-device, free, ~300MB model download)",
-        f"Keep current ({current_label})",
-    ]
-    idx = prompt_choice("Select TTS provider:", choices, len(choices) - 1)
+    choices = []
+    providers = []
+    if managed_nous_tools_enabled() and subscription_features.nous_auth_present:
+        choices.append("Nous Subscription (managed OpenAI TTS, billed to your subscription)")
+        providers.append("nous-openai")
+    choices.extend(
+        [
+            "Edge TTS (free, cloud-based, no setup needed)",
+            "ElevenLabs (premium quality, needs API key)",
+            "OpenAI TTS (good quality, needs API key)",
+            "MiniMax TTS (high quality with voice cloning, needs API key)",
+            "NeuTTS (local on-device, free, ~300MB model download)",
+        ]
+    )
+    providers.extend(["edge", "elevenlabs", "openai", "minimax", "neutts"])
+    choices.append(f"Keep current ({current_label})")
+    keep_current_idx = len(choices) - 1
+    idx = prompt_choice("Select TTS provider:", choices, keep_current_idx)

-    if idx == 4:  # Keep current
+    if idx == keep_current_idx:
        return

-    providers = ["edge", "elevenlabs", "openai", "neutts"]
    selected = providers[idx]
+    selected_via_nous = selected == "nous-openai"
+    if selected == "nous-openai":
+        selected = "openai"
+        print_info("OpenAI TTS will use the managed Nous gateway and bill to your subscription.")
+        if get_env_value("VOICE_TOOLS_OPENAI_KEY") or get_env_value("OPENAI_API_KEY"):
+            print_warning(
+                "Direct OpenAI credentials are still configured and may take precedence until removed from ~/.hermes/.env."
+            )

    if selected == "neutts":
        # Check if already installed
@@ -1175,8 +1260,8 @@ def _setup_tts_provider(config: dict):
                print_warning("No API key provided. Falling back to Edge TTS.")
                selected = "edge"

-    elif selected == "openai":
-        existing = get_env_value("VOICE_TOOLS_OPENAI_KEY")
+    elif selected == "openai" and not selected_via_nous:
+        existing = get_env_value("VOICE_TOOLS_OPENAI_KEY") or get_env_value("OPENAI_API_KEY")
        if not existing:
            print()
            api_key = prompt("OpenAI API key for TTS", password=True)
@@ -1187,6 +1272,18 @@ def _setup_tts_provider(config: dict):
                print_warning("No API key provided. Falling back to Edge TTS.")
                selected = "edge"

+    elif selected == "minimax":
+        existing = get_env_value("MINIMAX_API_KEY")
+        if not existing:
+            print()
+            api_key = prompt("MiniMax API key for TTS", password=True)
+            if api_key:
+                save_env_value("MINIMAX_API_KEY", api_key)
+                print_success("MiniMax TTS API key saved")
+            else:
+                print_warning("No API key provided. Falling back to Edge TTS.")
+                selected = "edge"
+
    # Save the selection
    if "tts" not in config:
        config["tts"] = {}
@@ -1331,63 +1428,99 @@ def setup_terminal_backend(config: dict):
    elif selected_backend == "modal":
        print_success("Terminal backend: Modal")
        print_info("Serverless cloud sandboxes. Each session gets its own container.")
-        print_info("Requires a Modal account: https://modal.com")
+        from tools.managed_tool_gateway import is_managed_tool_gateway_ready
+        from tools.tool_backend_helpers import normalize_modal_mode

-        # Check if modal SDK is installed
-        try:
-            __import__("modal")
-        except ImportError:
-            print_info("Installing modal SDK...")
-            import subprocess
-
-            uv_bin = shutil.which("uv")
-            if uv_bin:
-                result = subprocess.run(
-                    [
-                        uv_bin,
-                        "pip",
-                        "install",
-                        "--python",
-                        sys.executable,
-                        "modal",
-                    ],
-                    capture_output=True,
-                    text=True,
-                )
+        managed_modal_available = bool(
+            managed_nous_tools_enabled()
+            and
+            get_nous_subscription_features(config).nous_auth_present
+            and is_managed_tool_gateway_ready("modal")
+        )
+        modal_mode = normalize_modal_mode(config.get("terminal", {}).get("modal_mode"))
+        use_managed_modal = False
+        if managed_modal_available:
+            modal_choices = [
+                "Use my Nous subscription",
+                "Use my own Modal account",
+            ]
+            if modal_mode == "managed":
+                default_modal_idx = 0
+            elif modal_mode == "direct":
+                default_modal_idx = 1
            else:
-                result = subprocess.run(
-                    [sys.executable, "-m", "pip", "install", "modal"],
-                    capture_output=True,
-                    text=True,
-                )
-            if result.returncode == 0:
-                print_success("modal SDK installed")
-            else:
-                print_warning(
-                    "Install failed — run manually: pip install modal"
-                )
+                default_modal_idx = 1 if get_env_value("MODAL_TOKEN_ID") else 0
+            modal_mode_idx = prompt_choice(
+                "Select how Modal execution should be billed:",
+                modal_choices,
+                default_modal_idx,
+            )
+            use_managed_modal = modal_mode_idx == 0

-        # Modal token
-        print()
-        print_info("Modal authentication:")
-        print_info("  Get your token at: https://modal.com/settings")
-        existing_token = get_env_value("MODAL_TOKEN_ID")
-        if existing_token:
-            print_info("  Modal token: already configured")
-            if prompt_yes_no("  Update Modal credentials?", False):
+        if use_managed_modal:
+            config["terminal"]["modal_mode"] = "managed"
+            print_info("Modal execution will use the managed Nous gateway and bill to your subscription.")
+            if get_env_value("MODAL_TOKEN_ID") or get_env_value("MODAL_TOKEN_SECRET"):
+                print_info(
+                    "Direct Modal credentials are still configured, but this backend is pinned to managed mode."
+                )
+        else:
+            config["terminal"]["modal_mode"] = "direct"
+            print_info("Requires a Modal account: https://modal.com")
+
+            # Check if modal SDK is installed
+            try:
+                __import__("modal")
+            except ImportError:
+                print_info("Installing modal SDK...")
+                import subprocess
+
+                uv_bin = shutil.which("uv")
+                if uv_bin:
+                    result = subprocess.run(
+                        [
+                            uv_bin,
+                            "pip",
+                            "install",
+                            "--python",
+                            sys.executable,
+                            "modal",
+                        ],
+                        capture_output=True,
+                        text=True,
+                    )
+                else:
+                    result = subprocess.run(
+                        [sys.executable, "-m", "pip", "install", "modal"],
+                        capture_output=True,
+                        text=True,
+                    )
+                if result.returncode == 0:
+                    print_success("modal SDK installed")
+                else:
+                    print_warning("Install failed — run manually: pip install modal")
+
+            # Modal token
+            print()
+            print_info("Modal authentication:")
+            print_info("  Get your token at: https://modal.com/settings")
+            existing_token = get_env_value("MODAL_TOKEN_ID")
+            if existing_token:
+                print_info("  Modal token: already configured")
+                if prompt_yes_no("  Update Modal credentials?", False):
+                    token_id = prompt("    Modal Token ID", password=True)
+                    token_secret = prompt("    Modal Token Secret", password=True)
+                    if token_id:
+                        save_env_value("MODAL_TOKEN_ID", token_id)
+                    if token_secret:
+                        save_env_value("MODAL_TOKEN_SECRET", token_secret)
+            else:
                token_id = prompt("    Modal Token ID", password=True)
                token_secret = prompt("    Modal Token Secret", password=True)
                if token_id:
                    save_env_value("MODAL_TOKEN_ID", token_id)
                if token_secret:
                    save_env_value("MODAL_TOKEN_SECRET", token_secret)
-        else:
-            token_id = prompt("    Modal Token ID", password=True)
-            token_secret = prompt("    Modal Token Secret", password=True)
-            if token_id:
-                save_env_value("MODAL_TOKEN_ID", token_id)
-            if token_secret:
-                save_env_value("MODAL_TOKEN_SECRET", token_secret)

        _prompt_container_resources(config)

@@ -1501,6 +1634,8 @@ def setup_terminal_backend(config: dict):
    # Sync terminal backend to .env so terminal_tool picks it up directly.
    # config.yaml is the source of truth, but terminal_tool reads TERMINAL_ENV.
    save_env_value("TERMINAL_ENV", selected_backend)
+    if selected_backend == "modal":
+        save_env_value("TERMINAL_MODAL_MODE", config["terminal"].get("modal_mode", "auto"))
    save_config(config)
    print()
    print_success(f"Terminal backend set to: {selected_backend}")
@@ -1706,14 +1841,23 @@ def setup_gateway(config: dict):
            print_info("   1. Message @userinfobot on Telegram")
            print_info("   2. It will reply with your numeric ID (e.g., 123456789)")
            print()
+            existing_allowlist = get_env_value("TELEGRAM_ALLOWED_USERS")
+            if existing_allowlist:
+                print_info(f"   Current allowlist: {existing_allowlist}")
            allowed_users = prompt(
-                "Allowed user IDs (comma-separated, leave empty for open access)"
+                "Allowed user IDs (comma-separated, leave empty to "
+                + ("keep current" if existing_allowlist else "allow open access")
+                + ")"
            )
            if allowed_users:
                save_env_value("TELEGRAM_ALLOWED_USERS", allowed_users.replace(" ", ""))
                print_success(
                    "Telegram allowlist configured - only listed users can use the bot"
                )
+            elif existing_allowlist:
+                print_success(
+                    f"Keeping existing Telegram allowlist: {existing_allowlist}"
+                )
            else:
                print_info(
                    "⚠️  No allowlist set - anyone who finds your bot can use it!"
@@ -1785,8 +1929,13 @@ def setup_gateway(config: dict):
                "   You can also use Discord usernames (resolved on gateway start)."
            )
            print()
+            existing_allowlist = get_env_value("DISCORD_ALLOWED_USERS")
+            if existing_allowlist:
+                print_info(f"   Current allowlist: {existing_allowlist}")
            allowed_users = prompt(
-                "Allowed user IDs or usernames (comma-separated, leave empty for open access)"
+                "Allowed user IDs or usernames (comma-separated, leave empty to "
+                + ("keep current" if existing_allowlist else "allow open access")
+                + ")"
            )
            if allowed_users:
                # Clean up common prefixes (user:123, <@123>, <@!123>)
@@ -1801,6 +1950,10 @@ def setup_gateway(config: dict):
                        cleaned_ids.append(uid)
                save_env_value("DISCORD_ALLOWED_USERS", ",".join(cleaned_ids))
                print_success("Discord allowlist configured")
+            elif existing_allowlist:
+                print_success(
+                    f"Keeping existing Discord allowlist: {existing_allowlist}"
+                )
            else:
                print_info(
                    "⚠️  No allowlist set - anyone in servers with your bot can use it!"
@@ -1897,12 +2050,21 @@ def setup_gateway(config: dict):
                "   To find a Member ID: click a user's name → View full profile → ⋮ → Copy member ID"
            )
            print()
+            existing_allowlist = get_env_value("SLACK_ALLOWED_USERS")
+            if existing_allowlist:
+                print_info(f"   Current allowlist: {existing_allowlist}")
            allowed_users = prompt(
-                "Allowed user IDs (comma-separated, leave empty to deny everyone except paired users)"
+                "Allowed user IDs (comma-separated, leave empty to "
+                + ("keep current" if existing_allowlist else "deny everyone except paired users")
+                + ")"
            )
            if allowed_users:
                save_env_value("SLACK_ALLOWED_USERS", allowed_users.replace(" ", ""))
                print_success("Slack allowlist configured")
+            elif existing_allowlist:
+                print_success(
+                    f"Keeping existing Slack allowlist: {existing_allowlist}"
+                )
            else:
                print_warning(
                    "⚠️  No Slack allowlist set - unpaired users will be denied by default."
@@ -1986,12 +2148,21 @@ def setup_gateway(config: dict):
            print_info("🔒 Security: Restrict who can use your bot")
            print_info("   Matrix user IDs look like @username:server")
            print()
+            existing_allowlist = get_env_value("MATRIX_ALLOWED_USERS")
+            if existing_allowlist:
+                print_info(f"   Current allowlist: {existing_allowlist}")
            allowed_users = prompt(
-                "Allowed user IDs (comma-separated, leave empty for open access)"
+                "Allowed user IDs (comma-separated, leave empty to "
+                + ("keep current" if existing_allowlist else "allow open access")
+                + ")"
            )
            if allowed_users:
                save_env_value("MATRIX_ALLOWED_USERS", allowed_users.replace(" ", ""))
                print_success("Matrix allowlist configured")
+            elif existing_allowlist:
+                print_success(
+                    f"Keeping existing Matrix allowlist: {existing_allowlist}"
+                )
            else:
                print_info(
                    "⚠️  No allowlist set - anyone who can message the bot can use it!"
@@ -2032,12 +2203,21 @@ def setup_gateway(config: dict):
            print_info("   To find your user ID: click your avatar → Profile")
            print_info("   or use the API: GET /api/v4/users/me")
            print()
+            existing_allowlist = get_env_value("MATTERMOST_ALLOWED_USERS")
+            if existing_allowlist:
+                print_info(f"   Current allowlist: {existing_allowlist}")
            allowed_users = prompt(
-                "Allowed user IDs (comma-separated, leave empty for open access)"
+                "Allowed user IDs (comma-separated, leave empty to "
+                + ("keep current" if existing_allowlist else "allow open access")
+                + ")"
            )
            if allowed_users:
                save_env_value("MATTERMOST_ALLOWED_USERS", allowed_users.replace(" ", ""))
                print_success("Mattermost allowlist configured")
+            elif existing_allowlist:
+                print_success(
+                    f"Keeping existing Mattermost allowlist: {existing_allowlist}"
+                )
            else:
                print_info(
                    "⚠️  No allowlist set - anyone who can message the bot can use it!"
@@ -2472,6 +2652,17 @@ SETUP_SECTIONS = [
    ("agent", "Agent Settings", setup_agent_settings),
 ]

+# The returning-user menu intentionally omits standalone TTS because model setup
+# already includes TTS selection and tools setup covers the rest of the provider
+# configuration. Keep this list in the same order as the visible menu entries.
+RETURNING_USER_MENU_SECTION_KEYS = [
+    "model",
+    "terminal",
+    "gateway",
+    "tools",
+    "agent",
+]
+

 def run_setup_wizard(args):
    """Run the interactive setup wizard.
@@ -2622,8 +2813,7 @@ def run_setup_wizard(args):
            # Individual section — map by key, not by position.
            # SETUP_SECTIONS includes TTS but the returning-user menu skips it,
            # so positional indexing (choice - 3) would dispatch the wrong section.
-            _RETURNING_USER_SECTION_KEYS = ["model", "terminal", "gateway", "tools", "agent"]
-            section_key = _RETURNING_USER_SECTION_KEYS[choice - 3]
+            section_key = RETURNING_USER_MENU_SECTION_KEYS[choice - 3]
            section = next((s for s in SETUP_SECTIONS if s[0] == section_key), None)
            if section:
                _, label, func = section
@@ -30,6 +30,7 @@ PLATFORMS = {
    "dingtalk": "💬 DingTalk",
    "feishu": "🪽 Feishu",
    "wecom": "💬 WeCom",
+    "webhook": "🔗 Webhook",
 }

 # ─── Config Helpers ───────────────────────────────────────────────────────────
@@ -15,8 +15,10 @@ from hermes_cli.auth import AuthError, resolve_provider
 from hermes_cli.colors import Colors, color
 from hermes_cli.config import get_env_path, get_env_value, get_hermes_home, load_config
 from hermes_cli.models import provider_label
+from hermes_cli.nous_subscription import get_nous_subscription_features
 from hermes_cli.runtime_provider import resolve_requested_provider
 from hermes_constants import OPENROUTER_MODELS_URL
+from tools.tool_backend_helpers import managed_nous_tools_enabled

 def check_mark(ok: bool) -> str:
    if ok:
@@ -186,6 +188,31 @@ def show_status(args):
    if codex_status.get("error") and not codex_logged_in:
        print(f"    Error:      {codex_status.get('error')}")

+    # =========================================================================
+    # Nous Subscription Features
+    # =========================================================================
+    if managed_nous_tools_enabled():
+        features = get_nous_subscription_features(config)
+        print()
+        print(color("◆ Nous Subscription Features", Colors.CYAN, Colors.BOLD))
+        if not features.nous_auth_present:
+            print("  Nous Portal   ✗ not logged in")
+        else:
+            print("  Nous Portal   ✓ managed tools available")
+        for feature in features.items():
+            if feature.managed_by_nous:
+                state = "active via Nous subscription"
+            elif feature.active:
+                current = feature.current_provider or "configured provider"
+                state = f"active via {current}"
+            elif feature.included_by_default and features.nous_auth_present:
+                state = "included by subscription, not currently selected"
+            elif feature.key == "modal" and features.nous_auth_present:
+                state = "available via subscription (optional)"
+            else:
+                state = "not configured"
+            print(f"  {feature.label:<15} {check_mark(feature.available or feature.active or feature.managed_by_nous)} {state}")
+
    # =========================================================================
    # API-Key Providers
    # =========================================================================
@@ -20,6 +20,11 @@ from hermes_cli.config import (
    load_config, save_config, get_env_value, save_env_value,
 )
 from hermes_cli.colors import Colors, color
+from hermes_cli.nous_subscription import (
+    apply_nous_managed_defaults,
+    get_nous_subscription_features,
+)
+from tools.tool_backend_helpers import managed_nous_tools_enabled

 logger = logging.getLogger(__name__)

@@ -145,6 +150,7 @@ PLATFORMS = {
    "wecom": {"label": "💬 WeCom", "default_toolset": "hermes-wecom"},
    "api_server": {"label": "🌐 API Server", "default_toolset": "hermes-api-server"},
    "mattermost": {"label": "💬 Mattermost", "default_toolset": "hermes-mattermost"},
+    "webhook": {"label": "🔗 Webhook", "default_toolset": "hermes-webhook"},
 }


@@ -158,6 +164,15 @@ TOOL_CATEGORIES = {
        "name": "Text-to-Speech",
        "icon": "🔊",
        "providers": [
+            {
+                "name": "Nous Subscription",
+                "tag": "Managed OpenAI TTS billed to your subscription",
+                "env_vars": [],
+                "tts_provider": "openai",
+                "requires_nous_auth": True,
+                "managed_nous_feature": "tts",
+                "override_env_vars": ["VOICE_TOOLS_OPENAI_KEY", "OPENAI_API_KEY"],
+            },
            {
                "name": "Microsoft Edge TTS",
                "tag": "Free - no API key needed",
@@ -188,6 +203,15 @@ TOOL_CATEGORIES = {
        "setup_note": "A free DuckDuckGo search skill is also included — skip this if you don't need a premium provider.",
        "icon": "🔍",
        "providers": [
+            {
+                "name": "Nous Subscription",
+                "tag": "Managed Firecrawl billed to your subscription",
+                "web_backend": "firecrawl",
+                "env_vars": [],
+                "requires_nous_auth": True,
+                "managed_nous_feature": "web",
+                "override_env_vars": ["FIRECRAWL_API_KEY", "FIRECRAWL_API_URL"],
+            },
            {
                "name": "Firecrawl Cloud",
                "tag": "Hosted service - search, extract, and crawl",
@@ -234,6 +258,14 @@ TOOL_CATEGORIES = {
        "name": "Image Generation",
        "icon": "🎨",
        "providers": [
+            {
+                "name": "Nous Subscription",
+                "tag": "Managed FAL image generation billed to your subscription",
+                "env_vars": [],
+                "requires_nous_auth": True,
+                "managed_nous_feature": "image_gen",
+                "override_env_vars": ["FAL_KEY"],
+            },
            {
                "name": "FAL.ai",
                "tag": "FLUX 2 Pro with auto-upscaling",
@@ -247,11 +279,21 @@ TOOL_CATEGORIES = {
        "name": "Browser Automation",
        "icon": "🌐",
        "providers": [
+            {
+                "name": "Nous Subscription (Browserbase cloud)",
+                "tag": "Managed Browserbase billed to your subscription",
+                "env_vars": [],
+                "browser_provider": "browserbase",
+                "requires_nous_auth": True,
+                "managed_nous_feature": "browser",
+                "override_env_vars": ["BROWSERBASE_API_KEY", "BROWSERBASE_PROJECT_ID"],
+                "post_setup": "browserbase",
+            },
            {
                "name": "Local Browser",
                "tag": "Free headless Chromium (no API key needed)",
                "env_vars": [],
-                "browser_provider": None,
+                "browser_provider": "local",
                "post_setup": "browserbase",  # Same npm install for agent-browser
            },
            {
@@ -519,7 +561,7 @@ def _get_platform_tools(
    # MCP servers are expected to be available on all platforms by default.
    # If the platform explicitly lists one or more MCP server names, treat that
    # as an allowlist. Otherwise include every globally enabled MCP server.
-    mcp_servers = config.get("mcp_servers", {})
+    mcp_servers = config.get("mcp_servers") or {}
    enabled_mcp_servers = {
        name
        for name, server_cfg in mcp_servers.items()
@@ -581,8 +623,11 @@ def _save_platform_tools(config: dict, platform: str, enabled_toolset_keys: Set[
    save_config(config)


-def _toolset_has_keys(ts_key: str) -> bool:
+def _toolset_has_keys(ts_key: str, config: dict = None) -> bool:
    """Check if a toolset's required API keys are configured."""
+    if config is None:
+        config = load_config()
+
    if ts_key == "vision":
        try:
            from agent.auxiliary_client import resolve_vision_provider_client
@@ -592,10 +637,16 @@ def _toolset_has_keys(ts_key: str) -> bool:
        except Exception:
            return False

+    if ts_key in {"web", "image_gen", "tts", "browser"}:
+        features = get_nous_subscription_features(config)
+        feature = features.features.get(ts_key)
+        if feature and (feature.available or feature.managed_by_nous):
+            return True
+
    # Check TOOL_CATEGORIES first (provider-aware)
    cat = TOOL_CATEGORIES.get(ts_key)
    if cat:
-        for provider in cat.get("providers", []):
+        for provider in _visible_providers(cat, config):
            env_vars = provider.get("env_vars", [])
            if not env_vars:
                return True  # No-key provider (e.g. Local Browser, Edge TTS)
@@ -805,11 +856,45 @@ def _configure_toolset(ts_key: str, config: dict):
        _configure_simple_requirements(ts_key)


+def _visible_providers(cat: dict, config: dict) -> list[dict]:
+    """Return provider entries visible for the current auth/config state."""
+    features = get_nous_subscription_features(config)
+    visible = []
+    for provider in cat.get("providers", []):
+        if provider.get("managed_nous_feature") and not managed_nous_tools_enabled():
+            continue
+        if provider.get("requires_nous_auth") and not features.nous_auth_present:
+            continue
+        visible.append(provider)
+    return visible
+
+
+def _toolset_needs_configuration_prompt(ts_key: str, config: dict) -> bool:
+    """Return True when enabling this toolset should open provider setup."""
+    cat = TOOL_CATEGORIES.get(ts_key)
+    if not cat:
+        return not _toolset_has_keys(ts_key, config)
+
+    if ts_key == "tts":
+        tts_cfg = config.get("tts", {})
+        return not isinstance(tts_cfg, dict) or "provider" not in tts_cfg
+    if ts_key == "web":
+        web_cfg = config.get("web", {})
+        return not isinstance(web_cfg, dict) or "backend" not in web_cfg
+    if ts_key == "browser":
+        browser_cfg = config.get("browser", {})
+        return not isinstance(browser_cfg, dict) or "cloud_provider" not in browser_cfg
+    if ts_key == "image_gen":
+        return not get_env_value("FAL_KEY")
+
+    return not _toolset_has_keys(ts_key, config)
+
+
 def _configure_tool_category(ts_key: str, cat: dict, config: dict):
    """Configure a tool category with provider selection."""
    icon = cat.get("icon", "")
    name = cat["name"]
-    providers = cat["providers"]
+    providers = _visible_providers(cat, config)

    # Check Python version requirement
    if cat.get("requires_python"):
@@ -874,6 +959,27 @@ def _configure_tool_category(ts_key: str, cat: dict, config: dict):

 def _is_provider_active(provider: dict, config: dict) -> bool:
    """Check if a provider entry matches the currently active config."""
+    managed_feature = provider.get("managed_nous_feature")
+    if managed_feature:
+        features = get_nous_subscription_features(config)
+        feature = features.features.get(managed_feature)
+        if feature is None:
+            return False
+        if managed_feature == "image_gen":
+            return feature.managed_by_nous
+        if provider.get("tts_provider"):
+            return (
+                feature.managed_by_nous
+                and config.get("tts", {}).get("provider") == provider["tts_provider"]
+            )
+        if "browser_provider" in provider:
+            current = config.get("browser", {}).get("cloud_provider")
+            return feature.managed_by_nous and provider["browser_provider"] == current
+        if provider.get("web_backend"):
+            current = config.get("web", {}).get("backend")
+            return feature.managed_by_nous and current == provider["web_backend"]
+        return feature.managed_by_nous
+
    if provider.get("tts_provider"):
        return config.get("tts", {}).get("provider") == provider["tts_provider"]
    if "browser_provider" in provider:
@@ -900,6 +1006,13 @@ def _detect_active_provider_index(providers: list, config: dict) -> int:
 def _configure_provider(provider: dict, config: dict):
    """Configure a single provider - prompt for API keys and set config."""
    env_vars = provider.get("env_vars", [])
+    managed_feature = provider.get("managed_nous_feature")
+
+    if provider.get("requires_nous_auth"):
+        features = get_nous_subscription_features(config)
+        if not features.nous_auth_present:
+            _print_warning("  Nous Subscription is only available after logging into Nous Portal.")
+            return

    # Set TTS provider in config if applicable
    if provider.get("tts_provider"):
@@ -908,11 +1021,12 @@ def _configure_provider(provider: dict, config: dict):
    # Set browser cloud provider in config if applicable
    if "browser_provider" in provider:
        bp = provider["browser_provider"]
-        if bp:
+        if bp == "local":
+            config.setdefault("browser", {})["cloud_provider"] = "local"
+            _print_success("  Browser set to local mode")
+        elif bp:
            config.setdefault("browser", {})["cloud_provider"] = bp
            _print_success(f"  Browser cloud provider set to: {bp}")
-        else:
-            config.get("browser", {}).pop("cloud_provider", None)

    # Set web search backend in config if applicable
    if provider.get("web_backend"):
@@ -920,7 +1034,16 @@ def _configure_provider(provider: dict, config: dict):
        _print_success(f"  Web backend set to: {provider['web_backend']}")

    if not env_vars:
+        if provider.get("post_setup"):
+            _run_post_setup(provider["post_setup"])
        _print_success(f"  {provider['name']} - no configuration needed!")
+        if managed_feature:
+            _print_info("  Requests for this tool will be billed to your Nous subscription.")
+            override_envs = provider.get("override_env_vars", [])
+            if any(get_env_value(env_var) for env_var in override_envs):
+                _print_warning(
+                    "  Direct credentials are still configured and may take precedence until you remove them from ~/.hermes/.env."
+                )
        return

    # Prompt for each required env var
@@ -1028,7 +1151,7 @@ def _reconfigure_tool(config: dict):
        cat = TOOL_CATEGORIES.get(ts_key)
        reqs = TOOLSET_ENV_REQUIREMENTS.get(ts_key)
        if cat or reqs:
-            if _toolset_has_keys(ts_key):
+            if _toolset_has_keys(ts_key, config):
                configurable.append((ts_key, ts_label))

    if not configurable:
@@ -1058,7 +1181,7 @@ def _configure_tool_category_for_reconfig(ts_key: str, cat: dict, config: dict):
    """Reconfigure a tool category - provider selection + API key update."""
    icon = cat.get("icon", "")
    name = cat["name"]
-    providers = cat["providers"]
+    providers = _visible_providers(cat, config)

    if len(providers) == 1:
        provider = providers[0]
@@ -1093,6 +1216,13 @@ def _configure_tool_category_for_reconfig(ts_key: str, cat: dict, config: dict):
 def _reconfigure_provider(provider: dict, config: dict):
    """Reconfigure a provider - update API keys."""
    env_vars = provider.get("env_vars", [])
+    managed_feature = provider.get("managed_nous_feature")
+
+    if provider.get("requires_nous_auth"):
+        features = get_nous_subscription_features(config)
+        if not features.nous_auth_present:
+            _print_warning("  Nous Subscription is only available after logging into Nous Portal.")
+            return

    if provider.get("tts_provider"):
        config.setdefault("tts", {})["provider"] = provider["tts_provider"]
@@ -1100,12 +1230,12 @@ def _reconfigure_provider(provider: dict, config: dict):

    if "browser_provider" in provider:
        bp = provider["browser_provider"]
-        if bp:
+        if bp == "local":
+            config.setdefault("browser", {})["cloud_provider"] = "local"
+            _print_success("  Browser set to local mode")
+        elif bp:
            config.setdefault("browser", {})["cloud_provider"] = bp
            _print_success(f"  Browser cloud provider set to: {bp}")
-        else:
-            config.get("browser", {}).pop("cloud_provider", None)
-            _print_success("  Browser set to local mode")

    # Set web search backend in config if applicable
    if provider.get("web_backend"):
@@ -1113,7 +1243,16 @@ def _reconfigure_provider(provider: dict, config: dict):
        _print_success(f"  Web backend set to: {provider['web_backend']}")

    if not env_vars:
+        if provider.get("post_setup"):
+            _run_post_setup(provider["post_setup"])
        _print_success(f"  {provider['name']} - no configuration needed!")
+        if managed_feature:
+            _print_info("  Requests for this tool will be billed to your Nous subscription.")
+            override_envs = provider.get("override_env_vars", [])
+            if any(get_env_value(env_var) for env_var in override_envs):
+                _print_warning(
+                    "  Direct credentials are still configured and may take precedence until you remove them from ~/.hermes/.env."
+                )
        return

    for var in env_vars:
@@ -1222,13 +1361,23 @@ def tools_command(args=None, first_install: bool = False, config: dict = None):
                    label = next((l for k, l, _ in _get_effective_configurable_toolsets() if k == ts), ts)
                    print(color(f"  - {label}", Colors.RED))

+            auto_configured = apply_nous_managed_defaults(
+                config,
+                enabled_toolsets=new_enabled,
+            )
+            if managed_nous_tools_enabled():
+                for ts_key in sorted(auto_configured):
+                    label = next((l for k, l, _ in CONFIGURABLE_TOOLSETS if k == ts_key), ts_key)
+                    print(color(f"  ✓ {label}: using your Nous subscription defaults", Colors.GREEN))
+
            # Walk through ALL selected tools that have provider options or
            # need API keys.  This ensures browser (Local vs Browserbase),
            # TTS (Edge vs OpenAI vs ElevenLabs), etc. are shown even when
            # a free provider exists.
            to_configure = [
                ts_key for ts_key in sorted(new_enabled)
-                if TOOL_CATEGORIES.get(ts_key) or TOOLSET_ENV_REQUIREMENTS.get(ts_key)
+                if (TOOL_CATEGORIES.get(ts_key) or TOOLSET_ENV_REQUIREMENTS.get(ts_key))
+                and ts_key not in auto_configured
            ]

            if to_configure:
@@ -1321,7 +1470,7 @@ def tools_command(args=None, first_install: bool = False, config: dict = None):
                    # Configure API keys for newly enabled tools
                    for ts_key in sorted(added):
                        if (TOOL_CATEGORIES.get(ts_key) or TOOLSET_ENV_REQUIREMENTS.get(ts_key)):
-                            if not _toolset_has_keys(ts_key):
+                            if _toolset_needs_configuration_prompt(ts_key, config):
                                _configure_toolset(ts_key, config)
                    _save_platform_tools(config, pk, new_enabled)
                save_config(config)
@@ -1361,7 +1510,7 @@ def tools_command(args=None, first_install: bool = False, config: dict = None):
            # Configure newly enabled toolsets that need API keys
            for ts_key in sorted(added):
                if (TOOL_CATEGORIES.get(ts_key) or TOOLSET_ENV_REQUIREMENTS.get(ts_key)):
-                    if not _toolset_has_keys(ts_key):
+                    if _toolset_needs_configuration_prompt(ts_key, config):
                        _configure_toolset(ts_key, config)

            _save_platform_tools(config, pkey, new_enabled)
@@ -349,13 +349,6 @@ class SessionDB:

        self._conn.commit()

-    def close(self):
-        """Close the database connection."""
-        with self._lock:
-            if self._conn:
-                self._conn.close()
-                self._conn = None
-
    # =========================================================================
    # Session lifecycle
    # =========================================================================
@@ -1009,8 +1002,9 @@ class SessionDB:
        Strategy:
        - Preserve properly paired quoted phrases (``"exact phrase"``)
        - Strip unmatched FTS5-special characters that would cause errors
-        - Wrap unquoted hyphenated terms in quotes so FTS5 matches them
-          as exact phrases instead of splitting on the hyphen
+        - Wrap unquoted hyphenated and dotted terms in quotes so FTS5
+          matches them as exact phrases instead of splitting on the
+          hyphen/dot (e.g. ``chat-send``, ``P2.2``, ``my-app.config.ts``)
        """
        # Step 1: Extract balanced double-quoted phrases and protect them
        # from further processing via numbered placeholders.
@@ -1035,11 +1029,13 @@ class SessionDB:
        sanitized = re.sub(r"(?i)^(AND|OR|NOT)\b\s*", "", sanitized.strip())
        sanitized = re.sub(r"(?i)\s+(AND|OR|NOT)\s*$", "", sanitized.strip())

-        # Step 5: Wrap unquoted hyphenated terms (e.g. ``chat-send``) in
-        # double quotes.  FTS5's tokenizer splits on hyphens, turning
-        # ``chat-send`` into ``chat AND send``.  Quoting preserves the
-        # intended phrase match.
-        sanitized = re.sub(r"\b(\w+(?:-\w+)+)\b", r'"\1"', sanitized)
+        # Step 5: Wrap unquoted dotted and/or hyphenated terms in double
+        # quotes.  FTS5's tokenizer splits on dots and hyphens, turning
+        # ``chat-send`` into ``chat AND send`` and ``P2.2`` into ``p2 AND 2``.
+        # Quoting preserves phrase semantics.  A single pass avoids the
+        # double-quoting bug that would occur if dotted and hyphenated
+        # patterns were applied sequentially (e.g. ``my-app.config``).
+        sanitized = re.sub(r"\b(\w+(?:[.-]\w+)+)\b", r'"\1"', sanitized)

        # Step 6: Restore preserved quoted phrases
        for i, quoted in enumerate(_quoted_parts):
@@ -1,9 +0,0 @@
-"""Honcho integration for AI-native memory.
-
-This package is only active when honcho.enabled=true in config and
-HONCHO_API_KEY is set. All honcho-ai imports are deferred to avoid
-ImportError when the package is not installed.
-
-Named ``honcho_integration`` (not ``honcho``) to avoid shadowing the
-``honcho`` package installed by the ``honcho-ai`` SDK.
-"""
@@ -156,7 +156,7 @@ def _discover_tools():
        "tools.delegate_tool",
        "tools.process_registry",
        "tools.send_message_tool",
-        "tools.honcho_tools",
+        # "tools.honcho_tools",  # Removed — Honcho is now a memory provider plugin
        "tools.homeassistant_tool",
    ]
    import importlib
@@ -371,8 +371,6 @@ def handle_function_call(
    task_id: Optional[str] = None,
    user_task: Optional[str] = None,
    enabled_tools: Optional[List[str]] = None,
-    honcho_manager: Optional[Any] = None,
-    honcho_session_key: Optional[str] = None,
 ) -> str:
    """
    Main function call dispatcher that routes calls to the tool registry.
@@ -417,16 +415,12 @@ def handle_function_call(
                function_name, function_args,
                task_id=task_id,
                enabled_tools=sandbox_enabled,
-                honcho_manager=honcho_manager,
-                honcho_session_key=honcho_session_key,
            )
        else:
            result = registry.dispatch(
                function_name, function_args,
                task_id=task_id,
                user_task=user_task,
-                honcho_manager=honcho_manager,
-                honcho_session_key=honcho_session_key,
            )

        try:
@@ -0,0 +1 @@
+# Hermes plugins package
@@ -0,0 +1,213 @@
+"""Memory provider plugin discovery.
+
+Scans ``plugins/memory/<name>/`` directories for memory provider plugins.
+Each subdirectory must contain ``__init__.py`` with a class implementing
+the MemoryProvider ABC.
+
+Memory providers are separate from the general plugin system — they live
+in the repo and are always available without user installation. Only ONE
+can be active at a time, selected via ``memory.provider`` in config.yaml.
+
+Usage:
+    from plugins.memory import discover_memory_providers, load_memory_provider
+
+    available = discover_memory_providers()   # [(name, desc, available), ...]
+    provider = load_memory_provider("openviking")  # MemoryProvider instance
+"""
+
+from __future__ import annotations
+
+import importlib
+import importlib.util
+import logging
+import sys
+from pathlib import Path
+from typing import List, Optional, Tuple
+
+logger = logging.getLogger(__name__)
+
+_MEMORY_PLUGINS_DIR = Path(__file__).parent
+
+
+def discover_memory_providers() -> List[Tuple[str, str, bool]]:
+    """Scan plugins/memory/ for available providers.
+
+    Returns list of (name, description, is_available) tuples.
+    Does NOT import the providers — just reads plugin.yaml for metadata
+    and does a lightweight availability check.
+    """
+    results = []
+    if not _MEMORY_PLUGINS_DIR.is_dir():
+        return results
+
+    for child in sorted(_MEMORY_PLUGINS_DIR.iterdir()):
+        if not child.is_dir() or child.name.startswith(("_", ".")):
+            continue
+        init_file = child / "__init__.py"
+        if not init_file.exists():
+            continue
+
+        # Read description from plugin.yaml if available
+        desc = ""
+        yaml_file = child / "plugin.yaml"
+        if yaml_file.exists():
+            try:
+                import yaml
+                with open(yaml_file) as f:
+                    meta = yaml.safe_load(f) or {}
+                desc = meta.get("description", "")
+            except Exception:
+                pass
+
+        # Quick availability check — try loading and calling is_available()
+        available = True
+        try:
+            provider = _load_provider_from_dir(child)
+            if provider:
+                available = provider.is_available()
+            else:
+                available = False
+        except Exception:
+            available = False
+
+        results.append((child.name, desc, available))
+
+    return results
+
+
+def load_memory_provider(name: str) -> Optional["MemoryProvider"]:
+    """Load and return a MemoryProvider instance by name.
+
+    Returns None if the provider is not found or fails to load.
+    """
+    provider_dir = _MEMORY_PLUGINS_DIR / name
+    if not provider_dir.is_dir():
+        logger.debug("Memory provider '%s' not found in %s", name, _MEMORY_PLUGINS_DIR)
+        return None
+
+    try:
+        provider = _load_provider_from_dir(provider_dir)
+        if provider:
+            return provider
+        logger.warning("Memory provider '%s' loaded but no provider instance found", name)
+        return None
+    except Exception as e:
+        logger.warning("Failed to load memory provider '%s': %s", name, e)
+        return None
+
+
+def _load_provider_from_dir(provider_dir: Path) -> Optional["MemoryProvider"]:
+    """Import a provider module and extract the MemoryProvider instance.
+
+    The module must have either:
+    - A register(ctx) function (plugin-style) — we simulate a ctx
+    - A top-level class that extends MemoryProvider — we instantiate it
+    """
+    name = provider_dir.name
+    module_name = f"plugins.memory.{name}"
+    init_file = provider_dir / "__init__.py"
+
+    if not init_file.exists():
+        return None
+
+    # Check if already loaded
+    if module_name in sys.modules:
+        mod = sys.modules[module_name]
+    else:
+        # Handle relative imports within the plugin
+        # First ensure the parent packages are registered
+        for parent in ("plugins", "plugins.memory"):
+            if parent not in sys.modules:
+                parent_path = Path(__file__).parent
+                if parent == "plugins":
+                    parent_path = parent_path.parent
+                parent_init = parent_path / "__init__.py"
+                if parent_init.exists():
+                    spec = importlib.util.spec_from_file_location(
+                        parent, str(parent_init),
+                        submodule_search_locations=[str(parent_path)]
+                    )
+                    if spec:
+                        parent_mod = importlib.util.module_from_spec(spec)
+                        sys.modules[parent] = parent_mod
+                        try:
+                            spec.loader.exec_module(parent_mod)
+                        except Exception:
+                            pass
+
+        # Now load the provider module
+        spec = importlib.util.spec_from_file_location(
+            module_name, str(init_file),
+            submodule_search_locations=[str(provider_dir)]
+        )
+        if not spec:
+            return None
+
+        mod = importlib.util.module_from_spec(spec)
+        sys.modules[module_name] = mod
+
+        # Register submodules so relative imports work
+        # e.g., "from .store import MemoryStore" in holographic plugin
+        for sub_file in provider_dir.glob("*.py"):
+            if sub_file.name == "__init__.py":
+                continue
+            sub_name = sub_file.stem
+            full_sub_name = f"{module_name}.{sub_name}"
+            if full_sub_name not in sys.modules:
+                sub_spec = importlib.util.spec_from_file_location(
+                    full_sub_name, str(sub_file)
+                )
+                if sub_spec:
+                    sub_mod = importlib.util.module_from_spec(sub_spec)
+                    sys.modules[full_sub_name] = sub_mod
+                    try:
+                        sub_spec.loader.exec_module(sub_mod)
+                    except Exception as e:
+                        logger.debug("Failed to load submodule %s: %s", full_sub_name, e)
+
+        try:
+            spec.loader.exec_module(mod)
+        except Exception as e:
+            logger.debug("Failed to exec_module %s: %s", module_name, e)
+            sys.modules.pop(module_name, None)
+            return None
+
+    # Try register(ctx) pattern first (how our plugins are written)
+    if hasattr(mod, "register"):
+        collector = _ProviderCollector()
+        try:
+            mod.register(collector)
+            if collector.provider:
+                return collector.provider
+        except Exception as e:
+            logger.debug("register() failed for %s: %s", name, e)
+
+    # Fallback: find a MemoryProvider subclass and instantiate it
+    from agent.memory_provider import MemoryProvider
+    for attr_name in dir(mod):
+        attr = getattr(mod, attr_name, None)
+        if (isinstance(attr, type) and issubclass(attr, MemoryProvider)
+                and attr is not MemoryProvider):
+            try:
+                return attr()
+            except Exception:
+                pass
+
+    return None
+
+
+class _ProviderCollector:
+    """Fake plugin context that captures register_memory_provider calls."""
+
+    def __init__(self):
+        self.provider = None
+
+    def register_memory_provider(self, provider):
+        self.provider = provider
+
+    # No-op for other registration methods
+    def register_tool(self, *args, **kwargs):
+        pass
+
+    def register_hook(self, *args, **kwargs):
+        pass
@@ -0,0 +1,41 @@
+# ByteRover Memory Provider
+
+Persistent memory via the `brv` CLI — hierarchical knowledge tree with tiered retrieval (fuzzy text → LLM-driven search).
+
+## Requirements
+
+Install the ByteRover CLI:
+```bash
+curl -fsSL https://byterover.dev/install.sh | sh
+# or
+npm install -g byterover-cli
+```
+
+## Setup
+
+```bash
+hermes memory setup    # select "byterover"
+```
+
+Or manually:
+```bash
+hermes config set memory.provider byterover
+# Optional cloud sync:
+echo "BRV_API_KEY=your-key" >> ~/.hermes/.env
+```
+
+## Config
+
+| Env Var | Required | Description |
+|---------|----------|-------------|
+| `BRV_API_KEY` | No | Cloud sync key (optional, local-first by default) |
+
+Working directory: `$HERMES_HOME/byterover/` (profile-scoped).
+
+## Tools
+
+| Tool | Description |
+|------|-------------|
+| `brv_query` | Search the knowledge tree |
+| `brv_curate` | Store facts, decisions, patterns |
+| `brv_status` | CLI version, tree stats, sync state |
@@ -0,0 +1,383 @@
+"""ByteRover memory plugin — MemoryProvider interface.
+
+Persistent memory via the ByteRover CLI (``brv``). Organizes knowledge into
+a hierarchical context tree with tiered retrieval (fuzzy text → LLM-driven
+search). Local-first with optional cloud sync.
+
+Original PR #3499 by hieuntg81, adapted to MemoryProvider ABC.
+
+Requires: ``brv`` CLI installed (npm install -g byterover-cli or
+curl -fsSL https://byterover.dev/install.sh | sh).
+
+Config via environment variables (profile-scoped via each profile's .env):
+  BRV_API_KEY   — ByteRover API key (for cloud features, optional for local)
+
+Working directory: $HERMES_HOME/byterover/ (profile-scoped context tree)
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import shutil
+import subprocess
+import threading
+import time
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+# Timeouts
+_QUERY_TIMEOUT = 10   # brv query — should be fast
+_CURATE_TIMEOUT = 120  # brv curate — may involve LLM processing
+
+# Minimum lengths to filter noise
+_MIN_QUERY_LEN = 10
+_MIN_OUTPUT_LEN = 20
+
+
+# ---------------------------------------------------------------------------
+# brv binary resolution (cached, thread-safe)
+# ---------------------------------------------------------------------------
+
+_brv_path_lock = threading.Lock()
+_cached_brv_path: Optional[str] = None
+
+
+def _resolve_brv_path() -> Optional[str]:
+    """Find the brv binary on PATH or well-known install locations."""
+    global _cached_brv_path
+    with _brv_path_lock:
+        if _cached_brv_path is not None:
+            return _cached_brv_path if _cached_brv_path != "" else None
+
+    found = shutil.which("brv")
+    if not found:
+        home = Path.home()
+        candidates = [
+            home / ".brv-cli" / "bin" / "brv",
+            Path("/usr/local/bin/brv"),
+            home / ".npm-global" / "bin" / "brv",
+        ]
+        for c in candidates:
+            if c.exists():
+                found = str(c)
+                break
+
+    with _brv_path_lock:
+        if _cached_brv_path is not None:
+            return _cached_brv_path if _cached_brv_path != "" else None
+        _cached_brv_path = found or ""
+    return found
+
+
+def _run_brv(args: List[str], timeout: int = _QUERY_TIMEOUT,
+             cwd: str = None) -> dict:
+    """Run a brv CLI command. Returns {success, output, error}."""
+    brv_path = _resolve_brv_path()
+    if not brv_path:
+        return {"success": False, "error": "brv CLI not found. Install: npm install -g byterover-cli"}
+
+    cmd = [brv_path] + args
+    effective_cwd = cwd or str(_get_brv_cwd())
+    Path(effective_cwd).mkdir(parents=True, exist_ok=True)
+
+    env = os.environ.copy()
+    brv_bin_dir = str(Path(brv_path).parent)
+    env["PATH"] = brv_bin_dir + os.pathsep + env.get("PATH", "")
+
+    try:
+        result = subprocess.run(
+            cmd, capture_output=True, text=True,
+            timeout=timeout, cwd=effective_cwd, env=env,
+        )
+        stdout = result.stdout.strip()
+        stderr = result.stderr.strip()
+
+        if result.returncode == 0:
+            return {"success": True, "output": stdout}
+        return {"success": False, "error": stderr or stdout or f"brv exited {result.returncode}"}
+
+    except subprocess.TimeoutExpired:
+        return {"success": False, "error": f"brv timed out after {timeout}s"}
+    except FileNotFoundError:
+        global _cached_brv_path
+        with _brv_path_lock:
+            _cached_brv_path = None
+        return {"success": False, "error": "brv CLI not found"}
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+def _get_brv_cwd() -> Path:
+    """Profile-scoped working directory for the brv context tree."""
+    from hermes_constants import get_hermes_home
+    return get_hermes_home() / "byterover"
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas
+# ---------------------------------------------------------------------------
+
+QUERY_SCHEMA = {
+    "name": "brv_query",
+    "description": (
+        "Search ByteRover's persistent knowledge tree for relevant context. "
+        "Returns memories, project knowledge, architectural decisions, and "
+        "patterns from previous sessions. Use for any question where past "
+        "context would help."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "What to search for."},
+        },
+        "required": ["query"],
+    },
+}
+
+CURATE_SCHEMA = {
+    "name": "brv_curate",
+    "description": (
+        "Store important information in ByteRover's persistent knowledge tree. "
+        "Use for architectural decisions, bug fixes, user preferences, project "
+        "patterns — anything worth remembering across sessions. ByteRover's LLM "
+        "automatically categorizes and organizes the memory."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "content": {"type": "string", "description": "The information to remember."},
+        },
+        "required": ["content"],
+    },
+}
+
+STATUS_SCHEMA = {
+    "name": "brv_status",
+    "description": "Check ByteRover status — CLI version, context tree stats, cloud sync state.",
+    "parameters": {"type": "object", "properties": {}, "required": []},
+}
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class ByteRoverMemoryProvider(MemoryProvider):
+    """ByteRover persistent memory via the brv CLI."""
+
+    def __init__(self):
+        self._cwd = ""
+        self._session_id = ""
+        self._turn_count = 0
+        self._sync_thread: Optional[threading.Thread] = None
+
+    @property
+    def name(self) -> str:
+        return "byterover"
+
+    def is_available(self) -> bool:
+        """Check if brv CLI is installed. No network calls."""
+        return _resolve_brv_path() is not None
+
+    def get_config_schema(self):
+        return [
+            {
+                "key": "api_key",
+                "description": "ByteRover API key (optional, for cloud sync)",
+                "secret": True,
+                "env_var": "BRV_API_KEY",
+                "url": "https://app.byterover.dev",
+            },
+        ]
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        self._cwd = str(_get_brv_cwd())
+        self._session_id = session_id
+        self._turn_count = 0
+        Path(self._cwd).mkdir(parents=True, exist_ok=True)
+
+    def system_prompt_block(self) -> str:
+        if not _resolve_brv_path():
+            return ""
+        return (
+            "# ByteRover Memory\n"
+            "Active. Persistent knowledge tree with hierarchical context.\n"
+            "Use brv_query to search past knowledge, brv_curate to store "
+            "important facts, brv_status to check state."
+        )
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        """Run brv query synchronously before the agent's first LLM call.
+
+        Blocks until the query completes (up to _QUERY_TIMEOUT seconds), ensuring
+        the result is available as context before the model is called.
+        """
+        if not query or len(query.strip()) < _MIN_QUERY_LEN:
+            return ""
+        result = _run_brv(
+            ["query", "--", query.strip()[:5000]],
+            timeout=_QUERY_TIMEOUT, cwd=self._cwd,
+        )
+        if result["success"] and result.get("output"):
+            output = result["output"].strip()
+            if len(output) > _MIN_OUTPUT_LEN:
+                return f"## ByteRover Context\n{output}"
+        return ""
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        """No-op: prefetch() now runs synchronously at turn start."""
+        pass
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Curate the conversation turn in background (non-blocking)."""
+        self._turn_count += 1
+
+        # Only curate substantive turns
+        if len(user_content.strip()) < _MIN_QUERY_LEN:
+            return
+
+        def _sync():
+            try:
+                combined = f"User: {user_content[:2000]}\nAssistant: {assistant_content[:2000]}"
+                _run_brv(
+                    ["curate", "--", combined],
+                    timeout=_CURATE_TIMEOUT, cwd=self._cwd,
+                )
+            except Exception as e:
+                logger.debug("ByteRover sync failed: %s", e)
+
+        # Wait for previous sync
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=5.0)
+
+        self._sync_thread = threading.Thread(
+            target=_sync, daemon=True, name="brv-sync"
+        )
+        self._sync_thread.start()
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        """Mirror built-in memory writes to ByteRover."""
+        if action not in ("add", "replace") or not content:
+            return
+
+        def _write():
+            try:
+                label = "User profile" if target == "user" else "Agent memory"
+                _run_brv(
+                    ["curate", "--", f"[{label}] {content}"],
+                    timeout=_CURATE_TIMEOUT, cwd=self._cwd,
+                )
+            except Exception as e:
+                logger.debug("ByteRover memory mirror failed: %s", e)
+
+        t = threading.Thread(target=_write, daemon=True, name="brv-memwrite")
+        t.start()
+
+    def on_pre_compress(self, messages: List[Dict[str, Any]]) -> str:
+        """Extract insights before context compression discards turns."""
+        if not messages:
+            return ""
+
+        # Build a summary of messages about to be compressed
+        parts = []
+        for msg in messages[-10:]:  # last 10 messages
+            role = msg.get("role", "")
+            content = msg.get("content", "")
+            if isinstance(content, str) and content.strip() and role in ("user", "assistant"):
+                parts.append(f"{role}: {content[:500]}")
+
+        if not parts:
+            return ""
+
+        combined = "\n".join(parts)
+
+        def _flush():
+            try:
+                _run_brv(
+                    ["curate", "--", f"[Pre-compression context]\n{combined}"],
+                    timeout=_CURATE_TIMEOUT, cwd=self._cwd,
+                )
+                logger.info("ByteRover pre-compression flush: %d messages", len(parts))
+            except Exception as e:
+                logger.debug("ByteRover pre-compression flush failed: %s", e)
+
+        t = threading.Thread(target=_flush, daemon=True, name="brv-flush")
+        t.start()
+        return ""
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        return [QUERY_SCHEMA, CURATE_SCHEMA, STATUS_SCHEMA]
+
+    def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
+        if tool_name == "brv_query":
+            return self._tool_query(args)
+        elif tool_name == "brv_curate":
+            return self._tool_curate(args)
+        elif tool_name == "brv_status":
+            return self._tool_status()
+        return json.dumps({"error": f"Unknown tool: {tool_name}"})
+
+    def shutdown(self) -> None:
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=10.0)
+
+    # -- Tool implementations ------------------------------------------------
+
+    def _tool_query(self, args: dict) -> str:
+        query = args.get("query", "")
+        if not query:
+            return json.dumps({"error": "query is required"})
+
+        result = _run_brv(
+            ["query", "--", query.strip()[:5000]],
+            timeout=_QUERY_TIMEOUT, cwd=self._cwd,
+        )
+
+        if not result["success"]:
+            return json.dumps({"error": result.get("error", "Query failed")})
+
+        output = result.get("output", "").strip()
+        if not output or len(output) < _MIN_OUTPUT_LEN:
+            return json.dumps({"result": "No relevant memories found."})
+
+        # Truncate very long results
+        if len(output) > 8000:
+            output = output[:8000] + "\n\n[... truncated]"
+
+        return json.dumps({"result": output})
+
+    def _tool_curate(self, args: dict) -> str:
+        content = args.get("content", "")
+        if not content:
+            return json.dumps({"error": "content is required"})
+
+        result = _run_brv(
+            ["curate", "--", content],
+            timeout=_CURATE_TIMEOUT, cwd=self._cwd,
+        )
+
+        if not result["success"]:
+            return json.dumps({"error": result.get("error", "Curate failed")})
+
+        return json.dumps({"result": "Memory curated successfully."})
+
+    def _tool_status(self) -> str:
+        result = _run_brv(["status"], timeout=15, cwd=self._cwd)
+        if not result["success"]:
+            return json.dumps({"error": result.get("error", "Status check failed")})
+        return json.dumps({"status": result.get("output", "")})
+
+
+# ---------------------------------------------------------------------------
+# Plugin entry point
+# ---------------------------------------------------------------------------
+
+def register(ctx) -> None:
+    """Register ByteRover as a memory provider plugin."""
+    ctx.register_memory_provider(ByteRoverMemoryProvider())
@@ -0,0 +1,9 @@
+name: byterover
+version: 1.0.0
+description: "ByteRover — persistent knowledge tree with tiered retrieval via the brv CLI."
+external_dependencies:
+  - name: brv
+    install: "curl -fsSL https://byterover.dev/install.sh | sh"
+    check: "brv --version"
+hooks:
+  - on_pre_compress
@@ -0,0 +1,98 @@
+# Hindsight Memory Provider
+
+Long-term memory with knowledge graph, entity resolution, and multi-strategy retrieval. Supports cloud and local (embedded) modes.
+
+## Requirements
+
+- **Cloud:** API key from [ui.hindsight.vectorize.io](https://ui.hindsight.vectorize.io)
+- **Local:** API key for a supported LLM provider (OpenAI, Anthropic, Gemini, Groq, MiniMax, or Ollama). Embeddings and reranking run locally — no additional API keys needed.
+
+## Setup
+
+```bash
+hermes memory setup    # select "hindsight"
+```
+
+The setup wizard will install dependencies automatically via `uv` and walk you through configuration.
+
+Or manually (cloud mode with defaults):
+```bash
+hermes config set memory.provider hindsight
+echo "HINDSIGHT_API_KEY=your-key" >> ~/.hermes/.env
+```
+
+### Cloud Mode
+
+Connects to the Hindsight Cloud API. Requires an API key from [ui.hindsight.vectorize.io](https://ui.hindsight.vectorize.io).
+
+### Local Mode
+
+Runs an embedded Hindsight server with built-in PostgreSQL. Requires an LLM API key (e.g. Groq, OpenAI, Anthropic) for memory extraction and synthesis. The daemon starts automatically in the background on first use and stops after 5 minutes of inactivity.
+
+Daemon startup logs: `~/.hermes/logs/hindsight-embed.log`
+Daemon runtime logs: `~/.hindsight/profiles/<profile>.log`
+
+## Config
+
+Config file: `~/.hermes/hindsight/config.json`
+
+### Connection
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `mode` | `cloud` | `cloud` or `local` |
+| `api_url` | `https://api.hindsight.vectorize.io` | API URL (cloud mode) |
+| `api_url` | `http://localhost:8888` | API URL (local mode, unused — daemon manages its own port) |
+
+### Memory
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `bank_id` | `hermes` | Memory bank name |
+| `budget` | `mid` | Recall thoroughness: `low` / `mid` / `high` |
+
+### Integration
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `memory_mode` | `hybrid` | How memories are integrated into the agent |
+| `prefetch_method` | `recall` | Method for automatic context injection |
+
+**memory_mode:**
+- `hybrid` — automatic context injection + tools available to the LLM
+- `context` — automatic injection only, no tools exposed
+- `tools` — tools only, no automatic injection
+
+**prefetch_method:**
+- `recall` — injects raw memory facts (fast)
+- `reflect` — injects LLM-synthesized summary (slower, more coherent)
+
+### Local Mode LLM
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `llm_provider` | `openai` | LLM provider: `openai`, `anthropic`, `gemini`, `groq`, `minimax`, `ollama` |
+| `llm_model` | per-provider | Model name (e.g. `gpt-4o-mini`, `openai/gpt-oss-120b`) |
+
+The LLM API key is stored in `~/.hermes/.env` as `HINDSIGHT_LLM_API_KEY`.
+
+## Tools
+
+Available in `hybrid` and `tools` memory modes:
+
+| Tool | Description |
+|------|-------------|
+| `hindsight_retain` | Store information with auto entity extraction |
+| `hindsight_recall` | Multi-strategy search (semantic + entity graph) |
+| `hindsight_reflect` | Cross-memory synthesis (LLM-powered) |
+
+## Environment Variables
+
+| Variable | Description |
+|----------|-------------|
+| `HINDSIGHT_API_KEY` | API key for Hindsight Cloud |
+| `HINDSIGHT_LLM_API_KEY` | LLM API key for local mode |
+| `HINDSIGHT_API_URL` | Override API endpoint |
+| `HINDSIGHT_BANK_ID` | Override bank name |
+| `HINDSIGHT_BUDGET` | Override recall budget |
+| `HINDSIGHT_MODE` | Override mode (`cloud` / `local`) |
@@ -0,0 +1,515 @@
+"""Hindsight memory plugin — MemoryProvider interface.
+
+Long-term memory with knowledge graph, entity resolution, and multi-strategy
+retrieval. Supports cloud (API key) and local modes.
+
+Original PR #1811 by benfrank241, adapted to MemoryProvider ABC.
+
+Config via environment variables:
+  HINDSIGHT_API_KEY   — API key for Hindsight Cloud
+  HINDSIGHT_BANK_ID   — memory bank identifier (default: hermes)
+  HINDSIGHT_BUDGET    — recall budget: low/mid/high (default: mid)
+  HINDSIGHT_API_URL   — API endpoint
+  HINDSIGHT_MODE      — cloud or local (default: cloud)
+
+Or via $HERMES_HOME/hindsight/config.json (profile-scoped), falling back to
+~/.hindsight/config.json (legacy, shared) for backward compatibility.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import os
+import threading
+from typing import Any, Dict, List
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+_DEFAULT_API_URL = "https://api.hindsight.vectorize.io"
+_DEFAULT_LOCAL_URL = "http://localhost:8888"
+_VALID_BUDGETS = {"low", "mid", "high"}
+_PROVIDER_DEFAULT_MODELS = {
+    "openai": "gpt-4o-mini",
+    "anthropic": "claude-haiku-4-5",
+    "gemini": "gemini-2.5-flash",
+    "groq": "openai/gpt-oss-120b",
+    "minimax": "MiniMax-M2.7",
+    "ollama": "gemma3:12b",
+    "lmstudio": "local-model",
+}
+
+
+# ---------------------------------------------------------------------------
+# Dedicated event loop for Hindsight async calls (one per process, reused).
+# Avoids creating ephemeral loops that leak aiohttp sessions.
+# ---------------------------------------------------------------------------
+
+_loop: asyncio.AbstractEventLoop | None = None
+_loop_thread: threading.Thread | None = None
+_loop_lock = threading.Lock()
+
+
+def _get_loop() -> asyncio.AbstractEventLoop:
+    """Return a long-lived event loop running on a background thread."""
+    global _loop, _loop_thread
+    with _loop_lock:
+        if _loop is not None and _loop.is_running():
+            return _loop
+        _loop = asyncio.new_event_loop()
+
+        def _run():
+            asyncio.set_event_loop(_loop)
+            _loop.run_forever()
+
+        _loop_thread = threading.Thread(target=_run, daemon=True, name="hindsight-loop")
+        _loop_thread.start()
+        return _loop
+
+
+def _run_sync(coro, timeout: float = 120.0):
+    """Schedule *coro* on the shared loop and block until done."""
+    loop = _get_loop()
+    future = asyncio.run_coroutine_threadsafe(coro, loop)
+    return future.result(timeout=timeout)
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas
+# ---------------------------------------------------------------------------
+
+RETAIN_SCHEMA = {
+    "name": "hindsight_retain",
+    "description": (
+        "Store information to long-term memory. Hindsight automatically "
+        "extracts structured facts, resolves entities, and indexes for retrieval."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "content": {"type": "string", "description": "The information to store."},
+            "context": {"type": "string", "description": "Short label (e.g. 'user preference', 'project decision')."},
+        },
+        "required": ["content"],
+    },
+}
+
+RECALL_SCHEMA = {
+    "name": "hindsight_recall",
+    "description": (
+        "Search long-term memory. Returns memories ranked by relevance using "
+        "semantic search, keyword matching, entity graph traversal, and reranking."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "What to search for."},
+        },
+        "required": ["query"],
+    },
+}
+
+REFLECT_SCHEMA = {
+    "name": "hindsight_reflect",
+    "description": (
+        "Synthesize a reasoned answer from long-term memories. Unlike recall, "
+        "this reasons across all stored memories to produce a coherent response."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "The question to reflect on."},
+        },
+        "required": ["query"],
+    },
+}
+
+
+# ---------------------------------------------------------------------------
+# Config
+# ---------------------------------------------------------------------------
+
+def _load_config() -> dict:
+    """Load config from profile-scoped path, legacy path, or env vars.
+
+    Resolution order:
+      1. $HERMES_HOME/hindsight/config.json  (profile-scoped)
+      2. ~/.hindsight/config.json             (legacy, shared)
+      3. Environment variables
+    """
+    from pathlib import Path
+    from hermes_constants import get_hermes_home
+
+    # Profile-scoped path (preferred)
+    profile_path = get_hermes_home() / "hindsight" / "config.json"
+    if profile_path.exists():
+        try:
+            return json.loads(profile_path.read_text(encoding="utf-8"))
+        except Exception:
+            pass
+
+    # Legacy shared path (backward compat)
+    legacy_path = Path.home() / ".hindsight" / "config.json"
+    if legacy_path.exists():
+        try:
+            return json.loads(legacy_path.read_text(encoding="utf-8"))
+        except Exception:
+            pass
+
+    return {
+        "mode": os.environ.get("HINDSIGHT_MODE", "cloud"),
+        "apiKey": os.environ.get("HINDSIGHT_API_KEY", ""),
+        "banks": {
+            "hermes": {
+                "bankId": os.environ.get("HINDSIGHT_BANK_ID", "hermes"),
+                "budget": os.environ.get("HINDSIGHT_BUDGET", "mid"),
+                "enabled": True,
+            }
+        },
+    }
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class HindsightMemoryProvider(MemoryProvider):
+    """Hindsight long-term memory with knowledge graph and multi-strategy retrieval."""
+
+    def __init__(self):
+        self._config = None
+        self._api_key = None
+        self._api_url = _DEFAULT_API_URL
+        self._bank_id = "hermes"
+        self._budget = "mid"
+        self._mode = "cloud"
+        self._memory_mode = "hybrid"  # "context", "tools", or "hybrid"
+        self._prefetch_method = "recall"  # "recall" or "reflect"
+        self._client = None
+        self._prefetch_result = ""
+        self._prefetch_lock = threading.Lock()
+        self._prefetch_thread = None
+        self._sync_thread = None
+
+    @property
+    def name(self) -> str:
+        return "hindsight"
+
+    def is_available(self) -> bool:
+        try:
+            cfg = _load_config()
+            mode = cfg.get("mode", "cloud")
+            if mode == "local":
+                return True
+            has_key = bool(cfg.get("apiKey") or os.environ.get("HINDSIGHT_API_KEY", ""))
+            has_url = bool(cfg.get("api_url") or os.environ.get("HINDSIGHT_API_URL", ""))
+            return has_key or has_url
+        except Exception:
+            return False
+
+    def save_config(self, values, hermes_home):
+        """Write config to $HERMES_HOME/hindsight/config.json."""
+        import json
+        from pathlib import Path
+        config_dir = Path(hermes_home) / "hindsight"
+        config_dir.mkdir(parents=True, exist_ok=True)
+        config_path = config_dir / "config.json"
+        existing = {}
+        if config_path.exists():
+            try:
+                existing = json.loads(config_path.read_text())
+            except Exception:
+                pass
+        existing.update(values)
+        config_path.write_text(json.dumps(existing, indent=2))
+
+    def get_config_schema(self):
+        return [
+            {"key": "mode", "description": "Cloud API or local embedded mode", "default": "cloud", "choices": ["cloud", "local"]},
+            {"key": "api_url", "description": "Hindsight API URL", "default": _DEFAULT_API_URL, "when": {"mode": "cloud"}},
+            {"key": "api_key", "description": "Hindsight Cloud API key", "secret": True, "env_var": "HINDSIGHT_API_KEY", "url": "https://ui.hindsight.vectorize.io", "when": {"mode": "cloud"}},
+            {"key": "llm_provider", "description": "LLM provider for local mode", "default": "openai", "choices": ["openai", "anthropic", "gemini", "groq", "minimax", "ollama"], "when": {"mode": "local"}},
+            {"key": "llm_api_key", "description": "LLM API key for local Hindsight", "secret": True, "env_var": "HINDSIGHT_LLM_API_KEY", "when": {"mode": "local"}},
+            {"key": "llm_model", "description": "LLM model for local mode", "default": "gpt-4o-mini", "default_from": {"field": "llm_provider", "map": _PROVIDER_DEFAULT_MODELS}, "when": {"mode": "local"}},
+            {"key": "bank_id", "description": "Memory bank name", "default": "hermes"},
+            {"key": "budget", "description": "Recall thoroughness", "default": "mid", "choices": ["low", "mid", "high"]},
+            {"key": "memory_mode", "description": "Memory integration mode", "default": "hybrid", "choices": ["hybrid", "context", "tools"]},
+            {"key": "prefetch_method", "description": "Auto-recall method", "default": "recall", "choices": ["recall", "reflect"]},
+        ]
+
+    def _get_client(self):
+        """Return the cached Hindsight client (created once, reused)."""
+        if self._client is None:
+            if self._mode == "local":
+                from hindsight import HindsightEmbedded
+                # Disable __del__ on the class to prevent "attached to a
+                # different loop" errors during GC — we handle cleanup in
+                # shutdown() instead.
+                HindsightEmbedded.__del__ = lambda self: None
+                self._client = HindsightEmbedded(
+                    profile=self._config.get("profile", "hermes"),
+                    llm_provider=self._config.get("llm_provider", ""),
+                    llm_api_key=self._config.get("llmApiKey") or os.environ.get("HINDSIGHT_LLM_API_KEY", ""),
+                    llm_model=self._config.get("llm_model", ""),
+                )
+            else:
+                from hindsight_client import Hindsight
+                kwargs = {"base_url": self._api_url, "timeout": 30.0}
+                if self._api_key:
+                    kwargs["api_key"] = self._api_key
+                self._client = Hindsight(**kwargs)
+        return self._client
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        self._config = _load_config()
+        self._mode = self._config.get("mode", "cloud")
+        self._api_key = self._config.get("apiKey") or os.environ.get("HINDSIGHT_API_KEY", "")
+        default_url = _DEFAULT_LOCAL_URL if self._mode == "local" else _DEFAULT_API_URL
+        self._api_url = self._config.get("api_url") or os.environ.get("HINDSIGHT_API_URL", default_url)
+
+        banks = self._config.get("banks", {}).get("hermes", {})
+        self._bank_id = self._config.get("bank_id") or banks.get("bankId", "hermes")
+        budget = self._config.get("budget") or banks.get("budget", "mid")
+        self._budget = budget if budget in _VALID_BUDGETS else "mid"
+
+        memory_mode = self._config.get("memory_mode", "hybrid")
+        self._memory_mode = memory_mode if memory_mode in ("context", "tools", "hybrid") else "hybrid"
+
+        prefetch_method = self._config.get("prefetch_method", "recall")
+        self._prefetch_method = prefetch_method if prefetch_method in ("recall", "reflect") else "recall"
+
+        logger.info("Hindsight initialized: mode=%s, api_url=%s, bank=%s, budget=%s, memory_mode=%s, prefetch_method=%s",
+                     self._mode, self._api_url, self._bank_id, self._budget, self._memory_mode, self._prefetch_method)
+
+        # For local mode, start the embedded daemon in the background so it
+        # doesn't block the chat. Redirect stdout/stderr to a log file to
+        # prevent rich startup output from spamming the terminal.
+        if self._mode == "local":
+            def _start_daemon():
+                import traceback
+                from pathlib import Path
+                log_dir = Path(os.environ.get("HERMES_HOME", os.path.expanduser("~/.hermes"))) / "logs"
+                log_dir.mkdir(parents=True, exist_ok=True)
+                log_path = log_dir / "hindsight-embed.log"
+                try:
+                    # Redirect the daemon manager's Rich console to our log file
+                    # instead of stderr. This avoids global fd redirects that
+                    # would capture output from other threads.
+                    import hindsight_embed.daemon_embed_manager as dem
+                    from rich.console import Console
+                    dem.console = Console(file=open(log_path, "a"), force_terminal=False)
+
+                    client = self._get_client()
+                    profile = self._config.get("profile", "hermes")
+
+                    # Update the profile .env to match our current config so
+                    # the daemon always starts with the right settings.
+                    # If the config changed and the daemon is running, stop it.
+                    from pathlib import Path as _Path
+                    profile_env = _Path.home() / ".hindsight" / "profiles" / f"{profile}.env"
+                    current_key = self._config.get("llmApiKey") or os.environ.get("HINDSIGHT_LLM_API_KEY", "")
+                    current_provider = self._config.get("llm_provider", "")
+                    current_model = self._config.get("llm_model", "")
+
+                    # Read saved profile config
+                    saved = {}
+                    if profile_env.exists():
+                        for line in profile_env.read_text().splitlines():
+                            if "=" in line and not line.startswith("#"):
+                                k, v = line.split("=", 1)
+                                saved[k.strip()] = v.strip()
+
+                    config_changed = (
+                        saved.get("HINDSIGHT_API_LLM_PROVIDER") != current_provider or
+                        saved.get("HINDSIGHT_API_LLM_MODEL") != current_model or
+                        saved.get("HINDSIGHT_API_LLM_API_KEY") != current_key
+                    )
+
+                    if config_changed:
+                        # Write updated profile .env
+                        profile_env.parent.mkdir(parents=True, exist_ok=True)
+                        profile_env.write_text(
+                            f"HINDSIGHT_API_LLM_PROVIDER={current_provider}\n"
+                            f"HINDSIGHT_API_LLM_API_KEY={current_key}\n"
+                            f"HINDSIGHT_API_LLM_MODEL={current_model}\n"
+                            f"HINDSIGHT_API_LOG_LEVEL=info\n"
+                        )
+                        if client._manager.is_running(profile):
+                            with open(log_path, "a") as f:
+                                f.write("\n=== Config changed, restarting daemon ===\n")
+                            client._manager.stop(profile)
+
+                    client._ensure_started()
+                    with open(log_path, "a") as f:
+                        f.write("\n=== Daemon started successfully ===\n")
+                except Exception as e:
+                    with open(log_path, "a") as f:
+                        f.write(f"\n=== Daemon startup failed: {e} ===\n")
+                        traceback.print_exc(file=f)
+
+            t = threading.Thread(target=_start_daemon, daemon=True, name="hindsight-daemon-start")
+            t.start()
+
+    def system_prompt_block(self) -> str:
+        if self._memory_mode == "context":
+            return (
+                f"# Hindsight Memory\n"
+                f"Active (context mode). Bank: {self._bank_id}, budget: {self._budget}.\n"
+                f"Relevant memories are automatically injected into context."
+            )
+        if self._memory_mode == "tools":
+            return (
+                f"# Hindsight Memory\n"
+                f"Active (tools mode). Bank: {self._bank_id}, budget: {self._budget}.\n"
+                f"Use hindsight_recall to search, hindsight_reflect for synthesis, "
+                f"hindsight_retain to store facts."
+            )
+        return (
+            f"# Hindsight Memory\n"
+            f"Active. Bank: {self._bank_id}, budget: {self._budget}.\n"
+            f"Relevant memories are automatically injected into context. "
+            f"Use hindsight_recall to search, hindsight_reflect for synthesis, "
+            f"hindsight_retain to store facts."
+        )
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        if self._prefetch_thread and self._prefetch_thread.is_alive():
+            self._prefetch_thread.join(timeout=3.0)
+        with self._prefetch_lock:
+            result = self._prefetch_result
+            self._prefetch_result = ""
+        if not result:
+            return ""
+        return f"## Hindsight Memory\n{result}"
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        if self._memory_mode == "tools":
+            return
+        def _run():
+            try:
+                client = self._get_client()
+                if self._prefetch_method == "reflect":
+                    resp = _run_sync(client.areflect(bank_id=self._bank_id, query=query, budget=self._budget))
+                    text = resp.text or ""
+                else:
+                    resp = _run_sync(client.arecall(bank_id=self._bank_id, query=query, budget=self._budget))
+                    text = "\n".join(r.text for r in resp.results if r.text) if resp.results else ""
+                if text:
+                    with self._prefetch_lock:
+                        self._prefetch_result = text
+            except Exception as e:
+                logger.debug("Hindsight prefetch failed: %s", e)
+
+        self._prefetch_thread = threading.Thread(target=_run, daemon=True, name="hindsight-prefetch")
+        self._prefetch_thread.start()
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Retain conversation turn in background (non-blocking)."""
+        combined = f"User: {user_content}\nAssistant: {assistant_content}"
+
+        def _sync():
+            try:
+                client = self._get_client()
+                _run_sync(client.aretain(
+                    bank_id=self._bank_id, content=combined, context="conversation"
+                ))
+            except Exception as e:
+                logger.warning("Hindsight sync failed: %s", e)
+
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=5.0)
+        self._sync_thread = threading.Thread(target=_sync, daemon=True, name="hindsight-sync")
+        self._sync_thread.start()
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        if self._memory_mode == "context":
+            return []
+        return [RETAIN_SCHEMA, RECALL_SCHEMA, REFLECT_SCHEMA]
+
+    def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
+        try:
+            client = self._get_client()
+        except Exception as e:
+            logger.warning("Hindsight client init failed: %s", e)
+            return json.dumps({"error": f"Hindsight client unavailable: {e}"})
+
+        if tool_name == "hindsight_retain":
+            content = args.get("content", "")
+            if not content:
+                return json.dumps({"error": "Missing required parameter: content"})
+            context = args.get("context")
+            try:
+                _run_sync(client.aretain(
+                    bank_id=self._bank_id, content=content, context=context
+                ))
+                return json.dumps({"result": "Memory stored successfully."})
+            except Exception as e:
+                logger.warning("hindsight_retain failed: %s", e)
+                return json.dumps({"error": f"Failed to store memory: {e}"})
+
+        elif tool_name == "hindsight_recall":
+            query = args.get("query", "")
+            if not query:
+                return json.dumps({"error": "Missing required parameter: query"})
+            try:
+                resp = _run_sync(client.arecall(
+                    bank_id=self._bank_id, query=query, budget=self._budget
+                ))
+                if not resp.results:
+                    return json.dumps({"result": "No relevant memories found."})
+                lines = [f"{i}. {r.text}" for i, r in enumerate(resp.results, 1)]
+                return json.dumps({"result": "\n".join(lines)})
+            except Exception as e:
+                logger.warning("hindsight_recall failed: %s", e)
+                return json.dumps({"error": f"Failed to search memory: {e}"})
+
+        elif tool_name == "hindsight_reflect":
+            query = args.get("query", "")
+            if not query:
+                return json.dumps({"error": "Missing required parameter: query"})
+            try:
+                resp = _run_sync(client.areflect(
+                    bank_id=self._bank_id, query=query, budget=self._budget
+                ))
+                return json.dumps({"result": resp.text or "No relevant memories found."})
+            except Exception as e:
+                logger.warning("hindsight_reflect failed: %s", e)
+                return json.dumps({"error": f"Failed to reflect: {e}"})
+
+        return json.dumps({"error": f"Unknown tool: {tool_name}"})
+
+    def shutdown(self) -> None:
+        global _loop, _loop_thread
+        for t in (self._prefetch_thread, self._sync_thread):
+            if t and t.is_alive():
+                t.join(timeout=5.0)
+        if self._client is not None:
+            try:
+                if self._mode == "local":
+                    # Use the public close() API. The RuntimeError from
+                    # aiohttp's "attached to a different loop" is expected
+                    # and harmless — the daemon keeps running independently.
+                    try:
+                        self._client.close()
+                    except RuntimeError:
+                        pass
+                else:
+                    _run_sync(self._client.aclose())
+            except Exception:
+                pass
+            self._client = None
+        # Stop the background event loop so no tasks are pending at exit
+        if _loop is not None and _loop.is_running():
+            _loop.call_soon_threadsafe(_loop.stop)
+            if _loop_thread is not None:
+                _loop_thread.join(timeout=5.0)
+            _loop = None
+            _loop_thread = None
+
+
+def register(ctx) -> None:
+    """Register Hindsight as a memory provider plugin."""
+    ctx.register_memory_provider(HindsightMemoryProvider())
@@ -0,0 +1,10 @@
+name: hindsight
+version: 1.0.0
+description: "Hindsight — long-term memory with knowledge graph, entity resolution, and multi-strategy retrieval."
+pip_dependencies:
+  - hindsight-client
+  - hindsight-all
+requires_env:
+  - HINDSIGHT_API_KEY
+hooks:
+  - on_session_end
@@ -0,0 +1,36 @@
+# Holographic Memory Provider
+
+Local SQLite fact store with FTS5 search, trust scoring, entity resolution, and HRR-based compositional retrieval.
+
+## Requirements
+
+None — uses SQLite (always available). NumPy optional for HRR algebra.
+
+## Setup
+
+```bash
+hermes memory setup    # select "holographic"
+```
+
+Or manually:
+```bash
+hermes config set memory.provider holographic
+```
+
+## Config
+
+Config in `config.yaml` under `plugins.hermes-memory-store`:
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `db_path` | `$HERMES_HOME/memory_store.db` | SQLite database path |
+| `auto_extract` | `false` | Auto-extract facts at session end |
+| `default_trust` | `0.5` | Default trust score for new facts |
+| `hrr_dim` | `1024` | HRR vector dimensions |
+
+## Tools
+
+| Tool | Description |
+|------|-------------|
+| `fact_store` | 9 actions: add, search, probe, related, reason, contradict, update, remove, list |
+| `fact_feedback` | Rate facts as helpful/unhelpful (trains trust scores) |
@@ -0,0 +1,407 @@
+"""hermes-memory-store — holographic memory plugin using MemoryProvider interface.
+
+Registers as a MemoryProvider plugin, giving the agent structured fact storage
+with entity resolution, trust scoring, and HRR-based compositional retrieval.
+
+Original plugin by dusterbloom (PR #2351), adapted to the MemoryProvider ABC.
+
+Config in $HERMES_HOME/config.yaml (profile-scoped):
+  plugins:
+    hermes-memory-store:
+      db_path: $HERMES_HOME/memory_store.db   # omit to use the default
+      auto_extract: false
+      default_trust: 0.5
+      min_trust_threshold: 0.3
+      temporal_decay_half_life: 0
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import re
+from pathlib import Path
+from typing import Any, Dict, List
+
+from agent.memory_provider import MemoryProvider
+from .store import MemoryStore
+from .retrieval import FactRetriever
+
+logger = logging.getLogger(__name__)
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas (unchanged from original PR)
+# ---------------------------------------------------------------------------
+
+FACT_STORE_SCHEMA = {
+    "name": "fact_store",
+    "description": (
+        "Deep structured memory with algebraic reasoning. "
+        "Use alongside the memory tool — memory for always-on context, "
+        "fact_store for deep recall and compositional queries.\n\n"
+        "ACTIONS (simple → powerful):\n"
+        "• add — Store a fact the user would expect you to remember.\n"
+        "• search — Keyword lookup ('editor config', 'deploy process').\n"
+        "• probe — Entity recall: ALL facts about a person/thing.\n"
+        "• related — What connects to an entity? Structural adjacency.\n"
+        "• reason — Compositional: facts connected to MULTIPLE entities simultaneously.\n"
+        "• contradict — Memory hygiene: find facts making conflicting claims.\n"
+        "• update/remove/list — CRUD operations.\n\n"
+        "IMPORTANT: Before answering questions about the user, ALWAYS probe or reason first."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "action": {
+                "type": "string",
+                "enum": ["add", "search", "probe", "related", "reason", "contradict", "update", "remove", "list"],
+            },
+            "content": {"type": "string", "description": "Fact content (required for 'add')."},
+            "query": {"type": "string", "description": "Search query (required for 'search')."},
+            "entity": {"type": "string", "description": "Entity name for 'probe'/'related'."},
+            "entities": {"type": "array", "items": {"type": "string"}, "description": "Entity names for 'reason'."},
+            "fact_id": {"type": "integer", "description": "Fact ID for 'update'/'remove'."},
+            "category": {"type": "string", "enum": ["user_pref", "project", "tool", "general"]},
+            "tags": {"type": "string", "description": "Comma-separated tags."},
+            "trust_delta": {"type": "number", "description": "Trust adjustment for 'update'."},
+            "min_trust": {"type": "number", "description": "Minimum trust filter (default: 0.3)."},
+            "limit": {"type": "integer", "description": "Max results (default: 10)."},
+        },
+        "required": ["action"],
+    },
+}
+
+FACT_FEEDBACK_SCHEMA = {
+    "name": "fact_feedback",
+    "description": (
+        "Rate a fact after using it. Mark 'helpful' if accurate, 'unhelpful' if outdated. "
+        "This trains the memory — good facts rise, bad facts sink."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "action": {"type": "string", "enum": ["helpful", "unhelpful"]},
+            "fact_id": {"type": "integer", "description": "The fact ID to rate."},
+        },
+        "required": ["action", "fact_id"],
+    },
+}
+
+
+# ---------------------------------------------------------------------------
+# Config
+# ---------------------------------------------------------------------------
+
+def _load_plugin_config() -> dict:
+    from hermes_constants import get_hermes_home
+    config_path = get_hermes_home() / "config.yaml"
+    if not config_path.exists():
+        return {}
+    try:
+        import yaml
+        with open(config_path) as f:
+            all_config = yaml.safe_load(f) or {}
+        return all_config.get("plugins", {}).get("hermes-memory-store", {}) or {}
+    except Exception:
+        return {}
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class HolographicMemoryProvider(MemoryProvider):
+    """Holographic memory with structured facts, entity resolution, and HRR retrieval."""
+
+    def __init__(self, config: dict | None = None):
+        self._config = config or _load_plugin_config()
+        self._store = None
+        self._retriever = None
+        self._min_trust = float(self._config.get("min_trust_threshold", 0.3))
+
+    @property
+    def name(self) -> str:
+        return "holographic"
+
+    def is_available(self) -> bool:
+        return True  # SQLite is always available, numpy is optional
+
+    def save_config(self, values, hermes_home):
+        """Write config to config.yaml under plugins.hermes-memory-store."""
+        from pathlib import Path
+        config_path = Path(hermes_home) / "config.yaml"
+        try:
+            import yaml
+            existing = {}
+            if config_path.exists():
+                with open(config_path) as f:
+                    existing = yaml.safe_load(f) or {}
+            existing.setdefault("plugins", {})
+            existing["plugins"]["hermes-memory-store"] = values
+            with open(config_path, "w") as f:
+                yaml.dump(existing, f, default_flow_style=False)
+        except Exception:
+            pass
+
+    def get_config_schema(self):
+        from hermes_constants import display_hermes_home
+        _default_db = f"{display_hermes_home()}/memory_store.db"
+        return [
+            {"key": "db_path", "description": "SQLite database path", "default": _default_db},
+            {"key": "auto_extract", "description": "Auto-extract facts at session end", "default": "false", "choices": ["true", "false"]},
+            {"key": "default_trust", "description": "Default trust score for new facts", "default": "0.5"},
+            {"key": "hrr_dim", "description": "HRR vector dimensions", "default": "1024"},
+        ]
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        from hermes_constants import get_hermes_home
+        _hermes_home = str(get_hermes_home())
+        _default_db = _hermes_home + "/memory_store.db"
+        db_path = self._config.get("db_path", _default_db)
+        # Expand $HERMES_HOME in user-supplied paths so config values like
+        # "$HERMES_HOME/memory_store.db" or "~/.hermes/memory_store.db" both
+        # resolve to the active profile's directory.
+        if isinstance(db_path, str):
+            db_path = db_path.replace("$HERMES_HOME", _hermes_home)
+            db_path = db_path.replace("${HERMES_HOME}", _hermes_home)
+        default_trust = float(self._config.get("default_trust", 0.5))
+        hrr_dim = int(self._config.get("hrr_dim", 1024))
+        hrr_weight = float(self._config.get("hrr_weight", 0.3))
+        temporal_decay = int(self._config.get("temporal_decay_half_life", 0))
+
+        self._store = MemoryStore(db_path=db_path, default_trust=default_trust, hrr_dim=hrr_dim)
+        self._retriever = FactRetriever(
+            store=self._store,
+            temporal_decay_half_life=temporal_decay,
+            hrr_weight=hrr_weight,
+            hrr_dim=hrr_dim,
+        )
+        self._session_id = session_id
+
+    def system_prompt_block(self) -> str:
+        if not self._store:
+            return ""
+        try:
+            total = self._store._conn.execute(
+                "SELECT COUNT(*) FROM facts"
+            ).fetchone()[0]
+        except Exception:
+            total = 0
+        if total == 0:
+            return (
+                "# Holographic Memory\n"
+                "Active. Empty fact store — proactively add facts the user would expect you to remember.\n"
+                "Use fact_store(action='add') to store durable structured facts about people, projects, preferences, decisions.\n"
+                "Use fact_feedback to rate facts after using them (trains trust scores)."
+            )
+        return (
+            f"# Holographic Memory\n"
+            f"Active. {total} facts stored with entity resolution and trust scoring.\n"
+            f"Use fact_store to search, probe entities, reason across entities, or add facts.\n"
+            f"Use fact_feedback to rate facts after using them (trains trust scores)."
+        )
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        if not self._retriever or not query:
+            return ""
+        try:
+            results = self._retriever.search(query, min_trust=self._min_trust, limit=5)
+            if not results:
+                return ""
+            lines = []
+            for r in results:
+                trust = r.get("trust_score", r.get("trust", 0))
+                lines.append(f"- [{trust:.1f}] {r.get('content', '')}")
+            return "## Holographic Memory\n" + "\n".join(lines)
+        except Exception as e:
+            logger.debug("Holographic prefetch failed: %s", e)
+            return ""
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        # Holographic memory stores explicit facts via tools, not auto-sync.
+        # The on_session_end hook handles auto-extraction if configured.
+        pass
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        return [FACT_STORE_SCHEMA, FACT_FEEDBACK_SCHEMA]
+
+    def handle_tool_call(self, tool_name: str, args: Dict[str, Any], **kwargs) -> str:
+        if tool_name == "fact_store":
+            return self._handle_fact_store(args)
+        elif tool_name == "fact_feedback":
+            return self._handle_fact_feedback(args)
+        return json.dumps({"error": f"Unknown tool: {tool_name}"})
+
+    def on_session_end(self, messages: List[Dict[str, Any]]) -> None:
+        if not self._config.get("auto_extract", False):
+            return
+        if not self._store or not messages:
+            return
+        self._auto_extract_facts(messages)
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        """Mirror built-in memory writes as facts."""
+        if action == "add" and self._store and content:
+            try:
+                category = "user_pref" if target == "user" else "general"
+                self._store.add_fact(content, category=category)
+            except Exception as e:
+                logger.debug("Holographic memory_write mirror failed: %s", e)
+
+    def shutdown(self) -> None:
+        self._store = None
+        self._retriever = None
+
+    # -- Tool handlers -------------------------------------------------------
+
+    def _handle_fact_store(self, args: dict) -> str:
+        try:
+            action = args["action"]
+            store = self._store
+            retriever = self._retriever
+
+            if action == "add":
+                fact_id = store.add_fact(
+                    args["content"],
+                    category=args.get("category", "general"),
+                    tags=args.get("tags", ""),
+                )
+                return json.dumps({"fact_id": fact_id, "status": "added"})
+
+            elif action == "search":
+                results = retriever.search(
+                    args["query"],
+                    category=args.get("category"),
+                    min_trust=float(args.get("min_trust", self._min_trust)),
+                    limit=int(args.get("limit", 10)),
+                )
+                return json.dumps({"results": results, "count": len(results)})
+
+            elif action == "probe":
+                results = retriever.probe(
+                    args["entity"],
+                    category=args.get("category"),
+                    limit=int(args.get("limit", 10)),
+                )
+                return json.dumps({"results": results, "count": len(results)})
+
+            elif action == "related":
+                results = retriever.related(
+                    args["entity"],
+                    category=args.get("category"),
+                    limit=int(args.get("limit", 10)),
+                )
+                return json.dumps({"results": results, "count": len(results)})
+
+            elif action == "reason":
+                entities = args.get("entities", [])
+                if not entities:
+                    return json.dumps({"error": "reason requires 'entities' list"})
+                results = retriever.reason(
+                    entities,
+                    category=args.get("category"),
+                    limit=int(args.get("limit", 10)),
+                )
+                return json.dumps({"results": results, "count": len(results)})
+
+            elif action == "contradict":
+                results = retriever.contradict(
+                    category=args.get("category"),
+                    limit=int(args.get("limit", 10)),
+                )
+                return json.dumps({"results": results, "count": len(results)})
+
+            elif action == "update":
+                updated = store.update_fact(
+                    int(args["fact_id"]),
+                    content=args.get("content"),
+                    trust_delta=float(args["trust_delta"]) if "trust_delta" in args else None,
+                    tags=args.get("tags"),
+                    category=args.get("category"),
+                )
+                return json.dumps({"updated": updated})
+
+            elif action == "remove":
+                removed = store.remove_fact(int(args["fact_id"]))
+                return json.dumps({"removed": removed})
+
+            elif action == "list":
+                facts = store.list_facts(
+                    category=args.get("category"),
+                    min_trust=float(args.get("min_trust", 0.0)),
+                    limit=int(args.get("limit", 10)),
+                )
+                return json.dumps({"facts": facts, "count": len(facts)})
+
+            else:
+                return json.dumps({"error": f"Unknown action: {action}"})
+
+        except KeyError as exc:
+            return json.dumps({"error": f"Missing required argument: {exc}"})
+        except Exception as exc:
+            return json.dumps({"error": str(exc)})
+
+    def _handle_fact_feedback(self, args: dict) -> str:
+        try:
+            fact_id = int(args["fact_id"])
+            helpful = args["action"] == "helpful"
+            result = self._store.record_feedback(fact_id, helpful=helpful)
+            return json.dumps(result)
+        except KeyError as exc:
+            return json.dumps({"error": f"Missing required argument: {exc}"})
+        except Exception as exc:
+            return json.dumps({"error": str(exc)})
+
+    # -- Auto-extraction (on_session_end) ------------------------------------
+
+    def _auto_extract_facts(self, messages: list) -> None:
+        _PREF_PATTERNS = [
+            re.compile(r'\bI\s+(?:prefer|like|love|use|want|need)\s+(.+)', re.IGNORECASE),
+            re.compile(r'\bmy\s+(?:favorite|preferred|default)\s+\w+\s+is\s+(.+)', re.IGNORECASE),
+            re.compile(r'\bI\s+(?:always|never|usually)\s+(.+)', re.IGNORECASE),
+        ]
+        _DECISION_PATTERNS = [
+            re.compile(r'\bwe\s+(?:decided|agreed|chose)\s+(?:to\s+)?(.+)', re.IGNORECASE),
+            re.compile(r'\bthe\s+project\s+(?:uses|needs|requires)\s+(.+)', re.IGNORECASE),
+        ]
+
+        extracted = 0
+        for msg in messages:
+            if msg.get("role") != "user":
+                continue
+            content = msg.get("content", "")
+            if not isinstance(content, str) or len(content) < 10:
+                continue
+
+            for pattern in _PREF_PATTERNS:
+                if pattern.search(content):
+                    try:
+                        self._store.add_fact(content[:400], category="user_pref")
+                        extracted += 1
+                    except Exception:
+                        pass
+                    break
+
+            for pattern in _DECISION_PATTERNS:
+                if pattern.search(content):
+                    try:
+                        self._store.add_fact(content[:400], category="project")
+                        extracted += 1
+                    except Exception:
+                        pass
+                    break
+
+        if extracted:
+            logger.info("Auto-extracted %d facts from conversation", extracted)
+
+
+# ---------------------------------------------------------------------------
+# Plugin entry point
+# ---------------------------------------------------------------------------
+
+def register(ctx) -> None:
+    """Register the holographic memory provider with the plugin system."""
+    config = _load_plugin_config()
+    provider = HolographicMemoryProvider(config=config)
+    ctx.register_memory_provider(provider)
@@ -0,0 +1,203 @@
+"""Holographic Reduced Representations (HRR) with phase encoding.
+
+HRRs are a vector symbolic architecture for encoding compositional structure
+into fixed-width distributed representations. This module uses *phase vectors*:
+each concept is a vector of angles in [0, 2π). The algebraic operations are:
+
+  bind   — circular convolution (phase addition)  — associates two concepts
+  unbind — circular correlation (phase subtraction) — retrieves a bound value
+  bundle — superposition (circular mean)           — merges multiple concepts
+
+Phase encoding is numerically stable, avoids the magnitude collapse of
+traditional complex-number HRRs, and maps cleanly to cosine similarity.
+
+Atoms are generated deterministically from SHA-256 so representations are
+identical across processes, machines, and language versions.
+
+References:
+  Plate (1995) — Holographic Reduced Representations
+  Gayler (2004) — Vector Symbolic Architectures answer Jackendoff's challenges
+"""
+
+import hashlib
+import logging
+import struct
+import math
+
+try:
+    import numpy as np
+    _HAS_NUMPY = True
+except ImportError:
+    _HAS_NUMPY = False
+
+logger = logging.getLogger(__name__)
+
+_TWO_PI = 2.0 * math.pi
+
+
+def _require_numpy() -> None:
+    if not _HAS_NUMPY:
+        raise RuntimeError("numpy is required for holographic operations")
+
+
+def encode_atom(word: str, dim: int = 1024) -> "np.ndarray":
+    """Deterministic phase vector via SHA-256 counter blocks.
+
+    Uses hashlib (not numpy RNG) for cross-platform reproducibility.
+
+    Algorithm:
+    - Generate enough SHA-256 blocks by hashing f"{word}:{i}" for i=0,1,2,...
+    - Concatenate digests, interpret as uint16 values via struct.unpack
+    - Scale to [0, 2π): phases = values * (2π / 65536)
+    - Truncate to dim elements
+    - Returns np.float64 array of shape (dim,)
+    """
+    _require_numpy()
+
+    # Each SHA-256 digest is 32 bytes = 16 uint16 values.
+    values_per_block = 16
+    blocks_needed = math.ceil(dim / values_per_block)
+
+    uint16_values: list[int] = []
+    for i in range(blocks_needed):
+        digest = hashlib.sha256(f"{word}:{i}".encode()).digest()
+        uint16_values.extend(struct.unpack("<16H", digest))
+
+    phases = np.array(uint16_values[:dim], dtype=np.float64) * (_TWO_PI / 65536.0)
+    return phases
+
+
+def bind(a: "np.ndarray", b: "np.ndarray") -> "np.ndarray":
+    """Circular convolution = element-wise phase addition.
+
+    Binding associates two concepts into a single composite vector.
+    The result is dissimilar to both inputs (quasi-orthogonal).
+    """
+    _require_numpy()
+    return (a + b) % _TWO_PI
+
+
+def unbind(memory: "np.ndarray", key: "np.ndarray") -> "np.ndarray":
+    """Circular correlation = element-wise phase subtraction.
+
+    Unbinding retrieves the value associated with a key from a memory vector.
+    unbind(bind(a, b), a) ≈ b  (up to superposition noise)
+    """
+    _require_numpy()
+    return (memory - key) % _TWO_PI
+
+
+def bundle(*vectors: "np.ndarray") -> "np.ndarray":
+    """Superposition via circular mean of complex exponentials.
+
+    Bundling merges multiple vectors into one that is similar to each input.
+    The result can hold O(sqrt(dim)) items before similarity degrades.
+    """
+    _require_numpy()
+    complex_sum = np.sum([np.exp(1j * v) for v in vectors], axis=0)
+    return np.angle(complex_sum) % _TWO_PI
+
+
+def similarity(a: "np.ndarray", b: "np.ndarray") -> float:
+    """Phase cosine similarity. Range [-1, 1].
+
+    Returns 1.0 for identical vectors, near 0.0 for random (unrelated) vectors,
+    and -1.0 for perfectly anti-correlated vectors.
+    """
+    _require_numpy()
+    return float(np.mean(np.cos(a - b)))
+
+
+def encode_text(text: str, dim: int = 1024) -> "np.ndarray":
+    """Bag-of-words: bundle of atom vectors for each token.
+
+    Tokenizes by lowercasing, splitting on whitespace, and stripping
+    leading/trailing punctuation from each token.
+
+    Returns bundle of all token atom vectors.
+    If text is empty or produces no tokens, returns encode_atom("__hrr_empty__", dim).
+    """
+    _require_numpy()
+
+    tokens = [
+        token.strip(".,!?;:\"'()[]{}")
+        for token in text.lower().split()
+    ]
+    tokens = [t for t in tokens if t]
+
+    if not tokens:
+        return encode_atom("__hrr_empty__", dim)
+
+    atom_vectors = [encode_atom(token, dim) for token in tokens]
+    return bundle(*atom_vectors)
+
+
+def encode_fact(content: str, entities: list[str], dim: int = 1024) -> "np.ndarray":
+    """Structured encoding: content bound to ROLE_CONTENT, each entity bound to ROLE_ENTITY, all bundled.
+
+    Role vectors are reserved atoms: "__hrr_role_content__", "__hrr_role_entity__"
+
+    Components:
+    1. bind(encode_text(content, dim), encode_atom("__hrr_role_content__", dim))
+    2. For each entity: bind(encode_atom(entity.lower(), dim), encode_atom("__hrr_role_entity__", dim))
+    3. bundle all components together
+
+    This enables algebraic extraction:
+        unbind(fact, bind(entity, ROLE_ENTITY)) ≈ content_vector
+    """
+    _require_numpy()
+
+    role_content = encode_atom("__hrr_role_content__", dim)
+    role_entity = encode_atom("__hrr_role_entity__", dim)
+
+    components: list[np.ndarray] = [
+        bind(encode_text(content, dim), role_content)
+    ]
+
+    for entity in entities:
+        components.append(bind(encode_atom(entity.lower(), dim), role_entity))
+
+    return bundle(*components)
+
+
+def phases_to_bytes(phases: "np.ndarray") -> bytes:
+    """Serialize phase vector to bytes. float64 tobytes — 8 KB at dim=1024."""
+    _require_numpy()
+    return phases.tobytes()
+
+
+def bytes_to_phases(data: bytes) -> "np.ndarray":
+    """Deserialize bytes back to phase vector. Inverse of phases_to_bytes.
+
+    The .copy() call is required because frombuffer returns a read-only view
+    backed by the bytes object; callers expect a mutable array.
+    """
+    _require_numpy()
+    return np.frombuffer(data, dtype=np.float64).copy()
+
+
+def snr_estimate(dim: int, n_items: int) -> float:
+    """Signal-to-noise ratio estimate for holographic storage.
+
+    SNR = sqrt(dim / n_items) when n_items > 0, else inf.
+
+    The SNR falls below 2.0 when n_items > dim / 4, meaning retrieval
+    errors become likely. Logs a warning when this threshold is crossed.
+    """
+    _require_numpy()
+
+    if n_items <= 0:
+        return float("inf")
+
+    snr = math.sqrt(dim / n_items)
+
+    if snr < 2.0:
+        logger.warning(
+            "HRR storage near capacity: SNR=%.2f (dim=%d, n_items=%d). "
+            "Retrieval accuracy may degrade. Consider increasing dim or reducing stored items.",
+            snr,
+            dim,
+            n_items,
+        )
+
+    return snr
@@ -0,0 +1,5 @@
+name: holographic
+version: 0.1.0
+description: "Holographic memory — local SQLite fact store with FTS5 search, trust scoring, and HRR-based compositional retrieval."
+hooks:
+  - on_session_end
@@ -0,0 +1,593 @@
+"""Hybrid keyword/BM25 retrieval for the memory store.
+
+Ported from KIK memory_agent.py — combines FTS5 full-text search with
+Jaccard similarity reranking and trust-weighted scoring.
+"""
+
+from __future__ import annotations
+
+import math
+from datetime import datetime, timezone
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from .store import MemoryStore
+
+try:
+    from . import holographic as hrr
+except ImportError:
+    import holographic as hrr  # type: ignore[no-redef]
+
+
+class FactRetriever:
+    """Multi-strategy fact retrieval with trust-weighted scoring."""
+
+    def __init__(
+        self,
+        store: MemoryStore,
+        temporal_decay_half_life: int = 0,  # days, 0 = disabled
+        fts_weight: float = 0.4,
+        jaccard_weight: float = 0.3,
+        hrr_weight: float = 0.3,
+        hrr_dim: int = 1024,
+    ):
+        self.store = store
+        self.half_life = temporal_decay_half_life
+        self.hrr_dim = hrr_dim
+
+        # Auto-redistribute weights if numpy unavailable
+        if hrr_weight > 0 and not hrr._HAS_NUMPY:
+            fts_weight = 0.6
+            jaccard_weight = 0.4
+            hrr_weight = 0.0
+
+        self.fts_weight = fts_weight
+        self.jaccard_weight = jaccard_weight
+        self.hrr_weight = hrr_weight
+
+    def search(
+        self,
+        query: str,
+        category: str | None = None,
+        min_trust: float = 0.3,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Hybrid search: FTS5 candidates → Jaccard rerank → trust weighting.
+
+        Pipeline:
+        1. FTS5 search: Get limit*3 candidates from SQLite full-text search
+        2. Jaccard boost: Token overlap between query and fact content
+        3. Trust weighting: final_score = relevance * trust_score
+        4. Temporal decay (optional): decay = 0.5^(age_days / half_life)
+
+        Returns list of dicts with fact data + 'score' field, sorted by score desc.
+        """
+        # Stage 1: Get FTS5 candidates (more than limit for reranking headroom)
+        candidates = self._fts_candidates(query, category, min_trust, limit * 3)
+
+        if not candidates:
+            return []
+
+        # Stage 2: Rerank with Jaccard + trust + optional decay
+        query_tokens = self._tokenize(query)
+        scored = []
+
+        for fact in candidates:
+            content_tokens = self._tokenize(fact["content"])
+            tag_tokens = self._tokenize(fact.get("tags", ""))
+            all_tokens = content_tokens | tag_tokens
+
+            jaccard = self._jaccard_similarity(query_tokens, all_tokens)
+            fts_score = fact.get("fts_rank", 0.0)
+
+            # HRR similarity
+            if self.hrr_weight > 0 and fact.get("hrr_vector"):
+                fact_vec = hrr.bytes_to_phases(fact["hrr_vector"])
+                query_vec = hrr.encode_text(query, self.hrr_dim)
+                hrr_sim = (hrr.similarity(query_vec, fact_vec) + 1.0) / 2.0  # shift to [0,1]
+            else:
+                hrr_sim = 0.5  # neutral
+
+            # Combine FTS5 + Jaccard + HRR
+            relevance = (self.fts_weight * fts_score
+                        + self.jaccard_weight * jaccard
+                        + self.hrr_weight * hrr_sim)
+
+            # Trust weighting
+            score = relevance * fact["trust_score"]
+
+            # Optional temporal decay
+            if self.half_life > 0:
+                score *= self._temporal_decay(fact.get("updated_at") or fact.get("created_at"))
+
+            fact["score"] = score
+            scored.append(fact)
+
+        # Sort by score descending, return top limit
+        scored.sort(key=lambda x: x["score"], reverse=True)
+        results = scored[:limit]
+        # Strip raw HRR bytes — callers expect JSON-serializable dicts
+        for fact in results:
+            fact.pop("hrr_vector", None)
+        return results
+
+    def probe(
+        self,
+        entity: str,
+        category: str | None = None,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Compositional entity query using HRR algebra.
+
+        Unbinds entity from memory bank to extract associated content.
+        This is NOT keyword search — it uses algebraic structure to find facts
+        where the entity plays a structural role.
+
+        Falls back to FTS5 search if numpy unavailable.
+        """
+        if not hrr._HAS_NUMPY:
+            # Fallback to keyword search on entity name
+            return self.search(entity, category=category, limit=limit)
+
+        conn = self.store._conn
+
+        # Encode entity as role-bound vector
+        role_entity = hrr.encode_atom("__hrr_role_entity__", self.hrr_dim)
+        entity_vec = hrr.encode_atom(entity.lower(), self.hrr_dim)
+        probe_key = hrr.bind(entity_vec, role_entity)
+
+        # Try category-specific bank first, then all facts
+        if category:
+            bank_name = f"cat:{category}"
+            bank_row = conn.execute(
+                "SELECT vector FROM memory_banks WHERE bank_name = ?",
+                (bank_name,),
+            ).fetchone()
+            if bank_row:
+                bank_vec = hrr.bytes_to_phases(bank_row["vector"])
+                extracted = hrr.unbind(bank_vec, probe_key)
+                # Use extracted signal to score individual facts
+                return self._score_facts_by_vector(
+                    extracted, category=category, limit=limit
+                )
+
+        # Score against individual fact vectors directly
+        where = "WHERE hrr_vector IS NOT NULL"
+        params: list = []
+        if category:
+            where += " AND category = ?"
+            params.append(category)
+
+        rows = conn.execute(
+            f"""
+            SELECT fact_id, content, category, tags, trust_score,
+                   retrieval_count, helpful_count, created_at, updated_at,
+                   hrr_vector
+            FROM facts
+            {where}
+            """,
+            params,
+        ).fetchall()
+
+        if not rows:
+            # Final fallback: keyword search
+            return self.search(entity, category=category, limit=limit)
+
+        scored = []
+        for row in rows:
+            fact = dict(row)
+            fact_vec = hrr.bytes_to_phases(fact.pop("hrr_vector"))
+            # Unbind probe key from fact to see if entity is structurally present
+            residual = hrr.unbind(fact_vec, probe_key)
+            # Compare residual against content signal
+            role_content = hrr.encode_atom("__hrr_role_content__", self.hrr_dim)
+            content_vec = hrr.bind(hrr.encode_text(fact["content"], self.hrr_dim), role_content)
+            sim = hrr.similarity(residual, content_vec)
+            fact["score"] = (sim + 1.0) / 2.0 * fact["trust_score"]
+            scored.append(fact)
+
+        scored.sort(key=lambda x: x["score"], reverse=True)
+        return scored[:limit]
+
+    def related(
+        self,
+        entity: str,
+        category: str | None = None,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Discover facts that share structural connections with an entity.
+
+        Unlike probe (which finds facts *about* an entity), related finds
+        facts that are connected through shared context — e.g., other entities
+        mentioned alongside this one, or content that overlaps structurally.
+
+        Falls back to FTS5 search if numpy unavailable.
+        """
+        if not hrr._HAS_NUMPY:
+            return self.search(entity, category=category, limit=limit)
+
+        conn = self.store._conn
+
+        # Encode entity as a bare atom (not role-bound — we want ANY structural match)
+        entity_vec = hrr.encode_atom(entity.lower(), self.hrr_dim)
+
+        # Get all facts with vectors
+        where = "WHERE hrr_vector IS NOT NULL"
+        params: list = []
+        if category:
+            where += " AND category = ?"
+            params.append(category)
+
+        rows = conn.execute(
+            f"""
+            SELECT fact_id, content, category, tags, trust_score,
+                   retrieval_count, helpful_count, created_at, updated_at,
+                   hrr_vector
+            FROM facts
+            {where}
+            """,
+            params,
+        ).fetchall()
+
+        if not rows:
+            return self.search(entity, category=category, limit=limit)
+
+        # Score each fact by how much the entity's atom appears in its vector
+        # This catches both role-bound entity matches AND content word matches
+        scored = []
+        for row in rows:
+            fact = dict(row)
+            fact_vec = hrr.bytes_to_phases(fact.pop("hrr_vector"))
+
+            # Check structural similarity: unbind entity from fact
+            residual = hrr.unbind(fact_vec, entity_vec)
+            # A high-similarity residual to ANY known role vector means this entity
+            # plays a structural role in the fact
+            role_entity = hrr.encode_atom("__hrr_role_entity__", self.hrr_dim)
+            role_content = hrr.encode_atom("__hrr_role_content__", self.hrr_dim)
+
+            entity_role_sim = hrr.similarity(residual, role_entity)
+            content_role_sim = hrr.similarity(residual, role_content)
+            # Take the max — entity could appear in either role
+            best_sim = max(entity_role_sim, content_role_sim)
+
+            fact["score"] = (best_sim + 1.0) / 2.0 * fact["trust_score"]
+            scored.append(fact)
+
+        scored.sort(key=lambda x: x["score"], reverse=True)
+        return scored[:limit]
+
+    def reason(
+        self,
+        entities: list[str],
+        category: str | None = None,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Multi-entity compositional query — vector-space JOIN.
+
+        Given multiple entities, algebraically intersects their structural
+        connections to find facts related to ALL of them simultaneously.
+        This is compositional reasoning that no embedding DB can do.
+
+        Example: reason(["peppi", "backend"]) finds facts where peppi AND
+        backend both play structural roles — without keyword matching.
+
+        Falls back to FTS5 search if numpy unavailable.
+        """
+        if not hrr._HAS_NUMPY or not entities:
+            # Fallback: search with all entities as keywords
+            query = " ".join(entities)
+            return self.search(query, category=category, limit=limit)
+
+        conn = self.store._conn
+        role_entity = hrr.encode_atom("__hrr_role_entity__", self.hrr_dim)
+
+        # For each entity, compute what the bank "remembers" about it
+        # by unbinding entity+role from each fact vector
+        entity_residuals = []
+        for entity in entities:
+            entity_vec = hrr.encode_atom(entity.lower(), self.hrr_dim)
+            probe_key = hrr.bind(entity_vec, role_entity)
+            entity_residuals.append(probe_key)
+
+        # Get all facts with vectors
+        where = "WHERE hrr_vector IS NOT NULL"
+        params: list = []
+        if category:
+            where += " AND category = ?"
+            params.append(category)
+
+        rows = conn.execute(
+            f"""
+            SELECT fact_id, content, category, tags, trust_score,
+                   retrieval_count, helpful_count, created_at, updated_at,
+                   hrr_vector
+            FROM facts
+            {where}
+            """,
+            params,
+        ).fetchall()
+
+        if not rows:
+            query = " ".join(entities)
+            return self.search(query, category=category, limit=limit)
+
+        # Score each fact by how much EACH entity is structurally present.
+        # A fact scores high only if ALL entities have structural presence
+        # (AND semantics via min, vs OR which would use mean/max).
+        role_content = hrr.encode_atom("__hrr_role_content__", self.hrr_dim)
+
+        scored = []
+        for row in rows:
+            fact = dict(row)
+            fact_vec = hrr.bytes_to_phases(fact.pop("hrr_vector"))
+
+            entity_scores = []
+            for probe_key in entity_residuals:
+                residual = hrr.unbind(fact_vec, probe_key)
+                sim = hrr.similarity(residual, role_content)
+                entity_scores.append(sim)
+
+            min_sim = min(entity_scores)
+            fact["score"] = (min_sim + 1.0) / 2.0 * fact["trust_score"]
+            scored.append(fact)
+
+        scored.sort(key=lambda x: x["score"], reverse=True)
+        return scored[:limit]
+
+    def contradict(
+        self,
+        category: str | None = None,
+        threshold: float = 0.3,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Find potentially contradictory facts via entity overlap + content divergence.
+
+        Two facts contradict when they share entities (same subject) but have
+        low content-vector similarity (different claims). This is automated
+        memory hygiene — no other memory system does this.
+
+        Returns pairs of facts with a contradiction score.
+        Falls back to empty list if numpy unavailable.
+        """
+        if not hrr._HAS_NUMPY:
+            return []
+
+        conn = self.store._conn
+
+        # Get all facts with vectors and their linked entities
+        where = "WHERE f.hrr_vector IS NOT NULL"
+        params: list = []
+        if category:
+            where += " AND f.category = ?"
+            params.append(category)
+
+        rows = conn.execute(
+            f"""
+            SELECT f.fact_id, f.content, f.category, f.tags, f.trust_score,
+                   f.created_at, f.updated_at, f.hrr_vector
+            FROM facts f
+            {where}
+            """,
+            params,
+        ).fetchall()
+
+        if len(rows) < 2:
+            return []
+
+        # Guard against O(n²) explosion on large fact stores.
+        # At 500 facts, that's ~125K comparisons — acceptable.
+        # Above that, only check the most recently updated facts.
+        _MAX_CONTRADICT_FACTS = 500
+        if len(rows) > _MAX_CONTRADICT_FACTS:
+            rows = sorted(rows, key=lambda r: r["updated_at"] or r["created_at"], reverse=True)
+            rows = rows[:_MAX_CONTRADICT_FACTS]
+
+        # Build entity sets per fact
+        fact_entities: dict[int, set[str]] = {}
+        for row in rows:
+            fid = row["fact_id"]
+            entity_rows = conn.execute(
+                """
+                SELECT e.name FROM entities e
+                JOIN fact_entities fe ON fe.entity_id = e.entity_id
+                WHERE fe.fact_id = ?
+                """,
+                (fid,),
+            ).fetchall()
+            fact_entities[fid] = {r["name"].lower() for r in entity_rows}
+
+        # Compare all pairs: high entity overlap + low content similarity = contradiction
+        facts = [dict(r) for r in rows]
+        contradictions = []
+
+        for i in range(len(facts)):
+            for j in range(i + 1, len(facts)):
+                f1, f2 = facts[i], facts[j]
+                ents1 = fact_entities.get(f1["fact_id"], set())
+                ents2 = fact_entities.get(f2["fact_id"], set())
+
+                if not ents1 or not ents2:
+                    continue
+
+                # Entity overlap (Jaccard)
+                entity_overlap = len(ents1 & ents2) / len(ents1 | ents2) if (ents1 | ents2) else 0.0
+
+                if entity_overlap < 0.3:
+                    continue  # Not enough entity overlap to be contradictory
+
+                # Content similarity via HRR vectors
+                v1 = hrr.bytes_to_phases(f1["hrr_vector"])
+                v2 = hrr.bytes_to_phases(f2["hrr_vector"])
+                content_sim = hrr.similarity(v1, v2)
+
+                # High entity overlap + low content similarity = potential contradiction
+                # contradiction_score: higher = more contradictory
+                contradiction_score = entity_overlap * (1.0 - (content_sim + 1.0) / 2.0)
+
+                if contradiction_score >= threshold:
+                    # Strip hrr_vector from output (not JSON serializable)
+                    f1_clean = {k: v for k, v in f1.items() if k != "hrr_vector"}
+                    f2_clean = {k: v for k, v in f2.items() if k != "hrr_vector"}
+                    contradictions.append({
+                        "fact_a": f1_clean,
+                        "fact_b": f2_clean,
+                        "entity_overlap": round(entity_overlap, 3),
+                        "content_similarity": round(content_sim, 3),
+                        "contradiction_score": round(contradiction_score, 3),
+                        "shared_entities": sorted(ents1 & ents2),
+                    })
+
+        contradictions.sort(key=lambda x: x["contradiction_score"], reverse=True)
+        return contradictions[:limit]
+
+    def _score_facts_by_vector(
+        self,
+        target_vec: "np.ndarray",
+        category: str | None = None,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Score facts by similarity to a target vector."""
+        conn = self.store._conn
+
+        where = "WHERE hrr_vector IS NOT NULL"
+        params: list = []
+        if category:
+            where += " AND category = ?"
+            params.append(category)
+
+        rows = conn.execute(
+            f"""
+            SELECT fact_id, content, category, tags, trust_score,
+                   retrieval_count, helpful_count, created_at, updated_at,
+                   hrr_vector
+            FROM facts
+            {where}
+            """,
+            params,
+        ).fetchall()
+
+        scored = []
+        for row in rows:
+            fact = dict(row)
+            fact_vec = hrr.bytes_to_phases(fact.pop("hrr_vector"))
+            sim = hrr.similarity(target_vec, fact_vec)
+            fact["score"] = (sim + 1.0) / 2.0 * fact["trust_score"]
+            scored.append(fact)
+
+        scored.sort(key=lambda x: x["score"], reverse=True)
+        return scored[:limit]
+
+    def _fts_candidates(
+        self,
+        query: str,
+        category: str | None,
+        min_trust: float,
+        limit: int,
+    ) -> list[dict]:
+        """Get raw FTS5 candidates from the store.
+
+        Uses the store's database connection directly for FTS5 MATCH
+        with rank scoring. Normalizes FTS5 rank to [0, 1] range.
+        """
+        conn = self.store._conn
+
+        # Build query - FTS5 rank is negative (lower = better match)
+        # We need to join facts_fts with facts to get all columns
+        params: list = []
+        where_clauses = ["facts_fts MATCH ?"]
+        params.append(query)
+
+        if category:
+            where_clauses.append("f.category = ?")
+            params.append(category)
+
+        where_clauses.append("f.trust_score >= ?")
+        params.append(min_trust)
+
+        where_sql = " AND ".join(where_clauses)
+
+        sql = f"""
+            SELECT f.*, facts_fts.rank as fts_rank_raw
+            FROM facts_fts
+            JOIN facts f ON f.fact_id = facts_fts.rowid
+            WHERE {where_sql}
+            ORDER BY facts_fts.rank
+            LIMIT ?
+        """
+        params.append(limit)
+
+        try:
+            rows = conn.execute(sql, params).fetchall()
+        except Exception:
+            # FTS5 MATCH can fail on malformed queries — fall back to empty
+            return []
+
+        if not rows:
+            return []
+
+        # Normalize FTS5 rank: rank is negative, lower = better
+        # Convert to positive score in [0, 1] range
+        raw_ranks = [abs(row["fts_rank_raw"]) for row in rows]
+        max_rank = max(raw_ranks) if raw_ranks else 1.0
+        max_rank = max(max_rank, 1e-6)  # avoid div by zero
+
+        results = []
+        for row, raw_rank in zip(rows, raw_ranks):
+            fact = dict(row)
+            fact.pop("fts_rank_raw", None)
+            fact["fts_rank"] = raw_rank / max_rank  # normalize to [0, 1]
+            results.append(fact)
+
+        return results
+
+    @staticmethod
+    def _tokenize(text: str) -> set[str]:
+        """Simple whitespace tokenization with lowercasing.
+
+        Strips common punctuation. No stemming/lemmatization (Phase 1).
+        """
+        if not text:
+            return set()
+        # Split on whitespace, lowercase, strip punctuation
+        tokens = set()
+        for word in text.lower().split():
+            cleaned = word.strip(".,;:!?\"'()[]{}#@<>")
+            if cleaned:
+                tokens.add(cleaned)
+        return tokens
+
+    @staticmethod
+    def _jaccard_similarity(set_a: set, set_b: set) -> float:
+        """Jaccard similarity coefficient: |A ∩ B| / |A ∪ B|."""
+        if not set_a or not set_b:
+            return 0.0
+        intersection = len(set_a & set_b)
+        union = len(set_a | set_b)
+        return intersection / union if union > 0 else 0.0
+
+    def _temporal_decay(self, timestamp_str: str | None) -> float:
+        """Exponential decay: 0.5^(age_days / half_life_days).
+
+        Returns 1.0 if decay is disabled or timestamp is missing.
+        """
+        if not self.half_life or not timestamp_str:
+            return 1.0
+
+        try:
+            if isinstance(timestamp_str, str):
+                # Parse ISO format timestamp from SQLite
+                ts = datetime.fromisoformat(timestamp_str.replace("Z", "+00:00"))
+            else:
+                ts = timestamp_str
+
+            if ts.tzinfo is None:
+                ts = ts.replace(tzinfo=timezone.utc)
+
+            age_days = (datetime.now(timezone.utc) - ts).total_seconds() / 86400
+            if age_days < 0:
+                return 1.0
+
+            return math.pow(0.5, age_days / self.half_life)
+        except (ValueError, TypeError):
+            return 1.0
@@ -0,0 +1,575 @@
+"""
+SQLite-backed fact store with entity resolution and trust scoring.
+Single-user Hermes memory store plugin.
+"""
+
+import re
+import sqlite3
+import threading
+from datetime import datetime
+from pathlib import Path
+
+try:
+    from . import holographic as hrr
+except ImportError:
+    import holographic as hrr  # type: ignore[no-redef]
+
+_SCHEMA = """
+CREATE TABLE IF NOT EXISTS facts (
+    fact_id         INTEGER PRIMARY KEY AUTOINCREMENT,
+    content         TEXT NOT NULL UNIQUE,
+    category        TEXT DEFAULT 'general',
+    tags            TEXT DEFAULT '',
+    trust_score     REAL DEFAULT 0.5,
+    retrieval_count INTEGER DEFAULT 0,
+    helpful_count   INTEGER DEFAULT 0,
+    created_at      TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+    updated_at      TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+    hrr_vector      BLOB
+);
+
+CREATE TABLE IF NOT EXISTS entities (
+    entity_id   INTEGER PRIMARY KEY AUTOINCREMENT,
+    name        TEXT NOT NULL,
+    entity_type TEXT DEFAULT 'unknown',
+    aliases     TEXT DEFAULT '',
+    created_at  TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS fact_entities (
+    fact_id   INTEGER REFERENCES facts(fact_id),
+    entity_id INTEGER REFERENCES entities(entity_id),
+    PRIMARY KEY (fact_id, entity_id)
+);
+
+CREATE INDEX IF NOT EXISTS idx_facts_trust    ON facts(trust_score DESC);
+CREATE INDEX IF NOT EXISTS idx_facts_category ON facts(category);
+CREATE INDEX IF NOT EXISTS idx_entities_name  ON entities(name);
+
+CREATE VIRTUAL TABLE IF NOT EXISTS facts_fts
+    USING fts5(content, tags, content=facts, content_rowid=fact_id);
+
+CREATE TRIGGER IF NOT EXISTS facts_ai AFTER INSERT ON facts BEGIN
+    INSERT INTO facts_fts(rowid, content, tags)
+        VALUES (new.fact_id, new.content, new.tags);
+END;
+
+CREATE TRIGGER IF NOT EXISTS facts_ad AFTER DELETE ON facts BEGIN
+    INSERT INTO facts_fts(facts_fts, rowid, content, tags)
+        VALUES ('delete', old.fact_id, old.content, old.tags);
+END;
+
+CREATE TRIGGER IF NOT EXISTS facts_au AFTER UPDATE ON facts BEGIN
+    INSERT INTO facts_fts(facts_fts, rowid, content, tags)
+        VALUES ('delete', old.fact_id, old.content, old.tags);
+    INSERT INTO facts_fts(rowid, content, tags)
+        VALUES (new.fact_id, new.content, new.tags);
+END;
+
+CREATE TABLE IF NOT EXISTS memory_banks (
+    bank_id    INTEGER PRIMARY KEY AUTOINCREMENT,
+    bank_name  TEXT NOT NULL UNIQUE,
+    vector     BLOB NOT NULL,
+    dim        INTEGER NOT NULL,
+    fact_count INTEGER DEFAULT 0,
+    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+"""
+
+# Trust adjustment constants
+_HELPFUL_DELTA   =  0.05
+_UNHELPFUL_DELTA = -0.10
+_TRUST_MIN       =  0.0
+_TRUST_MAX       =  1.0
+
+# Entity extraction patterns
+_RE_CAPITALIZED  = re.compile(r'\b([A-Z][a-z]+(?:\s+[A-Z][a-z]+)+)\b')
+_RE_DOUBLE_QUOTE = re.compile(r'"([^"]+)"')
+_RE_SINGLE_QUOTE = re.compile(r"'([^']+)'")
+_RE_AKA          = re.compile(
+    r'(\w+(?:\s+\w+)*)\s+(?:aka|also known as)\s+(\w+(?:\s+\w+)*)',
+    re.IGNORECASE,
+)
+
+
+def _clamp_trust(value: float) -> float:
+    return max(_TRUST_MIN, min(_TRUST_MAX, value))
+
+
+class MemoryStore:
+    """SQLite-backed fact store with entity resolution and trust scoring."""
+
+    def __init__(
+        self,
+        db_path: "str | Path | None" = None,
+        default_trust: float = 0.5,
+        hrr_dim: int = 1024,
+    ) -> None:
+        if db_path is None:
+            from hermes_constants import get_hermes_home
+            db_path = str(get_hermes_home() / "memory_store.db")
+        self.db_path = Path(db_path).expanduser()
+        self.db_path.parent.mkdir(parents=True, exist_ok=True)
+        self.default_trust = _clamp_trust(default_trust)
+        self.hrr_dim = hrr_dim
+        self._hrr_available = hrr._HAS_NUMPY
+        self._conn: sqlite3.Connection = sqlite3.connect(
+            str(self.db_path),
+            check_same_thread=False,
+            timeout=10.0,
+        )
+        self._lock = threading.RLock()
+        self._conn.row_factory = sqlite3.Row
+        self._init_db()
+
+    # ------------------------------------------------------------------
+    # Initialisation
+    # ------------------------------------------------------------------
+
+    def _init_db(self) -> None:
+        """Create tables, indexes, and triggers if they do not exist. Enable WAL mode."""
+        self._conn.execute("PRAGMA journal_mode=WAL")
+        self._conn.executescript(_SCHEMA)
+        # Migrate: add hrr_vector column if missing (safe for existing databases)
+        columns = {row[1] for row in self._conn.execute("PRAGMA table_info(facts)").fetchall()}
+        if "hrr_vector" not in columns:
+            self._conn.execute("ALTER TABLE facts ADD COLUMN hrr_vector BLOB")
+        self._conn.commit()
+
+    # ------------------------------------------------------------------
+    # Public API
+    # ------------------------------------------------------------------
+
+    def add_fact(
+        self,
+        content: str,
+        category: str = "general",
+        tags: str = "",
+    ) -> int:
+        """Insert a fact and return its fact_id.
+
+        Deduplicates by content (UNIQUE constraint). On duplicate, returns
+        the existing fact_id without modifying the row. Extracts entities from
+        the content and links them to the fact.
+        """
+        with self._lock:
+            content = content.strip()
+            if not content:
+                raise ValueError("content must not be empty")
+
+            try:
+                cur = self._conn.execute(
+                    """
+                    INSERT INTO facts (content, category, tags, trust_score)
+                    VALUES (?, ?, ?, ?)
+                    """,
+                    (content, category, tags, self.default_trust),
+                )
+                self._conn.commit()
+                fact_id: int = cur.lastrowid  # type: ignore[assignment]
+            except sqlite3.IntegrityError:
+                # Duplicate content — return existing id
+                row = self._conn.execute(
+                    "SELECT fact_id FROM facts WHERE content = ?", (content,)
+                ).fetchone()
+                return int(row["fact_id"])
+
+            # Entity extraction and linking
+            for name in self._extract_entities(content):
+                entity_id = self._resolve_entity(name)
+                self._link_fact_entity(fact_id, entity_id)
+
+            # Compute HRR vector after entity linking
+            self._compute_hrr_vector(fact_id, content)
+            self._rebuild_bank(category)
+
+            return fact_id
+
+    def search_facts(
+        self,
+        query: str,
+        category: str | None = None,
+        min_trust: float = 0.3,
+        limit: int = 10,
+    ) -> list[dict]:
+        """Full-text search over facts using FTS5.
+
+        Returns a list of fact dicts ordered by FTS5 rank, then trust_score
+        descending. Also increments retrieval_count for matched facts.
+        """
+        with self._lock:
+            query = query.strip()
+            if not query:
+                return []
+
+            params: list = [query, min_trust]
+            category_clause = ""
+            if category is not None:
+                category_clause = "AND f.category = ?"
+                params.append(category)
+            params.append(limit)
+
+            sql = f"""
+                SELECT f.fact_id, f.content, f.category, f.tags,
+                       f.trust_score, f.retrieval_count, f.helpful_count,
+                       f.created_at, f.updated_at
+                FROM facts f
+                JOIN facts_fts fts ON fts.rowid = f.fact_id
+                WHERE facts_fts MATCH ?
+                  AND f.trust_score >= ?
+                  {category_clause}
+                ORDER BY fts.rank, f.trust_score DESC
+                LIMIT ?
+            """
+
+            rows = self._conn.execute(sql, params).fetchall()
+            results = [self._row_to_dict(r) for r in rows]
+
+            if results:
+                ids = [r["fact_id"] for r in results]
+                placeholders = ",".join("?" * len(ids))
+                self._conn.execute(
+                    f"UPDATE facts SET retrieval_count = retrieval_count + 1 WHERE fact_id IN ({placeholders})",
+                    ids,
+                )
+                self._conn.commit()
+
+            return results
+
+    def update_fact(
+        self,
+        fact_id: int,
+        content: str | None = None,
+        trust_delta: float | None = None,
+        tags: str | None = None,
+        category: str | None = None,
+    ) -> bool:
+        """Partially update a fact. Trust is clamped to [0, 1].
+
+        Returns True if the row existed, False otherwise.
+        """
+        with self._lock:
+            row = self._conn.execute(
+                "SELECT fact_id, trust_score FROM facts WHERE fact_id = ?", (fact_id,)
+            ).fetchone()
+            if row is None:
+                return False
+
+            assignments: list[str] = ["updated_at = CURRENT_TIMESTAMP"]
+            params: list = []
+
+            if content is not None:
+                assignments.append("content = ?")
+                params.append(content.strip())
+            if tags is not None:
+                assignments.append("tags = ?")
+                params.append(tags)
+            if category is not None:
+                assignments.append("category = ?")
+                params.append(category)
+            if trust_delta is not None:
+                new_trust = _clamp_trust(row["trust_score"] + trust_delta)
+                assignments.append("trust_score = ?")
+                params.append(new_trust)
+
+            params.append(fact_id)
+            self._conn.execute(
+                f"UPDATE facts SET {', '.join(assignments)} WHERE fact_id = ?",
+                params,
+            )
+            self._conn.commit()
+
+            # If content changed, re-extract entities
+            if content is not None:
+                self._conn.execute(
+                    "DELETE FROM fact_entities WHERE fact_id = ?", (fact_id,)
+                )
+                for name in self._extract_entities(content):
+                    entity_id = self._resolve_entity(name)
+                    self._link_fact_entity(fact_id, entity_id)
+                self._conn.commit()
+
+            # Recompute HRR vector if content changed
+            if content is not None:
+                self._compute_hrr_vector(fact_id, content)
+            # Rebuild bank for relevant category
+            cat = category or self._conn.execute(
+                "SELECT category FROM facts WHERE fact_id = ?", (fact_id,)
+            ).fetchone()["category"]
+            self._rebuild_bank(cat)
+
+            return True
+
+    def remove_fact(self, fact_id: int) -> bool:
+        """Delete a fact and its entity links. Returns True if the row existed."""
+        with self._lock:
+            row = self._conn.execute(
+                "SELECT fact_id, category FROM facts WHERE fact_id = ?", (fact_id,)
+            ).fetchone()
+            if row is None:
+                return False
+
+            self._conn.execute(
+                "DELETE FROM fact_entities WHERE fact_id = ?", (fact_id,)
+            )
+            self._conn.execute("DELETE FROM facts WHERE fact_id = ?", (fact_id,))
+            self._conn.commit()
+            self._rebuild_bank(row["category"])
+            return True
+
+    def list_facts(
+        self,
+        category: str | None = None,
+        min_trust: float = 0.0,
+        limit: int = 50,
+    ) -> list[dict]:
+        """Browse facts ordered by trust_score descending.
+
+        Optionally filter by category and minimum trust score.
+        """
+        with self._lock:
+            params: list = [min_trust]
+            category_clause = ""
+            if category is not None:
+                category_clause = "AND category = ?"
+                params.append(category)
+            params.append(limit)
+
+            sql = f"""
+                SELECT fact_id, content, category, tags, trust_score,
+                       retrieval_count, helpful_count, created_at, updated_at
+                FROM facts
+                WHERE trust_score >= ?
+                  {category_clause}
+                ORDER BY trust_score DESC
+                LIMIT ?
+            """
+            rows = self._conn.execute(sql, params).fetchall()
+            return [self._row_to_dict(r) for r in rows]
+
+    def record_feedback(self, fact_id: int, helpful: bool) -> dict:
+        """Record user feedback and adjust trust asymmetrically.
+
+        helpful=True  -> trust += 0.05, helpful_count += 1
+        helpful=False -> trust -= 0.10
+
+        Returns a dict with fact_id, old_trust, new_trust, helpful_count.
+        Raises KeyError if fact_id does not exist.
+        """
+        with self._lock:
+            row = self._conn.execute(
+                "SELECT fact_id, trust_score, helpful_count FROM facts WHERE fact_id = ?",
+                (fact_id,),
+            ).fetchone()
+            if row is None:
+                raise KeyError(f"fact_id {fact_id} not found")
+
+            old_trust: float = row["trust_score"]
+            delta = _HELPFUL_DELTA if helpful else _UNHELPFUL_DELTA
+            new_trust = _clamp_trust(old_trust + delta)
+
+            helpful_increment = 1 if helpful else 0
+            self._conn.execute(
+                """
+                UPDATE facts
+                SET trust_score    = ?,
+                    helpful_count  = helpful_count + ?,
+                    updated_at     = CURRENT_TIMESTAMP
+                WHERE fact_id = ?
+                """,
+                (new_trust, helpful_increment, fact_id),
+            )
+            self._conn.commit()
+
+            return {
+                "fact_id":      fact_id,
+                "old_trust":    old_trust,
+                "new_trust":    new_trust,
+                "helpful_count": row["helpful_count"] + helpful_increment,
+            }
+
+    # ------------------------------------------------------------------
+    # Entity helpers
+    # ------------------------------------------------------------------
+
+    def _extract_entities(self, text: str) -> list[str]:
+        """Extract entity candidates from text using simple regex rules.
+
+        Rules applied (in order):
+        1. Capitalized multi-word phrases  e.g. "John Doe"
+        2. Double-quoted terms             e.g. "Python"
+        3. Single-quoted terms             e.g. 'pytest'
+        4. AKA patterns                    e.g. "Guido aka BDFL" -> two entities
+
+        Returns a deduplicated list preserving first-seen order.
+        """
+        seen: set[str] = set()
+        candidates: list[str] = []
+
+        def _add(name: str) -> None:
+            stripped = name.strip()
+            if stripped and stripped.lower() not in seen:
+                seen.add(stripped.lower())
+                candidates.append(stripped)
+
+        for m in _RE_CAPITALIZED.finditer(text):
+            _add(m.group(1))
+
+        for m in _RE_DOUBLE_QUOTE.finditer(text):
+            _add(m.group(1))
+
+        for m in _RE_SINGLE_QUOTE.finditer(text):
+            _add(m.group(1))
+
+        for m in _RE_AKA.finditer(text):
+            _add(m.group(1))
+            _add(m.group(2))
+
+        return candidates
+
+    def _resolve_entity(self, name: str) -> int:
+        """Find an existing entity by name or alias (case-insensitive) or create one.
+
+        Returns the entity_id.
+        """
+        # Exact name match
+        row = self._conn.execute(
+            "SELECT entity_id FROM entities WHERE name LIKE ?", (name,)
+        ).fetchone()
+        if row is not None:
+            return int(row["entity_id"])
+
+        # Search aliases — aliases stored as comma-separated; use LIKE with % boundaries
+        alias_row = self._conn.execute(
+            """
+            SELECT entity_id FROM entities
+            WHERE ',' || aliases || ',' LIKE '%,' || ? || ',%'
+            """,
+            (name,),
+        ).fetchone()
+        if alias_row is not None:
+            return int(alias_row["entity_id"])
+
+        # Create new entity
+        cur = self._conn.execute(
+            "INSERT INTO entities (name) VALUES (?)", (name,)
+        )
+        self._conn.commit()
+        return int(cur.lastrowid)  # type: ignore[return-value]
+
+    def _link_fact_entity(self, fact_id: int, entity_id: int) -> None:
+        """Insert into fact_entities, silently ignore if the link already exists."""
+        self._conn.execute(
+            """
+            INSERT OR IGNORE INTO fact_entities (fact_id, entity_id)
+            VALUES (?, ?)
+            """,
+            (fact_id, entity_id),
+        )
+        self._conn.commit()
+
+    def _compute_hrr_vector(self, fact_id: int, content: str) -> None:
+        """Compute and store HRR vector for a fact. No-op if numpy unavailable."""
+        with self._lock:
+            if not self._hrr_available:
+                return
+
+            # Get entities linked to this fact
+            rows = self._conn.execute(
+                """
+                SELECT e.name FROM entities e
+                JOIN fact_entities fe ON fe.entity_id = e.entity_id
+                WHERE fe.fact_id = ?
+                """,
+                (fact_id,),
+            ).fetchall()
+            entities = [row["name"] for row in rows]
+
+            vector = hrr.encode_fact(content, entities, self.hrr_dim)
+            self._conn.execute(
+                "UPDATE facts SET hrr_vector = ? WHERE fact_id = ?",
+                (hrr.phases_to_bytes(vector), fact_id),
+            )
+            self._conn.commit()
+
+    def _rebuild_bank(self, category: str) -> None:
+        """Full rebuild of a category's memory bank from all its fact vectors."""
+        with self._lock:
+            if not self._hrr_available:
+                return
+
+            bank_name = f"cat:{category}"
+            rows = self._conn.execute(
+                "SELECT hrr_vector FROM facts WHERE category = ? AND hrr_vector IS NOT NULL",
+                (category,),
+            ).fetchall()
+
+            if not rows:
+                self._conn.execute("DELETE FROM memory_banks WHERE bank_name = ?", (bank_name,))
+                self._conn.commit()
+                return
+
+            vectors = [hrr.bytes_to_phases(row["hrr_vector"]) for row in rows]
+            bank_vector = hrr.bundle(*vectors)
+            fact_count = len(vectors)
+
+            # Check SNR
+            hrr.snr_estimate(self.hrr_dim, fact_count)
+
+            self._conn.execute(
+                """
+                INSERT INTO memory_banks (bank_name, vector, dim, fact_count, updated_at)
+                VALUES (?, ?, ?, ?, CURRENT_TIMESTAMP)
+                ON CONFLICT(bank_name) DO UPDATE SET
+                    vector = excluded.vector,
+                    dim = excluded.dim,
+                    fact_count = excluded.fact_count,
+                    updated_at = excluded.updated_at
+                """,
+                (bank_name, hrr.phases_to_bytes(bank_vector), self.hrr_dim, fact_count),
+            )
+            self._conn.commit()
+
+    def rebuild_all_vectors(self, dim: int | None = None) -> int:
+        """Recompute all HRR vectors + banks from text. For recovery/migration.
+
+        Returns the number of facts processed.
+        """
+        with self._lock:
+            if not self._hrr_available:
+                return 0
+
+            if dim is not None:
+                self.hrr_dim = dim
+
+            rows = self._conn.execute(
+                "SELECT fact_id, content, category FROM facts"
+            ).fetchall()
+
+            categories: set[str] = set()
+            for row in rows:
+                self._compute_hrr_vector(row["fact_id"], row["content"])
+                categories.add(row["category"])
+
+            for category in categories:
+                self._rebuild_bank(category)
+
+            return len(rows)
+
+    # ------------------------------------------------------------------
+    # Utilities
+    # ------------------------------------------------------------------
+
+    def _row_to_dict(self, row: sqlite3.Row) -> dict:
+        """Convert a sqlite3.Row to a plain dict."""
+        return dict(row)
+
+    def close(self) -> None:
+        """Close the database connection."""
+        self._conn.close()
+
+    def __enter__(self) -> "MemoryStore":
+        return self
+
+    def __exit__(self, *_: object) -> None:
+        self.close()
@@ -0,0 +1,35 @@
+# Honcho Memory Provider
+
+AI-native cross-session user modeling with dialectic Q&A, semantic search, peer cards, and persistent conclusions.
+
+## Requirements
+
+- `pip install honcho-ai`
+- Honcho API key from [app.honcho.dev](https://app.honcho.dev)
+
+## Setup
+
+```bash
+hermes memory setup    # select "honcho"
+```
+
+Or manually:
+```bash
+hermes config set memory.provider honcho
+echo "HONCHO_API_KEY=your-key" >> ~/.hermes/.env
+```
+
+## Config
+
+Config file: `$HERMES_HOME/honcho.json` (or `~/.honcho/config.json` legacy)
+
+Existing Honcho users: your config and data are preserved. Just set `memory.provider: honcho`.
+
+## Tools
+
+| Tool | Description |
+|------|-------------|
+| `honcho_profile` | User's peer card — key facts, no LLM |
+| `honcho_search` | Semantic search over stored context |
+| `honcho_context` | LLM-synthesized answer from memory |
+| `honcho_conclude` | Write a fact about the user to memory |
@@ -0,0 +1,692 @@
+"""Honcho memory plugin — MemoryProvider for Honcho AI-native memory.
+
+Provides cross-session user modeling with dialectic Q&A, semantic search,
+peer cards, and persistent conclusions via the Honcho SDK. Honcho provides AI-native cross-session user
+modeling with dialectic Q&A, semantic search, peer cards, and conclusions.
+
+The 4 tools (profile, search, context, conclude) are exposed through
+the MemoryProvider interface.
+
+Config: Uses the existing Honcho config chain:
+  1. $HERMES_HOME/honcho.json (profile-scoped)
+  2. ~/.honcho/config.json (legacy global)
+  3. Environment variables
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import threading
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas (moved from tools/honcho_tools.py)
+# ---------------------------------------------------------------------------
+
+PROFILE_SCHEMA = {
+    "name": "honcho_profile",
+    "description": (
+        "Retrieve the user's peer card from Honcho — a curated list of key facts "
+        "about them (name, role, preferences, communication style, patterns). "
+        "Fast, no LLM reasoning, minimal cost. "
+        "Use this at conversation start or when you need a quick factual snapshot."
+    ),
+    "parameters": {"type": "object", "properties": {}, "required": []},
+}
+
+SEARCH_SCHEMA = {
+    "name": "honcho_search",
+    "description": (
+        "Semantic search over Honcho's stored context about the user. "
+        "Returns raw excerpts ranked by relevance — no LLM synthesis. "
+        "Cheaper and faster than honcho_context. "
+        "Good when you want to find specific past facts and reason over them yourself."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {
+                "type": "string",
+                "description": "What to search for in Honcho's memory.",
+            },
+            "max_tokens": {
+                "type": "integer",
+                "description": "Token budget for returned context (default 800, max 2000).",
+            },
+        },
+        "required": ["query"],
+    },
+}
+
+CONTEXT_SCHEMA = {
+    "name": "honcho_context",
+    "description": (
+        "Ask Honcho a natural language question and get a synthesized answer. "
+        "Uses Honcho's LLM (dialectic reasoning) — higher cost than honcho_profile or honcho_search. "
+        "Can query about any peer: the user (default) or the AI assistant."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {
+                "type": "string",
+                "description": "A natural language question.",
+            },
+            "peer": {
+                "type": "string",
+                "description": "Which peer to query about: 'user' (default) or 'ai'.",
+            },
+        },
+        "required": ["query"],
+    },
+}
+
+CONCLUDE_SCHEMA = {
+    "name": "honcho_conclude",
+    "description": (
+        "Write a conclusion about the user back to Honcho's memory. "
+        "Conclusions are persistent facts that build the user's profile. "
+        "Use when the user states a preference, corrects you, or shares "
+        "something to remember across sessions."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "conclusion": {
+                "type": "string",
+                "description": "A factual statement about the user to persist.",
+            }
+        },
+        "required": ["conclusion"],
+    },
+}
+
+
+ALL_TOOL_SCHEMAS = [PROFILE_SCHEMA, SEARCH_SCHEMA, CONTEXT_SCHEMA, CONCLUDE_SCHEMA]
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class HonchoMemoryProvider(MemoryProvider):
+    """Honcho AI-native memory with dialectic Q&A and persistent user modeling."""
+
+    def __init__(self):
+        self._manager = None   # HonchoSessionManager
+        self._config = None    # HonchoClientConfig
+        self._session_key = ""
+        self._prefetch_result = ""
+        self._prefetch_lock = threading.Lock()
+        self._prefetch_thread: Optional[threading.Thread] = None
+        self._sync_thread: Optional[threading.Thread] = None
+
+        # B1: recall_mode — set during initialize from config
+        self._recall_mode = "hybrid"  # "context", "tools", or "hybrid"
+
+        # B4: First-turn context baking
+        self._first_turn_context: Optional[str] = None
+        self._first_turn_lock = threading.Lock()
+
+        # B5: Cost-awareness turn counting and cadence
+        self._turn_count = 0
+        self._injection_frequency = "every-turn"  # or "first-turn"
+        self._context_cadence = 1   # minimum turns between context API calls
+        self._dialectic_cadence = 1  # minimum turns between dialectic API calls
+        self._reasoning_level_cap: Optional[str] = None  # "minimal", "low", "mid", "high"
+        self._last_context_turn = -999
+        self._last_dialectic_turn = -999
+
+        # B2: peer_memory_mode gating (stub)
+        self._suppress_memory = False
+        self._suppress_user_profile = False
+
+        # Port #1957: lazy session init for tools-only mode
+        self._session_initialized = False
+        self._lazy_init_kwargs: Optional[dict] = None
+        self._lazy_init_session_id: Optional[str] = None
+
+        # Port #4053: cron guard — when True, plugin is fully inactive
+        self._cron_skipped = False
+
+    @property
+    def name(self) -> str:
+        return "honcho"
+
+    def is_available(self) -> bool:
+        """Check if Honcho is configured. No network calls."""
+        try:
+            from plugins.memory.honcho.client import HonchoClientConfig
+            cfg = HonchoClientConfig.from_global_config()
+            # Port #2645: baseUrl-only verification — api_key OR base_url suffices
+            return cfg.enabled and bool(cfg.api_key or cfg.base_url)
+        except Exception:
+            return False
+
+    def save_config(self, values, hermes_home):
+        """Write config to $HERMES_HOME/honcho.json (Honcho SDK native format)."""
+        import json
+        from pathlib import Path
+        config_path = Path(hermes_home) / "honcho.json"
+        existing = {}
+        if config_path.exists():
+            try:
+                existing = json.loads(config_path.read_text())
+            except Exception:
+                pass
+        existing.update(values)
+        config_path.write_text(json.dumps(existing, indent=2))
+
+    def get_config_schema(self):
+        return [
+            {"key": "api_key", "description": "Honcho API key", "secret": True, "env_var": "HONCHO_API_KEY", "url": "https://app.honcho.dev"},
+            {"key": "base_url", "description": "Honcho base URL", "default": "https://api.honcho.dev"},
+        ]
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        """Initialize Honcho session manager.
+
+        Handles: cron guard, recall_mode, session name resolution,
+        peer memory mode, SOUL.md ai_peer sync, memory file migration,
+        and pre-warming context at init.
+        """
+        try:
+            # ----- Port #4053: cron guard -----
+            agent_context = kwargs.get("agent_context", "")
+            platform = kwargs.get("platform", "cli")
+            if agent_context in ("cron", "flush") or platform == "cron":
+                logger.debug("Honcho skipped: cron/flush context (agent_context=%s, platform=%s)",
+                             agent_context, platform)
+                self._cron_skipped = True
+                return
+
+            from plugins.memory.honcho.client import HonchoClientConfig, get_honcho_client
+            from plugins.memory.honcho.session import HonchoSessionManager
+
+            cfg = HonchoClientConfig.from_global_config()
+            if not cfg.enabled or not (cfg.api_key or cfg.base_url):
+                logger.debug("Honcho not configured — plugin inactive")
+                return
+
+            self._config = cfg
+
+            # ----- B1: recall_mode from config -----
+            self._recall_mode = cfg.recall_mode  # "context", "tools", or "hybrid"
+            logger.debug("Honcho recall_mode: %s", self._recall_mode)
+
+            # ----- B5: cost-awareness config -----
+            try:
+                raw = cfg.raw or {}
+                self._injection_frequency = raw.get("injectionFrequency", "every-turn")
+                self._context_cadence = int(raw.get("contextCadence", 1))
+                self._dialectic_cadence = int(raw.get("dialecticCadence", 1))
+                cap = raw.get("reasoningLevelCap")
+                if cap and cap in ("minimal", "low", "mid", "high"):
+                    self._reasoning_level_cap = cap
+            except Exception as e:
+                logger.debug("Honcho cost-awareness config parse error: %s", e)
+
+            # ----- Port #1969: aiPeer sync from SOUL.md -----
+            try:
+                hermes_home = kwargs.get("hermes_home", "")
+                if hermes_home and not cfg.raw.get("aiPeer"):
+                    soul_path = Path(hermes_home) / "SOUL.md"
+                    if soul_path.exists():
+                        soul_text = soul_path.read_text(encoding="utf-8").strip()
+                        if soul_text:
+                            # Try YAML frontmatter: "name: Foo"
+                            first_line = soul_text.split("\n")[0].strip()
+                            if first_line.startswith("---"):
+                                # Look for name: in frontmatter
+                                for line in soul_text.split("\n")[1:]:
+                                    line = line.strip()
+                                    if line == "---":
+                                        break
+                                    if line.lower().startswith("name:"):
+                                        name_val = line.split(":", 1)[1].strip().strip("\"'")
+                                        if name_val:
+                                            cfg.ai_peer = name_val
+                                            logger.debug("Honcho ai_peer set from SOUL.md: %s", name_val)
+                                        break
+                            elif first_line.startswith("# "):
+                                # Markdown heading: "# AgentName"
+                                name_val = first_line[2:].strip()
+                                if name_val:
+                                    cfg.ai_peer = name_val
+                                    logger.debug("Honcho ai_peer set from SOUL.md heading: %s", name_val)
+            except Exception as e:
+                logger.debug("Honcho SOUL.md ai_peer sync failed: %s", e)
+
+            # ----- B2: peer_memory_mode gating (stub) -----
+            try:
+                ai_mode = cfg.peer_memory_mode(cfg.ai_peer)
+                user_mode = cfg.peer_memory_mode(cfg.peer_name or "user")
+                # "honcho" means Honcho owns memory; suppress built-in
+                self._suppress_memory = (ai_mode == "honcho")
+                self._suppress_user_profile = (user_mode == "honcho")
+                logger.debug("Honcho peer_memory_mode: ai=%s (suppress_memory=%s), user=%s (suppress_user_profile=%s)",
+                             ai_mode, self._suppress_memory, user_mode, self._suppress_user_profile)
+            except Exception as e:
+                logger.debug("Honcho peer_memory_mode check failed: %s", e)
+
+            # ----- Port #1957: lazy session init for tools-only mode -----
+            if self._recall_mode == "tools":
+                # Defer actual session creation until first tool call
+                self._lazy_init_kwargs = kwargs
+                self._lazy_init_session_id = session_id
+                # Still need a client reference for _ensure_session
+                self._config = cfg
+                logger.debug("Honcho tools-only mode — deferring session init until first tool call")
+                return
+
+            # ----- Eager init (context or hybrid mode) -----
+            self._do_session_init(cfg, session_id, **kwargs)
+
+        except ImportError:
+            logger.debug("honcho-ai package not installed — plugin inactive")
+        except Exception as e:
+            logger.warning("Honcho init failed: %s", e)
+            self._manager = None
+
+    def _do_session_init(self, cfg, session_id: str, **kwargs) -> None:
+        """Shared session initialization logic for both eager and lazy paths."""
+        from plugins.memory.honcho.client import get_honcho_client
+        from plugins.memory.honcho.session import HonchoSessionManager
+
+        client = get_honcho_client(cfg)
+        self._manager = HonchoSessionManager(
+            honcho=client,
+            config=cfg,
+            context_tokens=cfg.context_tokens,
+        )
+
+        # ----- B3: resolve_session_name -----
+        session_title = kwargs.get("session_title")
+        self._session_key = (
+            cfg.resolve_session_name(session_title=session_title, session_id=session_id)
+            or session_id
+            or "hermes-default"
+        )
+        logger.debug("Honcho session key resolved: %s", self._session_key)
+
+        # Create session eagerly
+        session = self._manager.get_or_create(self._session_key)
+        self._session_initialized = True
+
+        # ----- B6: Memory file migration (one-time, for new sessions) -----
+        try:
+            if not session.messages:
+                from hermes_constants import get_hermes_home
+                mem_dir = str(get_hermes_home() / "memories")
+                self._manager.migrate_memory_files(self._session_key, mem_dir)
+                logger.debug("Honcho memory file migration attempted for new session: %s", self._session_key)
+        except Exception as e:
+            logger.debug("Honcho memory file migration skipped: %s", e)
+
+        # ----- B7: Pre-warming context at init -----
+        if self._recall_mode in ("context", "hybrid"):
+            try:
+                self._manager.prefetch_context(self._session_key)
+                self._manager.prefetch_dialectic(self._session_key, "What should I know about this user?")
+                logger.debug("Honcho pre-warm threads started for session: %s", self._session_key)
+            except Exception as e:
+                logger.debug("Honcho pre-warm failed: %s", e)
+
+    def _ensure_session(self) -> bool:
+        """Lazily initialize the Honcho session (for tools-only mode).
+
+        Returns True if the manager is ready, False otherwise.
+        """
+        if self._manager and self._session_initialized:
+            return True
+        if self._cron_skipped:
+            return False
+        if not self._config or not self._lazy_init_kwargs:
+            return False
+
+        try:
+            self._do_session_init(
+                self._config,
+                self._lazy_init_session_id or "hermes-default",
+                **self._lazy_init_kwargs,
+            )
+            # Clear lazy refs
+            self._lazy_init_kwargs = None
+            self._lazy_init_session_id = None
+            return self._manager is not None
+        except Exception as e:
+            logger.warning("Honcho lazy session init failed: %s", e)
+            return False
+
+    def _format_first_turn_context(self, ctx: dict) -> str:
+        """Format the prefetch context dict into a readable system prompt block."""
+        parts = []
+
+        rep = ctx.get("representation", "")
+        if rep:
+            parts.append(f"## User Representation\n{rep}")
+
+        card = ctx.get("card", "")
+        if card:
+            parts.append(f"## User Peer Card\n{card}")
+
+        ai_rep = ctx.get("ai_representation", "")
+        if ai_rep:
+            parts.append(f"## AI Self-Representation\n{ai_rep}")
+
+        ai_card = ctx.get("ai_card", "")
+        if ai_card:
+            parts.append(f"## AI Identity Card\n{ai_card}")
+
+        if not parts:
+            return ""
+        return "\n\n".join(parts)
+
+    def system_prompt_block(self) -> str:
+        """Return system prompt text, adapted by recall_mode.
+
+        B4: On the FIRST call, fetch and bake the full Honcho context
+        (user representation, peer card, AI representation, continuity synthesis).
+        Subsequent calls return the cached block for prompt caching stability.
+        """
+        if self._cron_skipped:
+            return ""
+        if not self._manager or not self._session_key:
+            # tools-only mode without session yet still returns a minimal block
+            if self._recall_mode == "tools" and self._config:
+                return (
+                    "# Honcho Memory\n"
+                    "Active (tools-only mode). Use honcho_profile, honcho_search, "
+                    "honcho_context, and honcho_conclude tools to access user memory."
+                )
+            return ""
+
+        # ----- B4: First-turn context baking -----
+        first_turn_block = ""
+        if self._recall_mode in ("context", "hybrid"):
+            with self._first_turn_lock:
+                if self._first_turn_context is None:
+                    # First call — fetch and cache
+                    try:
+                        ctx = self._manager.get_prefetch_context(self._session_key)
+                        self._first_turn_context = self._format_first_turn_context(ctx) if ctx else ""
+                    except Exception as e:
+                        logger.debug("Honcho first-turn context fetch failed: %s", e)
+                        self._first_turn_context = ""
+                first_turn_block = self._first_turn_context
+
+        # ----- B1: adapt text based on recall_mode -----
+        if self._recall_mode == "context":
+            header = (
+                "# Honcho Memory\n"
+                "Active (context-injection mode). Relevant user context is automatically "
+                "injected before each turn. No memory tools are available — context is "
+                "managed automatically."
+            )
+        elif self._recall_mode == "tools":
+            header = (
+                "# Honcho Memory\n"
+                "Active (tools-only mode). Use honcho_profile for a quick factual snapshot, "
+                "honcho_search for raw excerpts, honcho_context for synthesized answers, "
+                "honcho_conclude to save facts about the user. "
+                "No automatic context injection — you must use tools to access memory."
+            )
+        else:  # hybrid
+            header = (
+                "# Honcho Memory\n"
+                "Active (hybrid mode). Relevant context is auto-injected AND memory tools are available. "
+                "Use honcho_profile for a quick factual snapshot, "
+                "honcho_search for raw excerpts, honcho_context for synthesized answers, "
+                "honcho_conclude to save facts about the user."
+            )
+
+        if first_turn_block:
+            return f"{header}\n\n{first_turn_block}"
+        return header
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        """Return prefetched dialectic context from background thread.
+
+        B1: Returns empty when recall_mode is "tools" (no injection).
+        B5: Respects injection_frequency — "first-turn" returns cached/empty after turn 0.
+        Port #3265: Truncates to context_tokens budget.
+        """
+        if self._cron_skipped:
+            return ""
+
+        # B1: tools-only mode — no auto-injection
+        if self._recall_mode == "tools":
+            return ""
+
+        # B5: injection_frequency — if "first-turn" and past first turn, return empty
+        if self._injection_frequency == "first-turn" and self._turn_count > 0:
+            return ""
+
+        if self._prefetch_thread and self._prefetch_thread.is_alive():
+            self._prefetch_thread.join(timeout=3.0)
+        with self._prefetch_lock:
+            result = self._prefetch_result
+            self._prefetch_result = ""
+        if not result:
+            return ""
+
+        # ----- Port #3265: token budget enforcement -----
+        result = self._truncate_to_budget(result)
+
+        return f"## Honcho Context\n{result}"
+
+    def _truncate_to_budget(self, text: str) -> str:
+        """Truncate text to fit within context_tokens budget if set."""
+        if not self._config or not self._config.context_tokens:
+            return text
+        budget_chars = self._config.context_tokens * 4  # conservative char estimate
+        if len(text) <= budget_chars:
+            return text
+        # Truncate at word boundary
+        truncated = text[:budget_chars]
+        last_space = truncated.rfind(" ")
+        if last_space > budget_chars * 0.8:
+            truncated = truncated[:last_space]
+        return truncated + " …"
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        """Fire a background dialectic query for the upcoming turn.
+
+        B5: Checks cadence before firing background threads.
+        """
+        if self._cron_skipped:
+            return
+        if not self._manager or not self._session_key or not query:
+            return
+
+        # B1: tools-only mode — no prefetch
+        if self._recall_mode == "tools":
+            return
+
+        # B5: cadence check — skip if too soon since last dialectic call
+        if self._dialectic_cadence > 1:
+            if (self._turn_count - self._last_dialectic_turn) < self._dialectic_cadence:
+                logger.debug("Honcho dialectic prefetch skipped: cadence %d, turns since last: %d",
+                             self._dialectic_cadence, self._turn_count - self._last_dialectic_turn)
+                return
+
+        self._last_dialectic_turn = self._turn_count
+
+        def _run():
+            try:
+                result = self._manager.dialectic_query(
+                    self._session_key, query, peer="user"
+                )
+                if result and result.strip():
+                    with self._prefetch_lock:
+                        self._prefetch_result = result
+            except Exception as e:
+                logger.debug("Honcho prefetch failed: %s", e)
+
+        self._prefetch_thread = threading.Thread(
+            target=_run, daemon=True, name="honcho-prefetch"
+        )
+        self._prefetch_thread.start()
+
+        # Also fire context prefetch if cadence allows
+        if self._context_cadence <= 1 or (self._turn_count - self._last_context_turn) >= self._context_cadence:
+            self._last_context_turn = self._turn_count
+            try:
+                self._manager.prefetch_context(self._session_key, query)
+            except Exception as e:
+                logger.debug("Honcho context prefetch failed: %s", e)
+
+    def on_turn_start(self, turn_number: int, message: str, **kwargs) -> None:
+        """Track turn count for cadence and injection_frequency logic."""
+        self._turn_count = turn_number
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Record the conversation turn in Honcho (non-blocking)."""
+        if self._cron_skipped:
+            return
+        if not self._manager or not self._session_key:
+            return
+
+        def _sync():
+            try:
+                session = self._manager.get_or_create(self._session_key)
+                session.add_message("user", user_content[:4000])
+                session.add_message("assistant", assistant_content[:4000])
+                # Flush to Honcho API
+                self._manager._flush_session(session)
+            except Exception as e:
+                logger.debug("Honcho sync_turn failed: %s", e)
+
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=5.0)
+        self._sync_thread = threading.Thread(
+            target=_sync, daemon=True, name="honcho-sync"
+        )
+        self._sync_thread.start()
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        """Mirror built-in user profile writes as Honcho conclusions."""
+        if action != "add" or target != "user" or not content:
+            return
+        if self._cron_skipped:
+            return
+        if not self._manager or not self._session_key:
+            return
+
+        def _write():
+            try:
+                self._manager.create_conclusion(self._session_key, content)
+            except Exception as e:
+                logger.debug("Honcho memory mirror failed: %s", e)
+
+        t = threading.Thread(target=_write, daemon=True, name="honcho-memwrite")
+        t.start()
+
+    def on_session_end(self, messages: List[Dict[str, Any]]) -> None:
+        """Flush all pending messages to Honcho on session end."""
+        if self._cron_skipped:
+            return
+        if not self._manager:
+            return
+        # Wait for pending sync
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=10.0)
+        try:
+            self._manager.flush_all()
+        except Exception as e:
+            logger.debug("Honcho session-end flush failed: %s", e)
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        """Return tool schemas, respecting recall_mode.
+
+        B1: context-only mode hides all tools.
+        """
+        if self._cron_skipped:
+            return []
+        if self._recall_mode == "context":
+            return []
+        return list(ALL_TOOL_SCHEMAS)
+
+    def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
+        """Handle a Honcho tool call, with lazy session init for tools-only mode."""
+        if self._cron_skipped:
+            return json.dumps({"error": "Honcho is not active (cron context)."})
+
+        # Port #1957: ensure session is initialized for tools-only mode
+        if not self._session_initialized:
+            if not self._ensure_session():
+                return json.dumps({"error": "Honcho session could not be initialized."})
+
+        if not self._manager or not self._session_key:
+            return json.dumps({"error": "Honcho is not active for this session."})
+
+        try:
+            if tool_name == "honcho_profile":
+                card = self._manager.get_peer_card(self._session_key)
+                if not card:
+                    return json.dumps({"result": "No profile facts available yet."})
+                return json.dumps({"result": card})
+
+            elif tool_name == "honcho_search":
+                query = args.get("query", "")
+                if not query:
+                    return json.dumps({"error": "Missing required parameter: query"})
+                max_tokens = min(int(args.get("max_tokens", 800)), 2000)
+                result = self._manager.search_context(
+                    self._session_key, query, max_tokens=max_tokens
+                )
+                if not result:
+                    return json.dumps({"result": "No relevant context found."})
+                return json.dumps({"result": result})
+
+            elif tool_name == "honcho_context":
+                query = args.get("query", "")
+                if not query:
+                    return json.dumps({"error": "Missing required parameter: query"})
+                peer = args.get("peer", "user")
+                result = self._manager.dialectic_query(
+                    self._session_key, query, peer=peer
+                )
+                return json.dumps({"result": result or "No result from Honcho."})
+
+            elif tool_name == "honcho_conclude":
+                conclusion = args.get("conclusion", "")
+                if not conclusion:
+                    return json.dumps({"error": "Missing required parameter: conclusion"})
+                ok = self._manager.create_conclusion(self._session_key, conclusion)
+                if ok:
+                    return json.dumps({"result": f"Conclusion saved: {conclusion}"})
+                return json.dumps({"error": "Failed to save conclusion."})
+
+            return json.dumps({"error": f"Unknown tool: {tool_name}"})
+
+        except Exception as e:
+            logger.error("Honcho tool %s failed: %s", tool_name, e)
+            return json.dumps({"error": f"Honcho {tool_name} failed: {e}"})
+
+    def shutdown(self) -> None:
+        for t in (self._prefetch_thread, self._sync_thread):
+            if t and t.is_alive():
+                t.join(timeout=5.0)
+        # Flush any remaining messages
+        if self._manager:
+            try:
+                self._manager.flush_all()
+            except Exception:
+                pass
+
+
+# ---------------------------------------------------------------------------
+# Plugin entry point
+# ---------------------------------------------------------------------------
+
+def register(ctx) -> None:
+    """Register Honcho as a memory provider plugin."""
+    ctx.register_memory_provider(HonchoMemoryProvider())
@@ -11,9 +11,228 @@ import sys
 from pathlib import Path

 from hermes_constants import get_hermes_home
-from honcho_integration.client import resolve_config_path, GLOBAL_CONFIG_PATH
+from plugins.memory.honcho.client import resolve_active_host, resolve_config_path, GLOBAL_CONFIG_PATH, HOST

-HOST = "hermes"
+
+def clone_honcho_for_profile(profile_name: str) -> bool:
+    """Auto-clone Honcho config for a new profile from the default host block.
+
+    Called during profile creation. If Honcho is configured on the default
+    host, creates a new host block for the profile with inherited settings
+    and auto-derived workspace/aiPeer.
+
+    Returns True if a host block was created, False if Honcho isn't configured.
+    """
+    cfg = _read_config()
+    if not cfg:
+        return False
+
+    hosts = cfg.get("hosts", {})
+    default_block = hosts.get(HOST, {})
+
+    # No default host block and no root-level API key = Honcho not configured
+    has_key = bool(cfg.get("apiKey") or os.environ.get("HONCHO_API_KEY"))
+    if not default_block and not has_key:
+        return False
+
+    new_host = f"{HOST}.{profile_name}"
+    if new_host in hosts:
+        return False  # already exists
+
+    # Clone settings from default block, override identity fields
+    new_block = {}
+    for key in ("memoryMode", "recallMode", "writeFrequency", "sessionStrategy",
+                "sessionPeerPrefix", "contextTokens", "dialecticReasoningLevel",
+                "dialecticMaxChars", "saveMessages"):
+        val = default_block.get(key)
+        if val is not None:
+            new_block[key] = val
+
+    # Inherit peer name from default
+    peer_name = default_block.get("peerName") or cfg.get("peerName")
+    if peer_name:
+        new_block["peerName"] = peer_name
+
+    # AI peer is profile-specific; workspace is shared so all profiles
+    # see the same user context, sessions, and project history.
+    # Use the bare profile name as the peer identity (not the host key)
+    # because Honcho's peer ID pattern is ^[a-zA-Z0-9_-]+$ (no dots).
+    new_block["aiPeer"] = profile_name
+    new_block["workspace"] = default_block.get("workspace") or cfg.get("workspace") or HOST
+    new_block["enabled"] = default_block.get("enabled", True)
+
+    cfg.setdefault("hosts", {})[new_host] = new_block
+    _write_config(cfg)
+
+    # Eagerly create the peer in Honcho so it exists before first message
+    _ensure_peer_exists(new_host)
+    return True
+
+
+def _ensure_peer_exists(host_key: str | None = None) -> bool:
+    """Create the AI peer in Honcho if it doesn't already exist.
+
+    Idempotent -- safe to call multiple times. Returns True if the peer
+    was created or already exists, False on failure.
+    """
+    try:
+        from plugins.memory.honcho.client import HonchoClientConfig, get_honcho_client
+        hcfg = HonchoClientConfig.from_global_config(host=host_key)
+        if not hcfg.enabled or not (hcfg.api_key or hcfg.base_url):
+            return False
+        client = get_honcho_client(hcfg)
+        # peer() is idempotent -- creates if missing, returns if exists
+        client.peer(hcfg.ai_peer)
+        if hcfg.peer_name:
+            client.peer(hcfg.peer_name)
+        return True
+    except Exception:
+        return False
+
+
+def cmd_enable(args) -> None:
+    """Enable Honcho for the active profile."""
+    cfg = _read_config()
+    host = _host_key()
+    label = f"[{host}] " if host != "hermes" else ""
+    block = cfg.setdefault("hosts", {}).setdefault(host, {})
+
+    if block.get("enabled") is True:
+        print(f"  {label}Honcho is already enabled.\n")
+        return
+
+    block["enabled"] = True
+
+    # If this is a new profile host block with no settings, clone from default
+    if not block.get("aiPeer"):
+        default_block = cfg.get("hosts", {}).get(HOST, {})
+        for key in ("memoryMode", "recallMode", "writeFrequency", "sessionStrategy",
+                    "contextTokens", "dialecticReasoningLevel", "dialecticMaxChars"):
+            val = default_block.get(key)
+            if val is not None and key not in block:
+                block[key] = val
+        peer_name = default_block.get("peerName") or cfg.get("peerName")
+        if peer_name and "peerName" not in block:
+            block["peerName"] = peer_name
+        # Use bare profile name as AI peer, not the host key
+        ai_peer = host.split(".", 1)[1] if "." in host else host
+        block.setdefault("aiPeer", ai_peer)
+        block.setdefault("workspace", default_block.get("workspace") or cfg.get("workspace") or HOST)
+
+    _write_config(cfg)
+    print(f"  {label}Honcho enabled.")
+
+    # Create peer eagerly
+    if _ensure_peer_exists(host):
+        print(f"  {label}Peer '{block.get('aiPeer', host)}' ready.")
+    else:
+        print(f"  {label}Peer creation deferred (no connection).")
+
+    print(f"  Saved to {_config_path()}\n")
+
+
+def cmd_disable(args) -> None:
+    """Disable Honcho for the active profile."""
+    cfg = _read_config()
+    host = _host_key()
+    label = f"[{host}] " if host != "hermes" else ""
+    block = cfg.get("hosts", {}).get(host, {})
+
+    if not block or block.get("enabled") is False:
+        print(f"  {label}Honcho is already disabled.\n")
+        return
+
+    block["enabled"] = False
+    _write_config(cfg)
+    print(f"  {label}Honcho disabled.")
+    print(f"  Saved to {_config_path()}\n")
+
+
+def cmd_sync(args) -> None:
+    """Sync Honcho config to all existing profiles.
+
+    Scans all Hermes profiles and creates host blocks for any that don't
+    have one yet. Inherits settings from the default host block.
+    """
+    try:
+        from hermes_cli.profiles import list_profiles
+        profiles = list_profiles()
+    except Exception as e:
+        print(f"  Could not list profiles: {e}\n")
+        return
+
+    cfg = _read_config()
+    if not cfg:
+        print("  No Honcho config found. Run 'hermes honcho setup' first.\n")
+        return
+
+    hosts = cfg.get("hosts", {})
+    default_block = hosts.get(HOST, {})
+    has_key = bool(cfg.get("apiKey") or os.environ.get("HONCHO_API_KEY"))
+
+    if not default_block and not has_key:
+        print("  Honcho not configured on default profile. Run 'hermes honcho setup' first.\n")
+        return
+
+    created = 0
+    skipped = 0
+    for p in profiles:
+        if p.name == "default":
+            continue
+        if clone_honcho_for_profile(p.name):
+            print(f"  + {p.name} -> hermes.{p.name}")
+            created += 1
+        else:
+            skipped += 1
+
+    if created:
+        print(f"\n  {created} profile(s) synced.")
+    else:
+        print("  All profiles already have Honcho config.")
+    if skipped:
+        print(f"  {skipped} profile(s) already configured (skipped).")
+    print()
+
+
+def sync_honcho_profiles_quiet() -> int:
+    """Sync Honcho host blocks for all profiles. Returns count of newly created blocks.
+
+    Called from `hermes update` -- no output, no exceptions.
+    """
+    try:
+        from hermes_cli.profiles import list_profiles
+        profiles = list_profiles()
+    except Exception:
+        return 0
+
+    cfg = _read_config()
+    if not cfg:
+        return 0
+
+    default_block = cfg.get("hosts", {}).get(HOST, {})
+    has_key = bool(cfg.get("apiKey") or os.environ.get("HONCHO_API_KEY"))
+    if not default_block and not has_key:
+        return 0
+
+    created = 0
+    for p in profiles:
+        if p.name == "default":
+            continue
+        if clone_honcho_for_profile(p.name):
+            created += 1
+    return created
+
+
+_profile_override: str | None = None
+
+
+def _host_key() -> str:
+    """Return the active Honcho host key, derived from the current Hermes profile."""
+    if _profile_override:
+        if _profile_override in ("default", "custom"):
+            return HOST
+        return f"{HOST}.{_profile_override}"
+    return resolve_active_host()


 def _config_path() -> Path:
@@ -52,7 +271,7 @@ def _write_config(cfg: dict, path: Path | None = None) -> None:

 def _resolve_api_key(cfg: dict) -> str:
    """Resolve API key with host -> root -> env fallback."""
-    host_key = ((cfg.get("hosts") or {}).get(HOST) or {}).get("apiKey")
+    host_key = ((cfg.get("hosts") or {}).get(_host_key()) or {}).get("apiKey")
    return host_key or cfg.get("apiKey", "") or os.environ.get("HONCHO_API_KEY", "")


@@ -118,10 +337,10 @@ def cmd_setup(args) -> None:
    if not _ensure_sdk_installed():
        return

-    # All writes go to hosts.hermes — root keys are managed by the user
-    # or the honcho CLI only.
+    # All writes go to the active host block — root keys are managed by
+    # the user or the honcho CLI only.
    hosts = cfg.setdefault("hosts", {})
-    hermes_host = hosts.setdefault(HOST, {})
+    hermes_host = hosts.setdefault(_host_key(), {})

    # API key — shared credential, lives at root so all hosts can read it
    current_key = cfg.get("apiKey", "")
@@ -148,7 +367,7 @@ def cmd_setup(args) -> None:
    if new_workspace:
        hermes_host["workspace"] = new_workspace

-    hermes_host.setdefault("aiPeer", HOST)
+    hermes_host.setdefault("aiPeer", _host_key())

    # Memory mode
    current_mode = hermes_host.get("memoryMode") or cfg.get("memoryMode", "hybrid")
@@ -205,9 +424,9 @@ def cmd_setup(args) -> None:
    # Test connection
    print("  Testing connection... ", end="", flush=True)
    try:
-        from honcho_integration.client import HonchoClientConfig, get_honcho_client, reset_honcho_client
+        from plugins.memory.honcho.client import HonchoClientConfig, get_honcho_client, reset_honcho_client
        reset_honcho_client()
-        hcfg = HonchoClientConfig.from_global_config()
+        hcfg = HonchoClientConfig.from_global_config(host=_host_key())
        get_honcho_client(hcfg)
        print("OK")
    except Exception as e:
@@ -237,8 +456,53 @@ def cmd_setup(args) -> None:
    print("    hermes honcho map <name> — map this directory to a session name\n")


+def _active_profile_name() -> str:
+    """Return the active Hermes profile name (respects --target-profile override)."""
+    if _profile_override:
+        return _profile_override
+    try:
+        from hermes_cli.profiles import get_active_profile_name
+        return get_active_profile_name()
+    except Exception:
+        return "default"
+
+
+def _all_profile_host_configs() -> list[tuple[str, str, dict]]:
+    """Return (profile_name, host_key, host_block) for every known profile.
+
+    Reads honcho.json once and maps each profile to its host block.
+    """
+    try:
+        from hermes_cli.profiles import list_profiles
+        profiles = list_profiles()
+    except Exception:
+        return [(_active_profile_name(), _host_key(), {})]
+
+    cfg = _read_config()
+    hosts = cfg.get("hosts", {})
+    results = []
+
+    # Default profile
+    default_block = hosts.get(HOST, {})
+    results.append(("default", HOST, default_block))
+
+    for p in profiles:
+        if p.name == "default":
+            continue
+        h = f"{HOST}.{p.name}"
+        results.append((p.name, h, hosts.get(h, {})))
+
+    return results
+
+
 def cmd_status(args) -> None:
    """Show current Honcho config and connection status."""
+    show_all = getattr(args, "all", False)
+
+    if show_all:
+        _cmd_status_all()
+        return
+
    try:
        import honcho  # noqa: F401
    except ImportError:
@@ -256,8 +520,8 @@ def cmd_status(args) -> None:
        return

    try:
-        from honcho_integration.client import HonchoClientConfig, get_honcho_client
-        hcfg = HonchoClientConfig.from_global_config()
+        from plugins.memory.honcho.client import HonchoClientConfig, get_honcho_client
+        hcfg = HonchoClientConfig.from_global_config(host=_host_key())
    except Exception as e:
        print(f"  Config error: {e}\n")
        return
@@ -265,11 +529,16 @@ def cmd_status(args) -> None:
    api_key = hcfg.api_key or ""
    masked = f"...{api_key[-8:]}" if len(api_key) > 8 else ("set" if api_key else "not set")

-    print("\nHoncho status\n" + "─" * 40)
+    profile = _active_profile_name()
+    profile_label = f" [{hcfg.host}]" if profile != "default" else ""
+
+    print(f"\nHoncho status{profile_label}\n" + "─" * 40)
+    if profile != "default":
+        print(f"  Profile:        {profile}")
+    print(f"  Host:           {hcfg.host}")
    print(f"  Enabled:        {hcfg.enabled}")
    print(f"  API key:        {masked}")
    print(f"  Workspace:      {hcfg.workspace_id}")
-    print(f"  Host:           {hcfg.host}")
    print(f"  Config path:    {active_path}")
    if write_path != active_path:
        print(f"  Write path:     {write_path}  (instance-local)")
@@ -287,8 +556,9 @@ def cmd_status(args) -> None:
    if hcfg.enabled and (hcfg.api_key or hcfg.base_url):
        print("\n  Connection... ", end="", flush=True)
        try:
-            get_honcho_client(hcfg)
-            print("OK\n")
+            client = get_honcho_client(hcfg)
+            print("OK")
+            _show_peer_cards(hcfg, client)
        except Exception as e:
            print(f"FAILED ({e})\n")
    else:
@@ -296,6 +566,90 @@ def cmd_status(args) -> None:
        print(f"\n  Not connected ({reason})\n")


+def _show_peer_cards(hcfg, client) -> None:
+    """Fetch and display peer cards for the active profile.
+
+    Uses get_or_create to ensure the session exists with peers configured.
+    This is idempotent -- if the session already exists on the server it's
+    just retrieved, not duplicated.
+    """
+    try:
+        from plugins.memory.honcho.session import HonchoSessionManager
+        mgr = HonchoSessionManager(honcho=client, config=hcfg)
+        session_key = hcfg.resolve_session_name()
+        mgr.get_or_create(session_key)
+
+        # User peer card
+        card = mgr.get_peer_card(session_key)
+        if card:
+            print(f"\n  User peer card ({len(card)} facts):")
+            for fact in card[:10]:
+                print(f"    - {fact}")
+            if len(card) > 10:
+                print(f"    ... and {len(card) - 10} more")
+
+        # AI peer representation
+        ai_rep = mgr.get_ai_representation(session_key)
+        ai_text = ai_rep.get("representation", "")
+        if ai_text:
+            # Truncate to first 200 chars
+            display = ai_text[:200] + ("..." if len(ai_text) > 200 else "")
+            print(f"\n  AI peer representation:")
+            print(f"    {display}")
+
+        if not card and not ai_text:
+            print("\n  No peer data yet (accumulates after first conversation)")
+
+        print()
+    except Exception as e:
+        print(f"\n  Peer data unavailable: {e}\n")
+
+
+def _cmd_status_all() -> None:
+    """Show Honcho config overview across all profiles."""
+    rows = _all_profile_host_configs()
+    cfg = _read_config()
+    active = _active_profile_name()
+
+    print(f"\nHoncho profiles ({len(rows)})\n" + "─" * 60)
+    print(f"  {'Profile':<14} {'Host':<22} {'Enabled':<9} {'Mode':<9} {'Recall':<9} {'Write'}")
+    print(f"  {'─' * 14} {'─' * 22} {'─' * 9} {'─' * 9} {'─' * 9} {'─' * 9}")
+
+    for name, host, block in rows:
+        enabled = block.get("enabled", cfg.get("enabled"))
+        if enabled is None:
+            # Auto-enable check: any credentials?
+            has_creds = bool(cfg.get("apiKey") or os.environ.get("HONCHO_API_KEY"))
+            enabled = has_creds if block else False
+        enabled_str = "yes" if enabled else "no"
+
+        mode = block.get("memoryMode") or cfg.get("memoryMode", "hybrid")
+        recall = block.get("recallMode") or cfg.get("recallMode", "hybrid")
+        write = block.get("writeFrequency") or cfg.get("writeFrequency", "async")
+
+        marker = " *" if name == active else ""
+        print(f"  {name + marker:<14} {host:<22} {enabled_str:<9} {mode:<9} {recall:<9} {write}")
+
+    print(f"\n  * active profile\n")
+
+
+def cmd_peers(args) -> None:
+    """Show peer identities across all profiles."""
+    rows = _all_profile_host_configs()
+    cfg = _read_config()
+
+    print(f"\nHoncho peer identities ({len(rows)} profiles)\n" + "─" * 50)
+    print(f"  {'Profile':<14} {'User peer':<16} {'AI peer'}")
+    print(f"  {'─' * 14} {'─' * 16} {'─' * 18}")
+
+    for name, host, block in rows:
+        user = block.get("peerName") or cfg.get("peerName") or "(not set)"
+        ai = block.get("aiPeer") or cfg.get("aiPeer") or host
+        print(f"  {name:<14} {user:<16} {ai}")
+
+    print()
+
+
 def cmd_sessions(args) -> None:
    """List known directory → session name mappings."""
    cfg = _read_config()
@@ -354,9 +708,9 @@ def cmd_peer(args) -> None:
    if user_name is None and ai_name is None and reasoning is None:
        # Show current values
        hosts = cfg.get("hosts", {})
-        hermes = hosts.get(HOST, {})
+        hermes = hosts.get(_host_key(), {})
        user = hermes.get('peerName') or cfg.get('peerName') or '(not set)'
-        ai = hermes.get('aiPeer') or cfg.get('aiPeer') or HOST
+        ai = hermes.get('aiPeer') or cfg.get('aiPeer') or _host_key()
        lvl = hermes.get("dialecticReasoningLevel") or cfg.get("dialecticReasoningLevel") or "low"
        max_chars = hermes.get("dialecticMaxChars") or cfg.get("dialecticMaxChars") or 600
        print("\nHoncho peers\n" + "─" * 40)
@@ -370,23 +724,26 @@ def cmd_peer(args) -> None:
        print(f"  Dialectic cap:        {max_chars} chars\n")
        return

+    host = _host_key()
+    label = f"[{host}] " if host != "hermes" else ""
+
    if user_name is not None:
-        cfg.setdefault("hosts", {}).setdefault(HOST, {})["peerName"] = user_name.strip()
+        cfg.setdefault("hosts", {}).setdefault(host, {})["peerName"] = user_name.strip()
        changed = True
-        print(f"  User peer → {user_name.strip()}")
+        print(f"  {label}User peer -> {user_name.strip()}")

    if ai_name is not None:
-        cfg.setdefault("hosts", {}).setdefault(HOST, {})["aiPeer"] = ai_name.strip()
+        cfg.setdefault("hosts", {}).setdefault(host, {})["aiPeer"] = ai_name.strip()
        changed = True
-        print(f"  AI peer   → {ai_name.strip()}")
+        print(f"  {label}AI peer   -> {ai_name.strip()}")

    if reasoning is not None:
        if reasoning not in REASONING_LEVELS:
            print(f"  Invalid reasoning level '{reasoning}'. Options: {', '.join(REASONING_LEVELS)}")
            return
-        cfg.setdefault("hosts", {}).setdefault(HOST, {})["dialecticReasoningLevel"] = reasoning
+        cfg.setdefault("hosts", {}).setdefault(host, {})["dialecticReasoningLevel"] = reasoning
        changed = True
-        print(f"  Dialectic reasoning level → {reasoning}")
+        print(f"  {label}Dialectic reasoning level -> {reasoning}")

    if changed:
        _write_config(cfg)
@@ -404,7 +761,7 @@ def cmd_mode(args) -> None:

    if mode_arg is None:
        current = (
-            (cfg.get("hosts") or {}).get(HOST, {}).get("memoryMode")
+            (cfg.get("hosts") or {}).get(_host_key(), {}).get("memoryMode")
            or cfg.get("memoryMode")
            or "hybrid"
        )
@@ -419,16 +776,18 @@ def cmd_mode(args) -> None:
        print(f"  Invalid mode '{mode_arg}'. Options: {', '.join(MODES)}\n")
        return

-    cfg.setdefault("hosts", {}).setdefault(HOST, {})["memoryMode"] = mode_arg
+    host = _host_key()
+    label = f"[{host}] " if host != "hermes" else ""
+    cfg.setdefault("hosts", {}).setdefault(host, {})["memoryMode"] = mode_arg
    _write_config(cfg)
-    print(f"  Memory mode → {mode_arg}  ({MODES[mode_arg]})\n")
+    print(f"  {label}Memory mode -> {mode_arg}  ({MODES[mode_arg]})\n")


 def cmd_tokens(args) -> None:
    """Show or set token budget settings."""
    cfg = _read_config()
    hosts = cfg.get("hosts", {})
-    hermes = hosts.get(HOST, {})
+    hermes = hosts.get(_host_key(), {})

    context = getattr(args, "context", None)
    dialectic = getattr(args, "dialectic", None)
@@ -451,14 +810,16 @@ def cmd_tokens(args) -> None:
        print("\n  Set with: hermes honcho tokens [--context N] [--dialectic N]\n")
        return

+    host = _host_key()
+    label = f"[{host}] " if host != "hermes" else ""
    changed = False
    if context is not None:
-        cfg.setdefault("hosts", {}).setdefault(HOST, {})["contextTokens"] = context
-        print(f"  context tokens → {context}")
+        cfg.setdefault("hosts", {}).setdefault(host, {})["contextTokens"] = context
+        print(f"  {label}context tokens -> {context}")
        changed = True
    if dialectic is not None:
-        cfg.setdefault("hosts", {}).setdefault(HOST, {})["dialecticMaxChars"] = dialectic
-        print(f"  dialectic cap  → {dialectic} chars")
+        cfg.setdefault("hosts", {}).setdefault(host, {})["dialecticMaxChars"] = dialectic
+        print(f"  {label}dialectic cap  -> {dialectic} chars")
        changed = True

    if changed:
@@ -477,9 +838,9 @@ def cmd_identity(args) -> None:
    show = getattr(args, "show", False)

    try:
-        from honcho_integration.client import HonchoClientConfig, get_honcho_client
-        from honcho_integration.session import HonchoSessionManager
-        hcfg = HonchoClientConfig.from_global_config()
+        from plugins.memory.honcho.client import HonchoClientConfig, get_honcho_client
+        from plugins.memory.honcho.session import HonchoSessionManager
+        hcfg = HonchoClientConfig.from_global_config(host=_host_key())
        client = get_honcho_client(hcfg)
        mgr = HonchoSessionManager(honcho=client, config=hcfg)
        session_key = hcfg.resolve_session_name()
@@ -642,12 +1003,12 @@ def cmd_migrate(args) -> None:
            answer = _prompt("  Upload user memory files to Honcho now?", default="y")
            if answer.lower() in ("y", "yes"):
                try:
-                    from honcho_integration.client import (
+                    from plugins.memory.honcho.client import (
                        HonchoClientConfig,
                        get_honcho_client,
                        reset_honcho_client,
                    )
-                    from honcho_integration.session import HonchoSessionManager
+                    from plugins.memory.honcho.session import HonchoSessionManager

                    reset_honcho_client()
                    hcfg = HonchoClientConfig.from_global_config()
@@ -692,12 +1053,12 @@ def cmd_migrate(args) -> None:
            answer = _prompt("  Seed AI identity from all detected files now?", default="y")
            if answer.lower() in ("y", "yes"):
                try:
-                    from honcho_integration.client import (
+                    from plugins.memory.honcho.client import (
                        HonchoClientConfig,
                        get_honcho_client,
                        reset_honcho_client,
                    )
-                    from honcho_integration.session import HonchoSessionManager
+                    from plugins.memory.honcho.session import HonchoSessionManager

                    reset_honcho_client()
                    hcfg = HonchoClientConfig.from_global_config()
@@ -770,11 +1131,16 @@ def cmd_migrate(args) -> None:

 def honcho_command(args) -> None:
    """Route honcho subcommands."""
+    global _profile_override
+    _profile_override = getattr(args, "target_profile", None)
+
    sub = getattr(args, "honcho_command", None)
    if sub == "setup" or sub is None:
        cmd_setup(args)
    elif sub == "status":
        cmd_status(args)
+    elif sub == "peers":
+        cmd_peers(args)
    elif sub == "sessions":
        cmd_sessions(args)
    elif sub == "map":
@@ -789,6 +1155,12 @@ def honcho_command(args) -> None:
        cmd_identity(args)
    elif sub == "migrate":
        cmd_migrate(args)
+    elif sub == "enable":
+        cmd_enable(args)
+    elif sub == "disable":
+        cmd_disable(args)
+    elif sub == "sync":
+        cmd_sync(args)
    else:
        print(f"  Unknown honcho command: {sub}")
-        print("  Available: setup, status, sessions, map, peer, mode, tokens, identity, migrate\n")
+        print("  Available: setup, status, sessions, map, peer, mode, tokens, identity, migrate, enable, disable, sync\n")
@@ -31,16 +31,47 @@ GLOBAL_CONFIG_PATH = Path.home() / ".honcho" / "config.json"
 HOST = "hermes"


+def resolve_active_host() -> str:
+    """Derive the Honcho host key from the active Hermes profile.
+
+    Resolution order:
+      1. HERMES_HONCHO_HOST env var (explicit override)
+      2. Active profile name via profiles system -> ``hermes.<profile>``
+      3. Fallback: ``"hermes"`` (default profile)
+    """
+    explicit = os.environ.get("HERMES_HONCHO_HOST", "").strip()
+    if explicit:
+        return explicit
+
+    try:
+        from hermes_cli.profiles import get_active_profile_name
+        profile = get_active_profile_name()
+        if profile and profile not in ("default", "custom"):
+            return f"{HOST}.{profile}"
+    except Exception:
+        pass
+    return HOST
+
+
 def resolve_config_path() -> Path:
    """Return the active Honcho config path.

-    Checks $HERMES_HOME/honcho.json first (instance-local), then falls back
-    to ~/.honcho/config.json (global).  Returns the global path if neither
-    exists (for first-time setup writes).
+    Resolution order:
+      1. $HERMES_HOME/honcho.json      (profile-local, if it exists)
+      2. ~/.hermes/honcho.json          (default profile — shared host blocks live here)
+      3. ~/.honcho/config.json          (global, cross-app interop)
+
+    Returns the global path if none exist (for first-time setup writes).
    """
    local_path = get_hermes_home() / "honcho.json"
    if local_path.exists():
        return local_path
+
+    # Default profile's config — host blocks accumulate here via setup/clone
+    default_path = Path.home() / ".hermes" / "honcho.json"
+    if default_path != local_path and default_path.exists():
+        return default_path
+
    return GLOBAL_CONFIG_PATH


@@ -54,6 +85,16 @@ def _normalize_recall_mode(val: str) -> str:
    return val if val in _VALID_RECALL_MODES else "hybrid"


+_VALID_OBSERVATION_MODES = {"unified", "directional"}
+_OBSERVATION_MODE_ALIASES = {"shared": "unified", "separate": "directional", "cross": "directional"}
+
+
+def _normalize_observation_mode(val: str) -> str:
+    """Normalize observation mode values."""
+    val = _OBSERVATION_MODE_ALIASES.get(val, val)
+    return val if val in _VALID_OBSERVATION_MODES else "unified"
+
+
 def _resolve_memory_mode(
    global_val: str | dict,
    host_val: str | dict | None,
@@ -123,6 +164,10 @@ class HonchoClientConfig:
    # "context" — auto-injected context only, Honcho tools removed
    # "tools"   — Honcho tools only, no auto-injected context
    recall_mode: str = "hybrid"
+    # Observation mode: how Honcho peers observe each other.
+    # "unified"      — user peer observes self; all agents share one observation pool
+    # "directional"  — AI peer observes user; each agent keeps its own view
+    observation_mode: str = "unified"
    # Session resolution
    session_strategy: str = "per-directory"
    session_peer_prefix: bool = False
@@ -135,40 +180,49 @@ class HonchoClientConfig:
    explicitly_configured: bool = False

    @classmethod
-    def from_env(cls, workspace_id: str = "hermes") -> HonchoClientConfig:
+    def from_env(
+        cls,
+        workspace_id: str = "hermes",
+        host: str | None = None,
+    ) -> HonchoClientConfig:
        """Create config from environment variables (fallback)."""
+        resolved_host = host or resolve_active_host()
        api_key = os.environ.get("HONCHO_API_KEY")
        base_url = os.environ.get("HONCHO_BASE_URL", "").strip() or None
        return cls(
+            host=resolved_host,
            workspace_id=workspace_id,
            api_key=api_key,
            environment=os.environ.get("HONCHO_ENVIRONMENT", "production"),
            base_url=base_url,
+            ai_peer=resolved_host,
            enabled=bool(api_key or base_url),
        )

    @classmethod
    def from_global_config(
        cls,
-        host: str = HOST,
+        host: str | None = None,
        config_path: Path | None = None,
    ) -> HonchoClientConfig:
        """Create config from the resolved Honcho config path.

        Resolution: $HERMES_HOME/honcho.json -> ~/.honcho/config.json -> env vars.
+        When host is None, derives it from the active Hermes profile.
        """
+        resolved_host = host or resolve_active_host()
        path = config_path or resolve_config_path()
        if not path.exists():
            logger.debug("No global Honcho config at %s, falling back to env", path)
-            return cls.from_env()
+            return cls.from_env(host=resolved_host)

        try:
            raw = json.loads(path.read_text(encoding="utf-8"))
        except (json.JSONDecodeError, OSError) as e:
            logger.warning("Failed to read %s: %s, falling back to env", path, e)
-            return cls.from_env()
+            return cls.from_env(host=resolved_host)

-        host_block = (raw.get("hosts") or {}).get(host, {})
+        host_block = (raw.get("hosts") or {}).get(resolved_host, {})
        # A hosts.hermes block or explicit enabled flag means the user
        # intentionally configured Honcho for this host.
        _explicitly_configured = bool(host_block) or raw.get("enabled") is True
@@ -177,12 +231,12 @@ class HonchoClientConfig:
        workspace = (
            host_block.get("workspace")
            or raw.get("workspace")
-            or host
+            or resolved_host
        )
        ai_peer = (
            host_block.get("aiPeer")
            or raw.get("aiPeer")
-            or host
+            or resolved_host
        )
        linked_hosts = host_block.get("linkedHosts", [])

@@ -242,7 +296,7 @@ class HonchoClientConfig:
        )

        return cls(
-            host=host,
+            host=resolved_host,
            workspace_id=workspace,
            api_key=api_key,
            environment=environment,
@@ -273,6 +327,11 @@ class HonchoClientConfig:
                or raw.get("recallMode")
                or "hybrid"
            ),
+            observation_mode=_normalize_observation_mode(
+                host_block.get("observationMode")
+                or raw.get("observationMode")
+                or "unified"
+            ),
            session_strategy=session_strategy,
            session_peer_prefix=session_peer_prefix,
            sessions=raw.get("sessions", {}),
@@ -0,0 +1,7 @@
+name: honcho
+version: 1.0.0
+description: "Honcho AI-native memory — cross-session user modeling with dialectic Q&A, semantic search, and persistent conclusions."
+pip_dependencies:
+  - honcho-ai
+hooks:
+  - on_session_end
@@ -10,7 +10,7 @@ from dataclasses import dataclass, field
 from datetime import datetime
 from typing import Any, TYPE_CHECKING

-from honcho_integration.client import get_honcho_client
+from plugins.memory.honcho.client import get_honcho_client

 if TYPE_CHECKING:
    from honcho import Honcho
@@ -110,6 +110,9 @@ class HonchoSessionManager:
        self._dialectic_max_chars: int = (
            config.dialectic_max_chars if config else 600
        )
+        self._observation_mode: str = (
+            config.observation_mode if config else "unified"
+        )

        # Async write queue — started lazily on first enqueue
        self._async_queue: queue.Queue | None = None
@@ -159,14 +162,25 @@ class HonchoSessionManager:

        session = self.honcho.session(session_id)

-        # Configure peer observation settings.
-        # observe_me=True for AI peer so Honcho watches what the agent says
-        # and builds its representation over time — enabling identity formation.
-        from honcho.session import SessionPeerConfig
-        user_config = SessionPeerConfig(observe_me=True, observe_others=True)
-        ai_config = SessionPeerConfig(observe_me=True, observe_others=True)
+        # Configure peer observation settings based on observation_mode.
+        # Unified: user peer observes self, AI peer passive — all agents share
+        #          one observation pool via user self-observations.
+        # Directional: AI peer observes user — each agent keeps its own view.
+        try:
+            from honcho.session import SessionPeerConfig
+            if self._observation_mode == "directional":
+                user_config = SessionPeerConfig(observe_me=True, observe_others=False)
+                ai_config = SessionPeerConfig(observe_me=False, observe_others=True)
+            else:  # unified (default)
+                user_config = SessionPeerConfig(observe_me=True, observe_others=False)
+                ai_config = SessionPeerConfig(observe_me=False, observe_others=False)

-        session.add_peers([(user_peer, user_config), (assistant_peer, ai_config)])
+            session.add_peers([(user_peer, user_config), (assistant_peer, ai_config)])
+        except Exception as e:
+            logger.warning(
+                "Honcho session '%s' add_peers failed (non-fatal): %s",
+                session_id, e,
+            )

        # Load existing messages via context() - single call for messages + metadata
        existing_messages = []
@@ -231,7 +245,7 @@ class HonchoSessionManager:
            chat_id = parts[1] if len(parts) > 1 else key
            user_peer_id = self._sanitize_id(f"user-{channel}-{chat_id}")

-        assistant_peer_id = (
+        assistant_peer_id = self._sanitize_id(
            self._config.ai_peer if self._config else "hermes-assistant"
        )

@@ -487,12 +501,27 @@ class HonchoSessionManager:
        if not session:
            return ""

-        peer_id = session.assistant_peer_id if peer == "ai" else session.user_peer_id
-        target_peer = self._get_or_create_peer(peer_id)
        level = reasoning_level or self._dynamic_reasoning_level(query)

        try:
-            result = target_peer.chat(query, reasoning_level=level) or ""
+            if self._observation_mode == "directional":
+                # AI peer queries about the user (cross-observation)
+                if peer == "ai":
+                    ai_peer_obj = self._get_or_create_peer(session.assistant_peer_id)
+                    result = ai_peer_obj.chat(query, reasoning_level=level) or ""
+                else:
+                    ai_peer_obj = self._get_or_create_peer(session.assistant_peer_id)
+                    result = ai_peer_obj.chat(
+                        query,
+                        target=session.user_peer_id,
+                        reasoning_level=level,
+                    ) or ""
+            else:
+                # Unified: user peer queries self, or AI peer queries self
+                peer_id = session.assistant_peer_id if peer == "ai" else session.user_peer_id
+                target_peer = self._get_or_create_peer(peer_id)
+                result = target_peer.chat(query, reasoning_level=level) or ""
+
            # Apply Hermes-side char cap before caching
            if result and self._dialectic_max_chars and len(result) > self._dialectic_max_chars:
                result = result[:self._dialectic_max_chars].rsplit(" ", 1)[0] + " …"
@@ -889,9 +918,16 @@ class HonchoSessionManager:
            logger.warning("No session cached for '%s', skipping conclusion", session_key)
            return False

-        assistant_peer = self._get_or_create_peer(session.assistant_peer_id)
        try:
-            conclusions_scope = assistant_peer.conclusions_of(session.user_peer_id)
+            if self._observation_mode == "directional":
+                # AI peer creates conclusion about user (cross-observation)
+                assistant_peer = self._get_or_create_peer(session.assistant_peer_id)
+                conclusions_scope = assistant_peer.conclusions_of(session.user_peer_id)
+            else:
+                # Unified: user peer creates self-conclusion
+                user_peer = self._get_or_create_peer(session.user_peer_id)
+                conclusions_scope = user_peer.conclusions_of(session.user_peer_id)
+
            conclusions_scope.create([{
                "content": content.strip(),
                "session_id": session.honcho_session_id,
@@ -0,0 +1,38 @@
+# Mem0 Memory Provider
+
+Server-side LLM fact extraction with semantic search, reranking, and automatic deduplication.
+
+## Requirements
+
+- `pip install mem0ai`
+- Mem0 API key from [app.mem0.ai](https://app.mem0.ai)
+
+## Setup
+
+```bash
+hermes memory setup    # select "mem0"
+```
+
+Or manually:
+```bash
+hermes config set memory.provider mem0
+echo "MEM0_API_KEY=your-key" >> ~/.hermes/.env
+```
+
+## Config
+
+Config file: `$HERMES_HOME/mem0.json`
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `user_id` | `hermes-user` | User identifier on Mem0 |
+| `agent_id` | `hermes` | Agent identifier |
+| `rerank` | `true` | Enable reranking for recall |
+
+## Tools
+
+| Tool | Description |
+|------|-------------|
+| `mem0_profile` | All stored memories about the user |
+| `mem0_search` | Semantic search with optional reranking |
+| `mem0_conclude` | Store a fact verbatim (no LLM extraction) |
@@ -0,0 +1,353 @@
+"""Mem0 memory plugin — MemoryProvider interface.
+
+Server-side LLM fact extraction, semantic search with reranking, and
+automatic deduplication via the Mem0 Platform API.
+
+Original PR #2933 by kartik-mem0, adapted to MemoryProvider ABC.
+
+Config via environment variables:
+  MEM0_API_KEY       — Mem0 Platform API key (required)
+  MEM0_USER_ID       — User identifier (default: hermes-user)
+  MEM0_AGENT_ID      — Agent identifier (default: hermes)
+
+Or via $HERMES_HOME/mem0.json.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import threading
+import time
+from pathlib import Path
+from typing import Any, Dict, List
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+# Circuit breaker: after this many consecutive failures, pause API calls
+# for _BREAKER_COOLDOWN_SECS to avoid hammering a down server.
+_BREAKER_THRESHOLD = 5
+_BREAKER_COOLDOWN_SECS = 120
+
+
+# ---------------------------------------------------------------------------
+# Config
+# ---------------------------------------------------------------------------
+
+def _load_config() -> dict:
+    """Load config from env vars, with $HERMES_HOME/mem0.json overrides.
+
+    Environment variables provide defaults; mem0.json (if present) overrides
+    individual keys.  This avoids a silent failure when the JSON file exists
+    but is missing fields like ``api_key`` that the user set in ``.env``.
+    """
+    from hermes_constants import get_hermes_home
+
+    config = {
+        "api_key": os.environ.get("MEM0_API_KEY", ""),
+        "user_id": os.environ.get("MEM0_USER_ID", "hermes-user"),
+        "agent_id": os.environ.get("MEM0_AGENT_ID", "hermes"),
+        "rerank": True,
+        "keyword_search": False,
+    }
+
+    config_path = get_hermes_home() / "mem0.json"
+    if config_path.exists():
+        try:
+            file_cfg = json.loads(config_path.read_text(encoding="utf-8"))
+            config.update({k: v for k, v in file_cfg.items()
+                           if v is not None and v != ""})
+        except Exception:
+            pass
+
+    return config
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas
+# ---------------------------------------------------------------------------
+
+PROFILE_SCHEMA = {
+    "name": "mem0_profile",
+    "description": (
+        "Retrieve all stored memories about the user — preferences, facts, "
+        "project context. Fast, no reranking. Use at conversation start."
+    ),
+    "parameters": {"type": "object", "properties": {}, "required": []},
+}
+
+SEARCH_SCHEMA = {
+    "name": "mem0_search",
+    "description": (
+        "Search memories by meaning. Returns relevant facts ranked by similarity. "
+        "Set rerank=true for higher accuracy on important queries."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "What to search for."},
+            "rerank": {"type": "boolean", "description": "Enable reranking for precision (default: false)."},
+            "top_k": {"type": "integer", "description": "Max results (default: 10, max: 50)."},
+        },
+        "required": ["query"],
+    },
+}
+
+CONCLUDE_SCHEMA = {
+    "name": "mem0_conclude",
+    "description": (
+        "Store a durable fact about the user. Stored verbatim (no LLM extraction). "
+        "Use for explicit preferences, corrections, or decisions."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "conclusion": {"type": "string", "description": "The fact to store."},
+        },
+        "required": ["conclusion"],
+    },
+}
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class Mem0MemoryProvider(MemoryProvider):
+    """Mem0 Platform memory with server-side extraction and semantic search."""
+
+    def __init__(self):
+        self._config = None
+        self._client = None
+        self._client_lock = threading.Lock()
+        self._api_key = ""
+        self._user_id = "hermes-user"
+        self._agent_id = "hermes"
+        self._rerank = True
+        self._prefetch_result = ""
+        self._prefetch_lock = threading.Lock()
+        self._prefetch_thread = None
+        self._sync_thread = None
+        # Circuit breaker state
+        self._consecutive_failures = 0
+        self._breaker_open_until = 0.0
+
+    @property
+    def name(self) -> str:
+        return "mem0"
+
+    def is_available(self) -> bool:
+        cfg = _load_config()
+        return bool(cfg.get("api_key"))
+
+    def save_config(self, values, hermes_home):
+        """Write config to $HERMES_HOME/mem0.json."""
+        import json
+        from pathlib import Path
+        config_path = Path(hermes_home) / "mem0.json"
+        existing = {}
+        if config_path.exists():
+            try:
+                existing = json.loads(config_path.read_text())
+            except Exception:
+                pass
+        existing.update(values)
+        config_path.write_text(json.dumps(existing, indent=2))
+
+    def get_config_schema(self):
+        return [
+            {"key": "api_key", "description": "Mem0 Platform API key", "secret": True, "required": True, "env_var": "MEM0_API_KEY", "url": "https://app.mem0.ai"},
+            {"key": "user_id", "description": "User identifier", "default": "hermes-user"},
+            {"key": "agent_id", "description": "Agent identifier", "default": "hermes"},
+            {"key": "rerank", "description": "Enable reranking for recall", "default": "true", "choices": ["true", "false"]},
+        ]
+
+    def _get_client(self):
+        """Thread-safe client accessor with lazy initialization."""
+        with self._client_lock:
+            if self._client is not None:
+                return self._client
+            try:
+                from mem0 import MemoryClient
+                self._client = MemoryClient(api_key=self._api_key)
+                return self._client
+            except ImportError:
+                raise RuntimeError("mem0 package not installed. Run: pip install mem0ai")
+
+    def _is_breaker_open(self) -> bool:
+        """Return True if the circuit breaker is tripped (too many failures)."""
+        if self._consecutive_failures < _BREAKER_THRESHOLD:
+            return False
+        if time.monotonic() >= self._breaker_open_until:
+            # Cooldown expired — reset and allow a retry
+            self._consecutive_failures = 0
+            return False
+        return True
+
+    def _record_success(self):
+        self._consecutive_failures = 0
+
+    def _record_failure(self):
+        self._consecutive_failures += 1
+        if self._consecutive_failures >= _BREAKER_THRESHOLD:
+            self._breaker_open_until = time.monotonic() + _BREAKER_COOLDOWN_SECS
+            logger.warning(
+                "Mem0 circuit breaker tripped after %d consecutive failures. "
+                "Pausing API calls for %ds.",
+                self._consecutive_failures, _BREAKER_COOLDOWN_SECS,
+            )
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        self._config = _load_config()
+        self._api_key = self._config.get("api_key", "")
+        self._user_id = self._config.get("user_id", "hermes-user")
+        self._agent_id = self._config.get("agent_id", "hermes")
+        self._rerank = self._config.get("rerank", True)
+
+    def system_prompt_block(self) -> str:
+        return (
+            "# Mem0 Memory\n"
+            f"Active. User: {self._user_id}.\n"
+            "Use mem0_search to find memories, mem0_conclude to store facts, "
+            "mem0_profile for a full overview."
+        )
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        if self._prefetch_thread and self._prefetch_thread.is_alive():
+            self._prefetch_thread.join(timeout=3.0)
+        with self._prefetch_lock:
+            result = self._prefetch_result
+            self._prefetch_result = ""
+        if not result:
+            return ""
+        return f"## Mem0 Memory\n{result}"
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        if self._is_breaker_open():
+            return
+
+        def _run():
+            try:
+                client = self._get_client()
+                results = client.search(
+                    query=query,
+                    user_id=self._user_id,
+                    rerank=self._rerank,
+                    top_k=5,
+                )
+                if results:
+                    lines = [r.get("memory", "") for r in results if r.get("memory")]
+                    with self._prefetch_lock:
+                        self._prefetch_result = "\n".join(f"- {l}" for l in lines)
+                self._record_success()
+            except Exception as e:
+                self._record_failure()
+                logger.debug("Mem0 prefetch failed: %s", e)
+
+        self._prefetch_thread = threading.Thread(target=_run, daemon=True, name="mem0-prefetch")
+        self._prefetch_thread.start()
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Send the turn to Mem0 for server-side fact extraction (non-blocking)."""
+        if self._is_breaker_open():
+            return
+
+        def _sync():
+            try:
+                client = self._get_client()
+                messages = [
+                    {"role": "user", "content": user_content},
+                    {"role": "assistant", "content": assistant_content},
+                ]
+                client.add(messages, user_id=self._user_id, agent_id=self._agent_id)
+                self._record_success()
+            except Exception as e:
+                self._record_failure()
+                logger.warning("Mem0 sync failed: %s", e)
+
+        # Wait for any previous sync before starting a new one
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=5.0)
+
+        self._sync_thread = threading.Thread(target=_sync, daemon=True, name="mem0-sync")
+        self._sync_thread.start()
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        return [PROFILE_SCHEMA, SEARCH_SCHEMA, CONCLUDE_SCHEMA]
+
+    def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
+        if self._is_breaker_open():
+            return json.dumps({
+                "error": "Mem0 API temporarily unavailable (multiple consecutive failures). Will retry automatically."
+            })
+
+        try:
+            client = self._get_client()
+        except Exception as e:
+            return json.dumps({"error": str(e)})
+
+        if tool_name == "mem0_profile":
+            try:
+                memories = client.get_all(user_id=self._user_id)
+                self._record_success()
+                if not memories:
+                    return json.dumps({"result": "No memories stored yet."})
+                lines = [m.get("memory", "") for m in memories if m.get("memory")]
+                return json.dumps({"result": "\n".join(lines), "count": len(lines)})
+            except Exception as e:
+                self._record_failure()
+                return json.dumps({"error": f"Failed to fetch profile: {e}"})
+
+        elif tool_name == "mem0_search":
+            query = args.get("query", "")
+            if not query:
+                return json.dumps({"error": "Missing required parameter: query"})
+            rerank = args.get("rerank", False)
+            top_k = min(int(args.get("top_k", 10)), 50)
+            try:
+                results = client.search(
+                    query=query, user_id=self._user_id,
+                    rerank=rerank, top_k=top_k,
+                )
+                self._record_success()
+                if not results:
+                    return json.dumps({"result": "No relevant memories found."})
+                items = [{"memory": r.get("memory", ""), "score": r.get("score", 0)} for r in results]
+                return json.dumps({"results": items, "count": len(items)})
+            except Exception as e:
+                self._record_failure()
+                return json.dumps({"error": f"Search failed: {e}"})
+
+        elif tool_name == "mem0_conclude":
+            conclusion = args.get("conclusion", "")
+            if not conclusion:
+                return json.dumps({"error": "Missing required parameter: conclusion"})
+            try:
+                client.add(
+                    [{"role": "user", "content": conclusion}],
+                    user_id=self._user_id,
+                    agent_id=self._agent_id,
+                    infer=False,
+                )
+                self._record_success()
+                return json.dumps({"result": "Fact stored."})
+            except Exception as e:
+                self._record_failure()
+                return json.dumps({"error": f"Failed to store: {e}"})
+
+        return json.dumps({"error": f"Unknown tool: {tool_name}"})
+
+    def shutdown(self) -> None:
+        for t in (self._prefetch_thread, self._sync_thread):
+            if t and t.is_alive():
+                t.join(timeout=5.0)
+        with self._client_lock:
+            self._client = None
+
+
+def register(ctx) -> None:
+    """Register Mem0 as a memory provider plugin."""
+    ctx.register_memory_provider(Mem0MemoryProvider())
@@ -0,0 +1,5 @@
+name: mem0
+version: 1.0.0
+description: "Mem0 — server-side LLM fact extraction with semantic search, reranking, and automatic deduplication."
+pip_dependencies:
+  - mem0ai
@@ -0,0 +1,40 @@
+# OpenViking Memory Provider
+
+Context database by Volcengine (ByteDance) with filesystem-style knowledge hierarchy, tiered retrieval, and automatic memory extraction.
+
+## Requirements
+
+- `pip install openviking`
+- OpenViking server running (`openviking-server`)
+- Embedding + VLM model configured in `~/.openviking/ov.conf`
+
+## Setup
+
+```bash
+hermes memory setup    # select "openviking"
+```
+
+Or manually:
+```bash
+hermes config set memory.provider openviking
+echo "OPENVIKING_ENDPOINT=http://localhost:1933" >> ~/.hermes/.env
+```
+
+## Config
+
+All config via environment variables in `.env`:
+
+| Env Var | Default | Description |
+|---------|---------|-------------|
+| `OPENVIKING_ENDPOINT` | `http://127.0.0.1:1933` | Server URL |
+| `OPENVIKING_API_KEY` | (none) | API key (optional) |
+
+## Tools
+
+| Tool | Description |
+|------|-------------|
+| `viking_search` | Semantic search with fast/deep/auto modes |
+| `viking_read` | Read content at a viking:// URI (abstract/overview/full) |
+| `viking_browse` | Filesystem-style navigation (list/tree/stat) |
+| `viking_remember` | Store a fact for extraction on session commit |
+| `viking_add_resource` | Ingest URLs/docs into the knowledge base |
@@ -0,0 +1,593 @@
+"""OpenViking memory plugin — full bidirectional MemoryProvider interface.
+
+Context database by Volcengine (ByteDance) that organizes agent knowledge
+into a filesystem hierarchy (viking:// URIs) with tiered context loading,
+automatic memory extraction, and session management.
+
+Original PR #3369 by Mibayy, rewritten to use the full OpenViking session
+lifecycle instead of read-only search endpoints.
+
+Config via environment variables (profile-scoped via each profile's .env):
+  OPENVIKING_ENDPOINT  — Server URL (default: http://127.0.0.1:1933)
+  OPENVIKING_API_KEY   — API key (required for authenticated servers)
+  OPENVIKING_ACCOUNT   — Tenant account (default: root)
+  OPENVIKING_USER      — Tenant user (default: default)
+
+Capabilities:
+  - Automatic memory extraction on session commit (6 categories)
+  - Tiered context: L0 (~100 tokens), L1 (~2k), L2 (full)
+  - Semantic search with hierarchical directory retrieval
+  - Filesystem-style browsing via viking:// URIs
+  - Resource ingestion (URLs, docs, code)
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import threading
+from typing import Any, Dict, List, Optional
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+_DEFAULT_ENDPOINT = "http://127.0.0.1:1933"
+_TIMEOUT = 30.0
+
+
+# ---------------------------------------------------------------------------
+# HTTP helper — uses httpx to avoid requiring the openviking SDK
+# ---------------------------------------------------------------------------
+
+def _get_httpx():
+    """Lazy import httpx."""
+    try:
+        import httpx
+        return httpx
+    except ImportError:
+        return None
+
+
+class _VikingClient:
+    """Thin HTTP client for the OpenViking REST API."""
+
+    def __init__(self, endpoint: str, api_key: str = "",
+                 account: str = "", user: str = ""):
+        self._endpoint = endpoint.rstrip("/")
+        self._api_key = api_key
+        self._account = account or os.environ.get("OPENVIKING_ACCOUNT", "root")
+        self._user = user or os.environ.get("OPENVIKING_USER", "default")
+        self._httpx = _get_httpx()
+        if self._httpx is None:
+            raise ImportError("httpx is required for OpenViking: pip install httpx")
+
+    def _headers(self) -> dict:
+        h = {
+            "Content-Type": "application/json",
+            "X-OpenViking-Account": self._account,
+            "X-OpenViking-User": self._user,
+        }
+        if self._api_key:
+            h["X-API-Key"] = self._api_key
+        return h
+
+    def _url(self, path: str) -> str:
+        return f"{self._endpoint}{path}"
+
+    def get(self, path: str, **kwargs) -> dict:
+        resp = self._httpx.get(
+            self._url(path), headers=self._headers(), timeout=_TIMEOUT, **kwargs
+        )
+        resp.raise_for_status()
+        return resp.json()
+
+    def post(self, path: str, payload: dict = None, **kwargs) -> dict:
+        resp = self._httpx.post(
+            self._url(path), json=payload or {}, headers=self._headers(),
+            timeout=_TIMEOUT, **kwargs
+        )
+        resp.raise_for_status()
+        return resp.json()
+
+    def health(self) -> bool:
+        try:
+            resp = self._httpx.get(
+                self._url("/health"), timeout=3.0
+            )
+            return resp.status_code == 200
+        except Exception:
+            return False
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas
+# ---------------------------------------------------------------------------
+
+SEARCH_SCHEMA = {
+    "name": "viking_search",
+    "description": (
+        "Semantic search over the OpenViking knowledge base. "
+        "Returns ranked results with viking:// URIs for deeper reading. "
+        "Use mode='deep' for complex queries that need reasoning across "
+        "multiple sources, 'fast' for simple lookups."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "Search query."},
+            "mode": {
+                "type": "string", "enum": ["auto", "fast", "deep"],
+                "description": "Search depth (default: auto).",
+            },
+            "scope": {
+                "type": "string",
+                "description": "Viking URI prefix to scope search (e.g. 'viking://resources/docs/').",
+            },
+            "limit": {"type": "integer", "description": "Max results (default: 10)."},
+        },
+        "required": ["query"],
+    },
+}
+
+READ_SCHEMA = {
+    "name": "viking_read",
+    "description": (
+        "Read content at a viking:// URI. Three detail levels:\n"
+        "  abstract — ~100 token summary (L0)\n"
+        "  overview — ~2k token key points (L1)\n"
+        "  full — complete content (L2)\n"
+        "Start with abstract/overview, only use full when you need details."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "uri": {"type": "string", "description": "viking:// URI to read."},
+            "level": {
+                "type": "string", "enum": ["abstract", "overview", "full"],
+                "description": "Detail level (default: overview).",
+            },
+        },
+        "required": ["uri"],
+    },
+}
+
+BROWSE_SCHEMA = {
+    "name": "viking_browse",
+    "description": (
+        "Browse the OpenViking knowledge store like a filesystem.\n"
+        "  list — show directory contents\n"
+        "  tree — show hierarchy\n"
+        "  stat — show metadata for a URI"
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "action": {
+                "type": "string", "enum": ["tree", "list", "stat"],
+                "description": "Browse action.",
+            },
+            "path": {
+                "type": "string",
+                "description": "Viking URI path (default: viking://). Examples: 'viking://resources/', 'viking://user/memories/'.",
+            },
+        },
+        "required": ["action"],
+    },
+}
+
+REMEMBER_SCHEMA = {
+    "name": "viking_remember",
+    "description": (
+        "Explicitly store a fact or memory in the OpenViking knowledge base. "
+        "Use for important information the agent should remember long-term. "
+        "The system automatically categorizes and indexes the memory."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "content": {"type": "string", "description": "The information to remember."},
+            "category": {
+                "type": "string",
+                "enum": ["preference", "entity", "event", "case", "pattern"],
+                "description": "Memory category (default: auto-detected).",
+            },
+        },
+        "required": ["content"],
+    },
+}
+
+ADD_RESOURCE_SCHEMA = {
+    "name": "viking_add_resource",
+    "description": (
+        "Add a URL or document to the OpenViking knowledge base. "
+        "Supports web pages, GitHub repos, PDFs, markdown, code files. "
+        "The system automatically parses, indexes, and generates summaries."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "url": {"type": "string", "description": "URL or path of the resource to add."},
+            "reason": {
+                "type": "string",
+                "description": "Why this resource is relevant (improves search).",
+            },
+        },
+        "required": ["url"],
+    },
+}
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class OpenVikingMemoryProvider(MemoryProvider):
+    """Full bidirectional memory via OpenViking context database."""
+
+    def __init__(self):
+        self._client: Optional[_VikingClient] = None
+        self._endpoint = ""
+        self._api_key = ""
+        self._session_id = ""
+        self._turn_count = 0
+        self._sync_thread: Optional[threading.Thread] = None
+        self._prefetch_result = ""
+        self._prefetch_lock = threading.Lock()
+        self._prefetch_thread: Optional[threading.Thread] = None
+
+    @property
+    def name(self) -> str:
+        return "openviking"
+
+    def is_available(self) -> bool:
+        """Check if OpenViking endpoint is configured. No network calls."""
+        return bool(os.environ.get("OPENVIKING_ENDPOINT"))
+
+    def get_config_schema(self):
+        return [
+            {
+                "key": "endpoint",
+                "description": "OpenViking server URL",
+                "required": True,
+                "default": _DEFAULT_ENDPOINT,
+                "env_var": "OPENVIKING_ENDPOINT",
+            },
+            {
+                "key": "api_key",
+                "description": "OpenViking API key",
+                "secret": True,
+                "env_var": "OPENVIKING_API_KEY",
+            },
+        ]
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        self._endpoint = os.environ.get("OPENVIKING_ENDPOINT", _DEFAULT_ENDPOINT)
+        self._api_key = os.environ.get("OPENVIKING_API_KEY", "")
+        self._session_id = session_id
+        self._turn_count = 0
+
+        try:
+            self._client = _VikingClient(self._endpoint, self._api_key)
+            if not self._client.health():
+                logger.warning("OpenViking server at %s is not reachable", self._endpoint)
+                self._client = None
+        except ImportError:
+            logger.warning("httpx not installed — OpenViking plugin disabled")
+            self._client = None
+
+    def system_prompt_block(self) -> str:
+        if not self._client:
+            return ""
+        # Provide brief info about the knowledge base
+        try:
+            # Check what's in the knowledge base via a root listing
+            resp = self._client.get("/api/v1/fs/ls", params={"uri": "viking://"})
+            result = resp.get("result", [])
+            children = len(result) if isinstance(result, list) else 0
+            if children == 0:
+                return ""
+            return (
+                "# OpenViking Knowledge Base\n"
+                f"Active. Endpoint: {self._endpoint}\n"
+                "Use viking_search to find information, viking_read for details "
+                "(abstract/overview/full), viking_browse to explore.\n"
+                "Use viking_remember to store facts, viking_add_resource to index URLs/docs."
+            )
+        except Exception:
+            return (
+                "# OpenViking Knowledge Base\n"
+                f"Active. Endpoint: {self._endpoint}\n"
+                "Use viking_search, viking_read, viking_browse, "
+                "viking_remember, viking_add_resource."
+            )
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        """Return prefetched results from the background thread."""
+        if self._prefetch_thread and self._prefetch_thread.is_alive():
+            self._prefetch_thread.join(timeout=3.0)
+        with self._prefetch_lock:
+            result = self._prefetch_result
+            self._prefetch_result = ""
+        if not result:
+            return ""
+        return f"## OpenViking Context\n{result}"
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        """Fire a background search to pre-load relevant context."""
+        if not self._client or not query:
+            return
+
+        def _run():
+            try:
+                client = _VikingClient(self._endpoint, self._api_key)
+                resp = client.post("/api/v1/search/find", {
+                    "query": query,
+                    "top_k": 5,
+                })
+                result = resp.get("result", {})
+                parts = []
+                for ctx_type in ("memories", "resources"):
+                    items = result.get(ctx_type, [])
+                    for item in items[:3]:
+                        uri = item.get("uri", "")
+                        abstract = item.get("abstract", "")
+                        score = item.get("score", 0)
+                        if abstract:
+                            parts.append(f"- [{score:.2f}] {abstract} ({uri})")
+                if parts:
+                    with self._prefetch_lock:
+                        self._prefetch_result = "\n".join(parts)
+            except Exception as e:
+                logger.debug("OpenViking prefetch failed: %s", e)
+
+        self._prefetch_thread = threading.Thread(
+            target=_run, daemon=True, name="openviking-prefetch"
+        )
+        self._prefetch_thread.start()
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Record the conversation turn in OpenViking's session (non-blocking)."""
+        if not self._client:
+            return
+
+        self._turn_count += 1
+
+        def _sync():
+            try:
+                client = _VikingClient(self._endpoint, self._api_key)
+                sid = self._session_id
+
+                # Add user message
+                client.post(f"/api/v1/sessions/{sid}/messages", {
+                    "role": "user",
+                    "content": user_content[:4000],  # trim very long messages
+                })
+                # Add assistant message
+                client.post(f"/api/v1/sessions/{sid}/messages", {
+                    "role": "assistant",
+                    "content": assistant_content[:4000],
+                })
+            except Exception as e:
+                logger.debug("OpenViking sync_turn failed: %s", e)
+
+        # Wait for any previous sync to finish before starting a new one
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=5.0)
+
+        self._sync_thread = threading.Thread(
+            target=_sync, daemon=True, name="openviking-sync"
+        )
+        self._sync_thread.start()
+
+    def on_session_end(self, messages: List[Dict[str, Any]]) -> None:
+        """Commit the session to trigger memory extraction.
+
+        OpenViking automatically extracts 6 categories of memories:
+        profile, preferences, entities, events, cases, and patterns.
+        """
+        if not self._client or self._turn_count == 0:
+            return
+
+        # Wait for any pending sync to finish first
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=10.0)
+
+        try:
+            self._client.post(f"/api/v1/sessions/{self._session_id}/commit")
+            logger.info("OpenViking session %s committed (%d turns)", self._session_id, self._turn_count)
+        except Exception as e:
+            logger.warning("OpenViking session commit failed: %s", e)
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        """Mirror built-in memory writes to OpenViking as explicit memories."""
+        if not self._client or action != "add" or not content:
+            return
+
+        def _write():
+            try:
+                client = _VikingClient(self._endpoint, self._api_key)
+                # Add as a user message with memory context so the commit
+                # picks it up as an explicit memory during extraction
+                client.post(f"/api/v1/sessions/{self._session_id}/messages", {
+                    "role": "user",
+                    "parts": [
+                        {"type": "text", "text": f"[Memory note — {target}] {content}"},
+                    ],
+                })
+            except Exception as e:
+                logger.debug("OpenViking memory mirror failed: %s", e)
+
+        t = threading.Thread(target=_write, daemon=True, name="openviking-memwrite")
+        t.start()
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        return [SEARCH_SCHEMA, READ_SCHEMA, BROWSE_SCHEMA, REMEMBER_SCHEMA, ADD_RESOURCE_SCHEMA]
+
+    def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
+        if not self._client:
+            return json.dumps({"error": "OpenViking server not connected"})
+
+        try:
+            if tool_name == "viking_search":
+                return self._tool_search(args)
+            elif tool_name == "viking_read":
+                return self._tool_read(args)
+            elif tool_name == "viking_browse":
+                return self._tool_browse(args)
+            elif tool_name == "viking_remember":
+                return self._tool_remember(args)
+            elif tool_name == "viking_add_resource":
+                return self._tool_add_resource(args)
+            return json.dumps({"error": f"Unknown tool: {tool_name}"})
+        except Exception as e:
+            return json.dumps({"error": str(e)})
+
+    def shutdown(self) -> None:
+        # Wait for background threads to finish
+        for t in (self._sync_thread, self._prefetch_thread):
+            if t and t.is_alive():
+                t.join(timeout=5.0)
+
+    # -- Tool implementations ------------------------------------------------
+
+    def _tool_search(self, args: dict) -> str:
+        query = args.get("query", "")
+        if not query:
+            return json.dumps({"error": "query is required"})
+
+        payload: Dict[str, Any] = {"query": query}
+        mode = args.get("mode", "auto")
+        if mode != "auto":
+            payload["mode"] = mode
+        if args.get("scope"):
+            payload["target_uri"] = args["scope"]
+        if args.get("limit"):
+            payload["top_k"] = args["limit"]
+
+        resp = self._client.post("/api/v1/search/find", payload)
+        result = resp.get("result", {})
+
+        # Format results for the model — keep it concise
+        formatted = []
+        for ctx_type in ("memories", "resources", "skills"):
+            items = result.get(ctx_type, [])
+            for item in items:
+                entry = {
+                    "uri": item.get("uri", ""),
+                    "type": ctx_type.rstrip("s"),
+                    "score": round(item.get("score", 0), 3),
+                    "abstract": item.get("abstract", ""),
+                }
+                if item.get("relations"):
+                    entry["related"] = [r.get("uri") for r in item["relations"][:3]]
+                formatted.append(entry)
+
+        return json.dumps({
+            "results": formatted,
+            "total": result.get("total", len(formatted)),
+        }, ensure_ascii=False)
+
+    def _tool_read(self, args: dict) -> str:
+        uri = args.get("uri", "")
+        if not uri:
+            return json.dumps({"error": "uri is required"})
+
+        level = args.get("level", "overview")
+        # Map our level names to OpenViking GET endpoints
+        if level == "abstract":
+            resp = self._client.get("/api/v1/content/abstract", params={"uri": uri})
+        elif level == "full":
+            resp = self._client.get("/api/v1/content/read", params={"uri": uri})
+        else:  # overview
+            resp = self._client.get("/api/v1/content/overview", params={"uri": uri})
+
+        result = resp.get("result", "")
+        # result is a plain string from the content endpoints
+        content = result if isinstance(result, str) else result.get("content", "")
+
+        # Truncate very long content to avoid flooding the context
+        if len(content) > 8000:
+            content = content[:8000] + "\n\n[... truncated, use a more specific URI or abstract level]"
+
+        return json.dumps({
+            "uri": uri,
+            "level": level,
+            "content": content,
+        }, ensure_ascii=False)
+
+    def _tool_browse(self, args: dict) -> str:
+        action = args.get("action", "list")
+        path = args.get("path", "viking://")
+
+        # Map action to the correct fs endpoint (all GET with uri= param)
+        endpoint_map = {"tree": "/api/v1/fs/tree", "list": "/api/v1/fs/ls", "stat": "/api/v1/fs/stat"}
+        endpoint = endpoint_map.get(action, "/api/v1/fs/ls")
+        resp = self._client.get(endpoint, params={"uri": path})
+        result = resp.get("result", {})
+
+        # Format list/tree results for readability
+        if action in ("list", "tree") and isinstance(result, list):
+            entries = []
+            for e in result[:50]:  # cap at 50 entries
+                entries.append({
+                    "name": e.get("rel_path", e.get("name", "")),
+                    "uri": e.get("uri", ""),
+                    "type": "dir" if e.get("isDir") else "file",
+                    "abstract": e.get("abstract", ""),
+                })
+            return json.dumps({"path": path, "entries": entries}, ensure_ascii=False)
+
+        return json.dumps(result, ensure_ascii=False)
+
+    def _tool_remember(self, args: dict) -> str:
+        content = args.get("content", "")
+        if not content:
+            return json.dumps({"error": "content is required"})
+
+        # Store as a session message that will be extracted during commit.
+        # The category hint helps OpenViking's extraction classify correctly.
+        category = args.get("category", "")
+        text = f"[Remember] {content}"
+        if category:
+            text = f"[Remember — {category}] {content}"
+
+        self._client.post(f"/api/v1/sessions/{self._session_id}/messages", {
+            "role": "user",
+            "parts": [
+                {"type": "text", "text": text},
+            ],
+        })
+
+        return json.dumps({
+            "status": "stored",
+            "message": "Memory recorded. Will be extracted and indexed on session commit.",
+        })
+
+    def _tool_add_resource(self, args: dict) -> str:
+        url = args.get("url", "")
+        if not url:
+            return json.dumps({"error": "url is required"})
+
+        payload: Dict[str, Any] = {"path": url}
+        if args.get("reason"):
+            payload["reason"] = args["reason"]
+
+        resp = self._client.post("/api/v1/resources", payload)
+        result = resp.get("result", {})
+
+        return json.dumps({
+            "status": "added",
+            "root_uri": result.get("root_uri", ""),
+            "message": "Resource queued for processing. Use viking_search after a moment to find it.",
+        }, ensure_ascii=False)
+
+
+# ---------------------------------------------------------------------------
+# Plugin entry point
+# ---------------------------------------------------------------------------
+
+def register(ctx) -> None:
+    """Register OpenViking as a memory provider plugin."""
+    ctx.register_memory_provider(OpenVikingMemoryProvider())
@@ -0,0 +1,9 @@
+name: openviking
+version: 2.0.0
+description: "OpenViking context database — session-managed memory with automatic extraction, tiered retrieval, and filesystem-style knowledge browsing."
+pip_dependencies:
+  - httpx
+requires_env:
+  - OPENVIKING_ENDPOINT
+hooks:
+  - on_session_end
@@ -0,0 +1,40 @@
+# RetainDB Memory Provider
+
+Cloud memory API with hybrid search (Vector + BM25 + Reranking) and 7 memory types.
+
+## Requirements
+
+- RetainDB account ($20/month) from [retaindb.com](https://www.retaindb.com)
+- `pip install requests`
+
+## Setup
+
+```bash
+hermes memory setup    # select "retaindb"
+```
+
+Or manually:
+```bash
+hermes config set memory.provider retaindb
+echo "RETAINDB_API_KEY=your-key" >> ~/.hermes/.env
+```
+
+## Config
+
+All config via environment variables in `.env`:
+
+| Env Var | Default | Description |
+|---------|---------|-------------|
+| `RETAINDB_API_KEY` | (required) | API key |
+| `RETAINDB_BASE_URL` | `https://api.retaindb.com` | API endpoint |
+| `RETAINDB_PROJECT` | auto (profile-scoped) | Project identifier |
+
+## Tools
+
+| Tool | Description |
+|------|-------------|
+| `retaindb_profile` | User's stable profile |
+| `retaindb_search` | Semantic search |
+| `retaindb_context` | Task-relevant context |
+| `retaindb_remember` | Store a fact with type + importance |
+| `retaindb_forget` | Delete a memory by ID |
@@ -0,0 +1,302 @@
+"""RetainDB memory plugin — MemoryProvider interface.
+
+Cross-session memory via RetainDB cloud API. Durable write-behind queue,
+semantic search with deduplication, and user profile retrieval.
+
+Original PR #2732 by Alinxus, adapted to MemoryProvider ABC.
+
+Config via environment variables:
+  RETAINDB_API_KEY    — API key (required)
+  RETAINDB_BASE_URL   — API endpoint (default: https://api.retaindb.com)
+  RETAINDB_PROJECT    — Project identifier (default: hermes)
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import threading
+from typing import Any, Dict, List
+
+from agent.memory_provider import MemoryProvider
+
+logger = logging.getLogger(__name__)
+
+_DEFAULT_BASE_URL = "https://api.retaindb.com"
+
+
+# ---------------------------------------------------------------------------
+# Tool schemas
+# ---------------------------------------------------------------------------
+
+PROFILE_SCHEMA = {
+    "name": "retaindb_profile",
+    "description": "Get the user's stable profile — preferences, facts, and patterns.",
+    "parameters": {"type": "object", "properties": {}, "required": []},
+}
+
+SEARCH_SCHEMA = {
+    "name": "retaindb_search",
+    "description": (
+        "Semantic search across stored memories. Returns ranked results "
+        "with relevance scores."
+    ),
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "What to search for."},
+            "top_k": {"type": "integer", "description": "Max results (default: 8, max: 20)."},
+        },
+        "required": ["query"],
+    },
+}
+
+CONTEXT_SCHEMA = {
+    "name": "retaindb_context",
+    "description": "Synthesized 'what matters now' context block for the current task.",
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "query": {"type": "string", "description": "Current task or question."},
+        },
+        "required": ["query"],
+    },
+}
+
+REMEMBER_SCHEMA = {
+    "name": "retaindb_remember",
+    "description": "Persist an explicit fact or preference to long-term memory.",
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "content": {"type": "string", "description": "The fact to remember."},
+            "memory_type": {
+                "type": "string",
+                "enum": ["preference", "fact", "decision", "context"],
+                "description": "Category (default: fact).",
+            },
+            "importance": {
+                "type": "number",
+                "description": "Importance 0-1 (default: 0.5).",
+            },
+        },
+        "required": ["content"],
+    },
+}
+
+FORGET_SCHEMA = {
+    "name": "retaindb_forget",
+    "description": "Delete a specific memory by ID.",
+    "parameters": {
+        "type": "object",
+        "properties": {
+            "memory_id": {"type": "string", "description": "Memory ID to delete."},
+        },
+        "required": ["memory_id"],
+    },
+}
+
+
+# ---------------------------------------------------------------------------
+# MemoryProvider implementation
+# ---------------------------------------------------------------------------
+
+class RetainDBMemoryProvider(MemoryProvider):
+    """RetainDB cloud memory with write-behind queue and semantic search."""
+
+    def __init__(self):
+        self._api_key = ""
+        self._base_url = _DEFAULT_BASE_URL
+        self._project = "hermes"
+        self._user_id = ""
+        self._prefetch_result = ""
+        self._prefetch_lock = threading.Lock()
+        self._prefetch_thread = None
+        self._sync_thread = None
+
+    @property
+    def name(self) -> str:
+        return "retaindb"
+
+    def is_available(self) -> bool:
+        return bool(os.environ.get("RETAINDB_API_KEY"))
+
+    def get_config_schema(self):
+        return [
+            {"key": "api_key", "description": "RetainDB API key", "secret": True, "required": True, "env_var": "RETAINDB_API_KEY", "url": "https://retaindb.com"},
+            {"key": "base_url", "description": "API endpoint", "default": "https://api.retaindb.com"},
+            {"key": "project", "description": "Project identifier", "default": "hermes"},
+        ]
+
+    def _headers(self) -> dict:
+        return {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+        }
+
+    def _api(self, method: str, path: str, **kwargs):
+        """Make an API call to RetainDB."""
+        import requests
+        url = f"{self._base_url}{path}"
+        resp = requests.request(method, url, headers=self._headers(), timeout=30, **kwargs)
+        resp.raise_for_status()
+        return resp.json()
+
+    def initialize(self, session_id: str, **kwargs) -> None:
+        self._api_key = os.environ.get("RETAINDB_API_KEY", "")
+        self._base_url = os.environ.get("RETAINDB_BASE_URL", _DEFAULT_BASE_URL)
+        self._user_id = kwargs.get("user_id", "default")
+        self._session_id = session_id
+
+        # Derive profile-scoped project name so different profiles don't
+        # share server-side memory.  Explicit RETAINDB_PROJECT always wins.
+        explicit_project = os.environ.get("RETAINDB_PROJECT")
+        if explicit_project:
+            self._project = explicit_project
+        else:
+            hermes_home = kwargs.get("hermes_home", "")
+            profile_name = os.path.basename(hermes_home) if hermes_home else ""
+            # Default profile (~/.hermes) → "hermes"; named profiles → "hermes-<name>"
+            if profile_name and profile_name != ".hermes":
+                self._project = f"hermes-{profile_name}"
+            else:
+                self._project = "hermes"
+
+    def system_prompt_block(self) -> str:
+        return (
+            "# RetainDB Memory\n"
+            f"Active. Project: {self._project}.\n"
+            "Use retaindb_search to find memories, retaindb_remember to store facts, "
+            "retaindb_profile for a user overview, retaindb_context for task-relevant context."
+        )
+
+    def prefetch(self, query: str, *, session_id: str = "") -> str:
+        if self._prefetch_thread and self._prefetch_thread.is_alive():
+            self._prefetch_thread.join(timeout=3.0)
+        with self._prefetch_lock:
+            result = self._prefetch_result
+            self._prefetch_result = ""
+        if not result:
+            return ""
+        return f"## RetainDB Memory\n{result}"
+
+    def queue_prefetch(self, query: str, *, session_id: str = "") -> None:
+        def _run():
+            try:
+                data = self._api("POST", "/v1/recall", json={
+                    "project": self._project,
+                    "query": query,
+                    "user_id": self._user_id,
+                    "top_k": 5,
+                })
+                results = data.get("results", [])
+                if results:
+                    lines = [r.get("content", "") for r in results if r.get("content")]
+                    with self._prefetch_lock:
+                        self._prefetch_result = "\n".join(f"- {l}" for l in lines)
+            except Exception as e:
+                logger.debug("RetainDB prefetch failed: %s", e)
+
+        self._prefetch_thread = threading.Thread(target=_run, daemon=True, name="retaindb-prefetch")
+        self._prefetch_thread.start()
+
+    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+        """Ingest conversation turn in background (non-blocking)."""
+        def _sync():
+            try:
+                self._api("POST", "/v1/ingest", json={
+                    "project": self._project,
+                    "user_id": self._user_id,
+                    "session_id": self._session_id,
+                    "messages": [
+                        {"role": "user", "content": user_content},
+                        {"role": "assistant", "content": assistant_content},
+                    ],
+                })
+            except Exception as e:
+                logger.warning("RetainDB sync failed: %s", e)
+
+        if self._sync_thread and self._sync_thread.is_alive():
+            self._sync_thread.join(timeout=5.0)
+        self._sync_thread = threading.Thread(target=_sync, daemon=True, name="retaindb-sync")
+        self._sync_thread.start()
+
+    def get_tool_schemas(self) -> List[Dict[str, Any]]:
+        return [PROFILE_SCHEMA, SEARCH_SCHEMA, CONTEXT_SCHEMA, REMEMBER_SCHEMA, FORGET_SCHEMA]
+
+    def handle_tool_call(self, tool_name: str, args: dict, **kwargs) -> str:
+        try:
+            if tool_name == "retaindb_profile":
+                data = self._api("GET", f"/v1/profile/{self._project}/{self._user_id}")
+                return json.dumps(data)
+
+            elif tool_name == "retaindb_search":
+                query = args.get("query", "")
+                if not query:
+                    return json.dumps({"error": "query is required"})
+                data = self._api("POST", "/v1/search", json={
+                    "project": self._project,
+                    "user_id": self._user_id,
+                    "query": query,
+                    "top_k": min(int(args.get("top_k", 8)), 20),
+                })
+                return json.dumps(data)
+
+            elif tool_name == "retaindb_context":
+                query = args.get("query", "")
+                if not query:
+                    return json.dumps({"error": "query is required"})
+                data = self._api("POST", "/v1/recall", json={
+                    "project": self._project,
+                    "user_id": self._user_id,
+                    "query": query,
+                    "top_k": 5,
+                })
+                return json.dumps(data)
+
+            elif tool_name == "retaindb_remember":
+                content = args.get("content", "")
+                if not content:
+                    return json.dumps({"error": "content is required"})
+                data = self._api("POST", "/v1/remember", json={
+                    "project": self._project,
+                    "user_id": self._user_id,
+                    "content": content,
+                    "memory_type": args.get("memory_type", "fact"),
+                    "importance": float(args.get("importance", 0.5)),
+                })
+                return json.dumps(data)
+
+            elif tool_name == "retaindb_forget":
+                memory_id = args.get("memory_id", "")
+                if not memory_id:
+                    return json.dumps({"error": "memory_id is required"})
+                data = self._api("DELETE", f"/v1/memory/{memory_id}")
+                return json.dumps(data)
+
+            return json.dumps({"error": f"Unknown tool: {tool_name}"})
+        except Exception as e:
+            return json.dumps({"error": str(e)})
+
+    def on_memory_write(self, action: str, target: str, content: str) -> None:
+        if action == "add":
+            try:
+                self._api("POST", "/v1/remember", json={
+                    "project": self._project,
+                    "user_id": self._user_id,
+                    "content": content,
+                    "memory_type": "preference" if target == "user" else "fact",
+                })
+            except Exception as e:
+                logger.debug("RetainDB memory bridge failed: %s", e)
+
+    def shutdown(self) -> None:
+        for t in (self._prefetch_thread, self._sync_thread):
+            if t and t.is_alive():
+                t.join(timeout=5.0)
+
+
+def register(ctx) -> None:
+    """Register RetainDB as a memory provider plugin."""
+    ctx.register_memory_provider(RetainDBMemoryProvider())
@@ -0,0 +1,7 @@
+name: retaindb
+version: 1.0.0
+description: "RetainDB — cloud memory API with hybrid search and 7 memory types."
+pip_dependencies:
+  - requests
+requires_env:
+  - RETAINDB_API_KEY
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"

 [project]
 name = "hermes-agent"
-version = "0.6.0"
+version = "0.7.0"
 description = "The self-improving AI agent — creates skills from experience, improves them during use, and runs anywhere"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -39,7 +39,7 @@ dependencies = [
 [project.optional-dependencies]
 modal = ["modal>=1.0.0,<2"]
 daytona = ["daytona>=0.148.0,<1"]
-dev = ["pytest>=9.0.2,<10", "pytest-asyncio>=1.3.0,<2", "pytest-xdist>=3.0,<4", "mcp>=1.2.0,<2"]
+dev = ["debugpy>=1.8.0,<2", "pytest>=9.0.2,<10", "pytest-asyncio>=1.3.0,<2", "pytest-xdist>=3.0,<4", "mcp>=1.2.0,<2"]
 messaging = ["python-telegram-bot>=22.6,<23", "discord.py[voice]>=2.7.1,<3", "aiohttp>=3.13.3,<4", "slack-bolt>=1.18.0,<2", "slack-sdk>=3.27.0,<4"]
 cron = ["croniter>=6.0.0,<7"]
 slack = ["slack-bolt>=1.18.0,<2", "slack-sdk>=3.27.0,<4"]
@@ -76,7 +76,10 @@ all = [
  "hermes-agent[modal]",
  "hermes-agent[daytona]",
  "hermes-agent[messaging]",
-  "hermes-agent[matrix]",
+  # matrix excluded: python-olm (required by matrix-nio[e2e]) is upstream-broken
+  # on modern macOS (archived libolm, C++ errors with Clang 21+). Including it
+  # here causes the entire [all] install to fail, dropping all other extras.
+  # Users who need Matrix can install manually: pip install 'hermes-agent[matrix]'
  "hermes-agent[cron]",
  "hermes-agent[cli]",
  "hermes-agent[dev]",
@@ -102,7 +105,7 @@ hermes-acp = "acp_adapter.entry:main"
 py-modules = ["run_agent", "model_tools", "toolsets", "batch_runner", "trajectory_compressor", "toolset_distributions", "cli", "hermes_constants", "hermes_state", "hermes_time", "rl_cli", "utils"]

 [tool.setuptools.packages.find]
-include = ["agent", "tools", "tools.*", "hermes_cli", "gateway", "gateway.*", "cron", "honcho_integration", "acp_adapter"]
+include = ["agent", "tools", "tools.*", "hermes_cli", "gateway", "gateway.*", "cron", "acp_adapter", "plugins", "plugins.*"]

 [tool.pytest.ini_options]
 testpaths = ["tests"]
@@ -15,6 +15,7 @@ requests
 jinja2
 pydantic>=2.0
 PyJWT[crypto]
+debugpy

 # Web tools
 firecrawl-py
@@ -62,6 +62,33 @@ function formatOutgoingMessage(message) {
  return REPLY_PREFIX ? `${REPLY_PREFIX}${message}` : message;
 }

+function normalizeWhatsAppId(value) {
+  if (!value) return '';
+  return String(value).replace(':', '@');
+}
+
+function getMessageContent(msg) {
+  const content = msg?.message || {};
+  if (content.ephemeralMessage?.message) return content.ephemeralMessage.message;
+  if (content.viewOnceMessage?.message) return content.viewOnceMessage.message;
+  if (content.viewOnceMessageV2?.message) return content.viewOnceMessageV2.message;
+  if (content.documentWithCaptionMessage?.message) return content.documentWithCaptionMessage.message;
+  if (content.templateMessage?.hydratedTemplate) return content.templateMessage.hydratedTemplate;
+  if (content.buttonsMessage) return content.buttonsMessage;
+  if (content.listMessage) return content.listMessage;
+  return content;
+}
+
+function getContextInfo(messageContent) {
+  if (!messageContent || typeof messageContent !== 'object') return {};
+  for (const value of Object.values(messageContent)) {
+    if (value && typeof value === 'object' && value.contextInfo) {
+      return value.contextInfo;
+    }
+  }
+  return {};
+}
+
 mkdirSync(SESSION_DIR, { recursive: true });

 // Build LID → phone reverse map from session files (lid-mapping-{phone}.json)
@@ -157,6 +184,11 @@ async function startSocket() {
    // than 'notify'. Accept both and filter agent echo-backs below.
    if (type !== 'notify' && type !== 'append') return;

+    const botIds = Array.from(new Set([
+      normalizeWhatsAppId(sock.user?.id),
+      normalizeWhatsAppId(sock.user?.lid),
+    ].filter(Boolean)));
+
    for (const msg of messages) {
      if (!msg.message) continue;

@@ -200,23 +232,28 @@ async function startSocket() {
        continue;
      }

+      const messageContent = getMessageContent(msg);
+      const contextInfo = getContextInfo(messageContent);
+      const mentionedIds = Array.from(new Set((contextInfo?.mentionedJid || []).map(normalizeWhatsAppId).filter(Boolean)));
+      const quotedParticipant = normalizeWhatsAppId(contextInfo?.participant || contextInfo?.remoteJid || '');
+
      // Extract message body
      let body = '';
      let hasMedia = false;
      let mediaType = '';
      const mediaUrls = [];

-      if (msg.message.conversation) {
-        body = msg.message.conversation;
-      } else if (msg.message.extendedTextMessage?.text) {
-        body = msg.message.extendedTextMessage.text;
-      } else if (msg.message.imageMessage) {
-        body = msg.message.imageMessage.caption || '';
+      if (messageContent.conversation) {
+        body = messageContent.conversation;
+      } else if (messageContent.extendedTextMessage?.text) {
+        body = messageContent.extendedTextMessage.text;
+      } else if (messageContent.imageMessage) {
+        body = messageContent.imageMessage.caption || '';
        hasMedia = true;
        mediaType = 'image';
        try {
          const buf = await downloadMediaMessage(msg, 'buffer', {}, { logger, reuploadRequest: sock.updateMediaMessage });
-          const mime = msg.message.imageMessage.mimetype || 'image/jpeg';
+          const mime = messageContent.imageMessage.mimetype || 'image/jpeg';
          const extMap = { 'image/jpeg': '.jpg', 'image/png': '.png', 'image/webp': '.webp', 'image/gif': '.gif' };
          const ext = extMap[mime] || '.jpg';
          mkdirSync(IMAGE_CACHE_DIR, { recursive: true });
@@ -226,13 +263,13 @@ async function startSocket() {
        } catch (err) {
          console.error('[bridge] Failed to download image:', err.message);
        }
-      } else if (msg.message.videoMessage) {
-        body = msg.message.videoMessage.caption || '';
+      } else if (messageContent.videoMessage) {
+        body = messageContent.videoMessage.caption || '';
        hasMedia = true;
        mediaType = 'video';
        try {
          const buf = await downloadMediaMessage(msg, 'buffer', {}, { logger, reuploadRequest: sock.updateMediaMessage });
-          const mime = msg.message.videoMessage.mimetype || 'video/mp4';
+          const mime = messageContent.videoMessage.mimetype || 'video/mp4';
          const ext = mime.includes('mp4') ? '.mp4' : '.mkv';
          mkdirSync(DOCUMENT_CACHE_DIR, { recursive: true });
          const filePath = path.join(DOCUMENT_CACHE_DIR, `vid_${randomBytes(6).toString('hex')}${ext}`);
@@ -241,11 +278,11 @@ async function startSocket() {
        } catch (err) {
          console.error('[bridge] Failed to download video:', err.message);
        }
-      } else if (msg.message.audioMessage || msg.message.pttMessage) {
+      } else if (messageContent.audioMessage || messageContent.pttMessage) {
        hasMedia = true;
-        mediaType = msg.message.pttMessage ? 'ptt' : 'audio';
+        mediaType = messageContent.pttMessage ? 'ptt' : 'audio';
        try {
-          const audioMsg = msg.message.pttMessage || msg.message.audioMessage;
+          const audioMsg = messageContent.pttMessage || messageContent.audioMessage;
          const buf = await downloadMediaMessage(msg, 'buffer', {}, { logger, reuploadRequest: sock.updateMediaMessage });
          const mime = audioMsg.mimetype || 'audio/ogg';
          const ext = mime.includes('ogg') ? '.ogg' : mime.includes('mp4') ? '.m4a' : '.ogg';
@@ -256,11 +293,11 @@ async function startSocket() {
        } catch (err) {
          console.error('[bridge] Failed to download audio:', err.message);
        }
-      } else if (msg.message.documentMessage) {
-        body = msg.message.documentMessage.caption || '';
+      } else if (messageContent.documentMessage) {
+        body = messageContent.documentMessage.caption || '';
        hasMedia = true;
        mediaType = 'document';
-        const fileName = msg.message.documentMessage.fileName || 'document';
+        const fileName = messageContent.documentMessage.fileName || 'document';
        try {
          const buf = await downloadMediaMessage(msg, 'buffer', {}, { logger, reuploadRequest: sock.updateMediaMessage });
          mkdirSync(DOCUMENT_CACHE_DIR, { recursive: true });
@@ -309,6 +346,9 @@ async function startSocket() {
        hasMedia,
        mediaType,
        mediaUrls,
+        mentionedIds,
+        quotedParticipant,
+        botIds,
        timestamp: msg.messageTimestamp,
      };

@@ -1,6 +1,10 @@
 ---
 name: youtube-content
-description: Fetch YouTube video transcripts and transform them into structured content (chapters, summaries, threads, blog posts).
+description: >
+  Fetch YouTube video transcripts and transform them into structured content
+  (chapters, summaries, threads, blog posts). Use when the user shares a YouTube
+  URL or video link, asks to summarize a video, requests a transcript, or wants
+  to extract and reformat content from any YouTube video.
 ---

 # YouTube Content Tool
@@ -13,59 +17,56 @@ Extract transcripts from YouTube videos and convert them into useful formats.
 pip install youtube-transcript-api
 ```

-## Helper script
+## Helper Script

-This skill includes `fetch_transcript.py` — use it to fetch transcripts quickly:
+`SKILL_DIR` is the directory containing this SKILL.md file. The script accepts any standard YouTube URL format, short links (youtu.be), shorts, embeds, live links, or a raw 11-character video ID.

 ```bash
 # JSON output with metadata
 python3 SKILL_DIR/scripts/fetch_transcript.py "https://youtube.com/watch?v=VIDEO_ID"

+# Plain text (good for piping into further processing)
+python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --text-only
+
 # With timestamps
-python3 SKILL_DIR/scripts/fetch_transcript.py "https://youtube.com/watch?v=VIDEO_ID" --timestamps
+python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --timestamps

-# Plain text output (good for piping into further processing)
-python3 SKILL_DIR/scripts/fetch_transcript.py "https://youtube.com/watch?v=VIDEO_ID" --text-only
-
-# Specific language with fallback
-python3 SKILL_DIR/scripts/fetch_transcript.py "https://youtube.com/watch?v=VIDEO_ID" --language tr,en
-
-# Timestamped plain text
-python3 SKILL_DIR/scripts/fetch_transcript.py "https://youtube.com/watch?v=VIDEO_ID" --text-only --timestamps
+# Specific language with fallback chain
+python3 SKILL_DIR/scripts/fetch_transcript.py "URL" --language tr,en
 ```

-`SKILL_DIR` is the directory containing this SKILL.md file.
-
-## URL formats supported
-
-The script accepts any of these formats (or a raw 11-character video ID):
-
- `https://www.youtube.com/watch?v=VIDEO_ID`
- `https://youtu.be/VIDEO_ID`
- `https://youtube.com/shorts/VIDEO_ID`
- `https://youtube.com/embed/VIDEO_ID`
- `https://youtube.com/live/VIDEO_ID`
-
-## Output formats
+## Output Formats

 After fetching the transcript, format it based on what the user asks for:

- **Chapters**: Group by topic shifts, output timestamped chapter list (`00:00 Introduction`, `03:45 Main Topic`, etc.)
+- **Chapters**: Group by topic shifts, output timestamped chapter list
 - **Summary**: Concise 5-10 sentence overview of the entire video
 - **Chapter summaries**: Chapters with a short paragraph summary for each
 - **Thread**: Twitter/X thread format — numbered posts, each under 280 chars
 - **Blog post**: Full article with title, sections, and key takeaways
 - **Quotes**: Notable quotes with timestamps

+### Example — Chapters Output
+
+```
+00:00 Introduction — host opens with the problem statement
+03:45 Background — prior work and why existing solutions fall short
+12:20 Core method — walkthrough of the proposed approach
+24:10 Results — benchmark comparisons and key takeaways
+31:55 Q&A — audience questions on scalability and next steps
+```
+
 ## Workflow

-1. Fetch the transcript using the helper script
-2. If the transcript is very long (>50K chars), summarize in chunks
-3. Transform into the requested output format using your own reasoning
+1. **Fetch** the transcript using the helper script with `--text-only --timestamps`.
+2. **Validate**: confirm the output is non-empty and in the expected language. If empty, retry without `--language` to get any available transcript. If still empty, tell the user the video likely has transcripts disabled.
+3. **Chunk if needed**: if the transcript exceeds ~50K characters, split into overlapping chunks (~40K with 2K overlap) and summarize each chunk before merging.
+4. **Transform** into the requested output format. If the user did not specify a format, default to a summary.
+5. **Verify**: re-read the transformed output to check for coherence, correct timestamps, and completeness before presenting.

-## Error handling
+## Error Handling

- **Transcript disabled**: Some videos have transcripts turned off — tell the user
- **Private/unavailable**: The API will raise an error — relay it clearly
- **No matching language**: Try without specifying a language to get whatever's available
- **Dependency missing**: Run `pip install youtube-transcript-api` first
+- **Transcript disabled**: tell the user; suggest they check if subtitles are available on the video page.
+- **Private/unavailable video**: relay the error and ask the user to verify the URL.
+- **No matching language**: retry without `--language` to fetch any available transcript, then note the actual language to the user.
+- **Dependency missing**: run `pip install youtube-transcript-api` and retry.
@@ -48,7 +48,11 @@ def format_timestamp(seconds: float) -> str:


 def fetch_transcript(video_id: str, languages: list = None):
-    """Fetch transcript segments from YouTube."""
+    """Fetch transcript segments from YouTube.
+
+    Returns a list of dicts with 'text', 'start', and 'duration' keys.
+    Compatible with youtube-transcript-api v1.x.
+    """
    try:
        from youtube_transcript_api import YouTubeTranscriptApi
    except ImportError:
@@ -56,9 +60,17 @@ def fetch_transcript(video_id: str, languages: list = None):
              file=sys.stderr)
        sys.exit(1)

+    api = YouTubeTranscriptApi()
    if languages:
-        return YouTubeTranscriptApi.get_transcript(video_id, languages=languages)
-    return YouTubeTranscriptApi.get_transcript(video_id)
+        result = api.fetch(video_id, languages=languages)
+    else:
+        result = api.fetch(video_id)
+
+    # v1.x returns FetchedTranscriptSnippet objects; normalize to dicts
+    return [
+        {"text": seg.text, "start": seg.start, "duration": seg.duration}
+        for seg in result
+    ]


 def main():
@@ -125,8 +125,9 @@ Should print `AUTHENTICATED`. Setup is complete — token refreshes automaticall

 ### Notes

- Token is stored at `~/.hermes/google_token.json` and auto-refreshes.
- Pending OAuth session state/verifier are stored temporarily at `~/.hermes/google_oauth_pending.json` until exchange completes.
+- Token is stored at `google_token.json` under the active profile's `HERMES_HOME` and auto-refreshes.
+- Pending OAuth session state/verifier are stored temporarily at `google_oauth_pending.json` under the active profile's `HERMES_HOME` until exchange completes.
+- Hermes now refuses to overwrite a full Google Workspace token with a narrower re-auth token missing Gmail scopes, so one profile's partial consent cannot silently break email actions later.
 - To revoke: `$GSETUP --revoke`

 ## Usage
@@ -22,13 +22,14 @@ Usage:
 import argparse
 import base64
 import json
-import os
 import sys
 from datetime import datetime, timedelta, timezone
 from email.mime.text import MIMEText
 from pathlib import Path

-HERMES_HOME = Path(os.getenv("HERMES_HOME", Path.home() / ".hermes"))
+from hermes_constants import display_hermes_home, get_hermes_home
+
+HERMES_HOME = get_hermes_home()
 TOKEN_PATH = HERMES_HOME / "google_token.json"

 SCOPES = [
@@ -43,6 +44,18 @@ SCOPES = [
 ]


+def _missing_scopes() -> list[str]:
+    try:
+        payload = json.loads(TOKEN_PATH.read_text())
+    except Exception:
+        return []
+    raw = payload.get("scopes") or payload.get("scope")
+    if not raw:
+        return []
+    granted = {s.strip() for s in (raw.split() if isinstance(raw, str) else raw) if s.strip()}
+    return sorted(scope for scope in SCOPES if scope not in granted)
+
+
 def get_credentials():
    """Load and refresh credentials from token file."""
    if not TOKEN_PATH.exists():
@@ -60,6 +73,20 @@ def get_credentials():
    if not creds.valid:
        print("Token is invalid. Re-run setup.", file=sys.stderr)
        sys.exit(1)
+
+    missing_scopes = _missing_scopes()
+    if missing_scopes:
+        print(
+            "Token is valid but missing Google Workspace scopes required by this skill.",
+            file=sys.stderr,
+        )
+        for scope in missing_scopes:
+            print(f"  - {scope}", file=sys.stderr)
+        print(
+            f"Re-run setup.py from the active Hermes profile ({display_hermes_home()}) to restore full access.",
+            file=sys.stderr,
+        )
+        sys.exit(1)
    return creds


@@ -23,12 +23,13 @@ Agent workflow:

 import argparse
 import json
-import os
 import subprocess
 import sys
 from pathlib import Path

-HERMES_HOME = Path(os.getenv("HERMES_HOME", Path.home() / ".hermes"))
+from hermes_constants import display_hermes_home, get_hermes_home
+
+HERMES_HOME = get_hermes_home()
 TOKEN_PATH = HERMES_HOME / "google_token.json"
 CLIENT_SECRET_PATH = HERMES_HOME / "google_client_secret.json"
 PENDING_AUTH_PATH = HERMES_HOME / "google_oauth_pending.json"
@@ -52,6 +53,30 @@ REQUIRED_PACKAGES = ["google-api-python-client", "google-auth-oauthlib", "google
 REDIRECT_URI = "http://localhost:1"


+def _load_token_payload(path: Path = TOKEN_PATH) -> dict:
+    try:
+        return json.loads(path.read_text())
+    except Exception:
+        return {}
+
+
+def _missing_scopes_from_payload(payload: dict) -> list[str]:
+    raw = payload.get("scopes") or payload.get("scope")
+    if not raw:
+        return []
+    granted = {s.strip() for s in (raw.split() if isinstance(raw, str) else raw) if s.strip()}
+    return sorted(scope for scope in SCOPES if scope not in granted)
+
+
+def _format_missing_scopes(missing_scopes: list[str]) -> str:
+    bullets = "\n".join(f"  - {scope}" for scope in missing_scopes)
+    return (
+        "Token is valid but missing required Google Workspace scopes:\n"
+        f"{bullets}\n"
+        "Run the Google Workspace setup again from this same Hermes profile to refresh consent."
+    )
+
+
 def install_deps():
    """Install Google API packages if missing. Returns True on success."""
    try:
@@ -102,7 +127,12 @@ def check_auth():
        print(f"TOKEN_CORRUPT: {e}")
        return False

+    payload = _load_token_payload(TOKEN_PATH)
    if creds.valid:
+        missing_scopes = _missing_scopes_from_payload(payload)
+        if missing_scopes:
+            print(f"AUTH_SCOPE_MISMATCH: {_format_missing_scopes(missing_scopes)}")
+            return False
        print(f"AUTHENTICATED: Token valid at {TOKEN_PATH}")
        return True

@@ -110,6 +140,10 @@ def check_auth():
        try:
            creds.refresh(Request())
            TOKEN_PATH.write_text(creds.to_json())
+            missing_scopes = _missing_scopes_from_payload(_load_token_payload(TOKEN_PATH))
+            if missing_scopes:
+                print(f"AUTH_SCOPE_MISMATCH: {_format_missing_scopes(missing_scopes)}")
+                return False
            print(f"AUTHENTICATED: Token refreshed at {TOKEN_PATH}")
            return True
        except Exception as e:
@@ -249,9 +283,17 @@ def exchange_auth_code(code: str):
        sys.exit(1)

    creds = flow.credentials
-    TOKEN_PATH.write_text(creds.to_json())
+    token_payload = json.loads(creds.to_json())
+    missing_scopes = _missing_scopes_from_payload(token_payload)
+    if missing_scopes:
+        print(f"ERROR: Refusing to save incomplete Google Workspace token. {_format_missing_scopes(missing_scopes)}")
+        print(f"Existing token at {TOKEN_PATH} was left unchanged.")
+        sys.exit(1)
+
+    TOKEN_PATH.write_text(json.dumps(token_payload, indent=2))
    PENDING_AUTH_PATH.unlink(missing_ok=True)
    print(f"OK: Authenticated. Token saved to {TOKEN_PATH}")
+    print(f"Profile-scoped token location: {display_hermes_home()}/google_token.json")


 def revoke():
@@ -1,940 +0,0 @@
---
-name: ml-paper-writing
-description: Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
-version: 1.0.0
-author: Orchestra Research
-license: MIT
-dependencies: [semanticscholar, arxiv, habanero, requests]
-metadata:
-  hermes:
-    tags: [Academic Writing, NeurIPS, ICML, ICLR, ACL, AAAI, COLM, LaTeX, Paper Writing, Citations, Research]
-
---
-
-# ML Paper Writing for Top AI Conferences
-
-Expert-level guidance for writing publication-ready papers targeting **NeurIPS, ICML, ICLR, ACL, AAAI, and COLM**. This skill combines writing philosophy from top researchers (Nanda, Farquhar, Karpathy, Lipton, Steinhardt) with practical tools: LaTeX templates, citation verification APIs, and conference checklists.
-
-## Core Philosophy: Collaborative Writing
-
-**Paper writing is collaborative, but Claude should be proactive in delivering drafts.**
-
-The typical workflow starts with a research repository containing code, results, and experimental artifacts. Claude's role is to:
-
-1. **Understand the project** by exploring the repo, results, and existing documentation
-2. **Deliver a complete first draft** when confident about the contribution
-3. **Search literature** using web search and APIs to find relevant citations
-4. **Refine through feedback cycles** when the scientist provides input
-5. **Ask for clarification** only when genuinely uncertain about key decisions
-
-**Key Principle**: Be proactive. If the repo and results are clear, deliver a full draft. Don't block waiting for feedback on every section—scientists are busy. Produce something concrete they can react to, then iterate based on their response.
-
---
-
-## ⚠️ CRITICAL: Never Hallucinate Citations
-
-**This is the most important rule in academic writing with AI assistance.**
-
-### The Problem
-AI-generated citations have a **~40% error rate**. Hallucinated references—papers that don't exist, wrong authors, incorrect years, fabricated DOIs—are a serious form of academic misconduct that can result in desk rejection or retraction.
-
-### The Rule
-**NEVER generate BibTeX entries from memory. ALWAYS fetch programmatically.**
-
-| Action | ✅ Correct | ❌ Wrong |
-|--------|-----------|----------|
-| Adding a citation | Search API → verify → fetch BibTeX | Write BibTeX from memory |
-| Uncertain about a paper | Mark as `[CITATION NEEDED]` | Guess the reference |
-| Can't find exact paper | Note: "placeholder - verify" | Invent similar-sounding paper |
-
-### When You Can't Verify a Citation
-
-If you cannot programmatically verify a citation, you MUST:
-
-```latex
-% EXPLICIT PLACEHOLDER - requires human verification
-\cite{PLACEHOLDER_author2024_verify_this}  % TODO: Verify this citation exists
-```
-
-**Always tell the scientist**: "I've marked [X] citations as placeholders that need verification. I could not confirm these papers exist."
-
-### Recommended: Install Exa MCP for Paper Search
-
-For the best paper search experience, install **Exa MCP** which provides real-time academic search:
-
-**Claude Code:**
-```bash
-claude mcp add exa -- npx -y mcp-remote "https://mcp.exa.ai/mcp"
-```
-
-**Cursor / VS Code** (add to MCP settings):
-```json
-{
-  "mcpServers": {
-    "exa": {
-      "type": "http",
-      "url": "https://mcp.exa.ai/mcp"
-    }
-  }
-}
-```
-
-Exa MCP enables searches like:
- "Find papers on RLHF for language models published after 2023"
- "Search for transformer architecture papers by Vaswani"
- "Get recent work on sparse autoencoders for interpretability"
-
-Then verify results with Semantic Scholar API and fetch BibTeX via DOI.
-
---
-
-## Workflow 0: Starting from a Research Repository
-
-When beginning paper writing, start by understanding the project:
-
-```
-Project Understanding:
- [ ] Step 1: Explore the repository structure
- [ ] Step 2: Read README, existing docs, and key results
- [ ] Step 3: Identify the main contribution with the scientist
- [ ] Step 4: Find papers already cited in the codebase
- [ ] Step 5: Search for additional relevant literature
- [ ] Step 6: Outline the paper structure together
- [ ] Step 7: Draft sections iteratively with feedback
-```
-
-**Step 1: Explore the Repository**
-
-```bash
-# Understand project structure
-ls -la
-find . -name "*.py" | head -20
-find . -name "*.md" -o -name "*.txt" | xargs grep -l -i "result\|conclusion\|finding"
-```
-
-Look for:
- `README.md` - Project overview and claims
- `results/`, `outputs/`, `experiments/` - Key findings
- `configs/` - Experimental settings
- Existing `.bib` files or citation references
- Any draft documents or notes
-
-**Step 2: Identify Existing Citations**
-
-Check for papers already referenced in the codebase:
-
-```bash
-# Find existing citations
-grep -r "arxiv\|doi\|cite" --include="*.md" --include="*.bib" --include="*.py"
-find . -name "*.bib"
-```
-
-These are high-signal starting points for Related Work—the scientist has already deemed them relevant.
-
-**Step 3: Clarify the Contribution**
-
-Before writing, explicitly confirm with the scientist:
-
-> "Based on my understanding of the repo, the main contribution appears to be [X].
-> The key results show [Y]. Is this the framing you want for the paper,
-> or should we emphasize different aspects?"
-
-**Never assume the narrative—always verify with the human.**
-
-**Step 4: Search for Additional Literature**
-
-Use web search to find relevant papers:
-
-```
-Search queries to try:
- "[main technique] + [application domain]"
- "[baseline method] comparison"
- "[problem name] state-of-the-art"
- Author names from existing citations
-```
-
-Then verify and retrieve BibTeX using the citation workflow below.
-
-**Step 5: Deliver a First Draft**
-
-**Be proactive—deliver a complete draft rather than asking permission for each section.**
-
-If the repo provides clear results and the contribution is apparent:
-1. Write the full first draft end-to-end
-2. Present the complete draft for feedback
-3. Iterate based on scientist's response
-
-If genuinely uncertain about framing or major claims:
-1. Draft what you can confidently
-2. Flag specific uncertainties: "I framed X as the main contribution—let me know if you'd prefer to emphasize Y instead"
-3. Continue with the draft rather than blocking
-
-**Questions to include with the draft** (not before):
- "I emphasized X as the main contribution—adjust if needed"
- "I highlighted results A, B, C—let me know if others are more important"
- "Related work section includes [papers]—add any I missed"
-
---
-
-## When to Use This Skill
-
-Use this skill when:
- **Starting from a research repo** to write a paper
- **Drafting or revising** specific sections
- **Finding and verifying citations** for related work
- **Formatting** for conference submission
- **Resubmitting** to a different venue (format conversion)
- **Iterating** on drafts with scientist feedback
-
-**Always remember**: First drafts are starting points for discussion, not final outputs.
-
---
-
-## Balancing Proactivity and Collaboration
-
-**Default: Be proactive. Deliver drafts, then iterate.**
-
-| Confidence Level | Action |
-|-----------------|--------|
-| **High** (clear repo, obvious contribution) | Write full draft, deliver, iterate on feedback |
-| **Medium** (some ambiguity) | Write draft with flagged uncertainties, continue |
-| **Low** (major unknowns) | Ask 1-2 targeted questions, then draft |
-
-**Draft first, ask with the draft** (not before):
-
-| Section | Draft Autonomously | Flag With Draft |
-|---------|-------------------|-----------------|
-| Abstract | Yes | "Framed contribution as X—adjust if needed" |
-| Introduction | Yes | "Emphasized problem Y—correct if wrong" |
-| Methods | Yes | "Included details A, B, C—add missing pieces" |
-| Experiments | Yes | "Highlighted results 1, 2, 3—reorder if needed" |
-| Related Work | Yes | "Cited papers X, Y, Z—add any I missed" |
-
-**Only block for input when:**
- Target venue is unclear (affects page limits, framing)
- Multiple contradictory framings seem equally valid
- Results seem incomplete or inconsistent
- Explicit request to review before continuing
-
-**Don't block for:**
- Word choice decisions
- Section ordering
- Which specific results to show (make a choice, flag it)
- Citation completeness (draft with what you find, note gaps)
-
---
-
-## The Narrative Principle
-
-**The single most critical insight**: Your paper is not a collection of experiments—it's a story with one clear contribution supported by evidence.
-
-Every successful ML paper centers on what Neel Nanda calls "the narrative": a short, rigorous, evidence-based technical story with a takeaway readers care about.
-
-**Three Pillars (must be crystal clear by end of introduction):**
-
-| Pillar | Description | Example |
-|--------|-------------|---------|
-| **The What** | 1-3 specific novel claims within cohesive theme | "We prove that X achieves Y under condition Z" |
-| **The Why** | Rigorous empirical evidence supporting claims | Strong baselines, experiments distinguishing hypotheses |
-| **The So What** | Why readers should care | Connection to recognized community problems |
-
-**If you cannot state your contribution in one sentence, you don't yet have a paper.**
-
---
-
-## Paper Structure Workflow
-
-### Workflow 1: Writing a Complete Paper (Iterative)
-
-Copy this checklist and track progress. **Each step involves drafting → feedback → revision:**
-
-```
-Paper Writing Progress:
- [ ] Step 1: Define the one-sentence contribution (with scientist)
- [ ] Step 2: Draft Figure 1 → get feedback → revise
- [ ] Step 3: Draft abstract → get feedback → revise
- [ ] Step 4: Draft introduction → get feedback → revise
- [ ] Step 5: Draft methods → get feedback → revise
- [ ] Step 6: Draft experiments → get feedback → revise
- [ ] Step 7: Draft related work → get feedback → revise
- [ ] Step 8: Draft limitations → get feedback → revise
- [ ] Step 9: Complete paper checklist (required)
- [ ] Step 10: Final review cycle and submission
-```
-
-**Step 1: Define the One-Sentence Contribution**
-
-**This step requires explicit confirmation from the scientist.**
-
-Before writing anything, articulate and verify:
- What is the single thing your paper contributes?
- What was not obvious or present before your work?
-
-> "I propose framing the contribution as: '[one sentence]'. Does this capture
-> what you see as the main takeaway? Should we adjust the emphasis?"
-
-**Step 2: Draft Figure 1**
-
-Figure 1 deserves special attention—many readers skip directly to it.
- Convey core idea, approach, or most compelling result
- Use vector graphics (PDF/EPS for plots)
- Write captions that stand alone without main text
- Ensure readability in black-and-white (8% of men have color vision deficiency)
-
-**Step 3: Write Abstract (5-Sentence Formula)**
-
-From Sebastian Farquhar (DeepMind):
-
-```
-1. What you achieved: "We introduce...", "We prove...", "We demonstrate..."
-2. Why this is hard and important
-3. How you do it (with specialist keywords for discoverability)
-4. What evidence you have
-5. Your most remarkable number/result
-```
-
-**Delete** generic openings like "Large language models have achieved remarkable success..."
-
-**Step 4: Write Introduction (1-1.5 pages max)**
-
-Must include:
- 2-4 bullet contribution list (max 1-2 lines each in two-column format)
- Clear problem statement
- Brief approach overview
- Methods should start by page 2-3 maximum
-
-**Step 5: Methods Section**
-
-Enable reimplementation:
- Conceptual outline or pseudocode
- All hyperparameters listed
- Architectural details sufficient for reproduction
- Present final design decisions; ablations go in experiments
-
-**Step 6: Experiments Section**
-
-For each experiment, explicitly state:
- What claim it supports
- How it connects to main contribution
- Experimental setting (details in appendix)
- What to observe: "the blue line shows X, which demonstrates Y"
-
-Requirements:
- Error bars with methodology (standard deviation vs standard error)
- Hyperparameter search ranges
- Compute infrastructure (GPU type, total hours)
- Seed-setting methods
-
-**Step 7: Related Work**
-
-Organize methodologically, not paper-by-paper:
-
-**Good:** "One line of work uses Floogledoodle's assumption [refs] whereas we use Doobersnoddle's assumption because..."
-
-**Bad:** "Snap et al. introduced X while Crackle et al. introduced Y."
-
-Cite generously—reviewers likely authored relevant papers.
-
-**Step 8: Limitations Section (REQUIRED)**
-
-All major conferences require this. Counter-intuitively, honesty helps:
- Reviewers are instructed not to penalize honest limitation acknowledgment
- Pre-empt criticisms by identifying weaknesses first
- Explain why limitations don't undermine core claims
-
-**Step 9: Paper Checklist**
-
-NeurIPS, ICML, and ICLR all require paper checklists. See [references/checklists.md](references/checklists.md).
-
---
-
-## Writing Philosophy for Top ML Conferences
-
-**This section distills the most important writing principles from leading ML researchers.** These aren't optional style suggestions—they're what separates accepted papers from rejected ones.
-
-> "A paper is a short, rigorous, evidence-based technical story with a takeaway readers care about." — Neel Nanda
-
-### The Sources Behind This Guidance
-
-This skill synthesizes writing philosophy from researchers who have published extensively at top venues:
-
-| Source | Key Contribution | Link |
-|--------|-----------------|------|
-| **Neel Nanda** (Google DeepMind) | The Narrative Principle, What/Why/So What framework | [How to Write ML Papers](https://www.alignmentforum.org/posts/eJGptPbbFPZGLpjsp/highly-opinionated-advice-on-how-to-write-ml-papers) |
-| **Sebastian Farquhar** (DeepMind) | 5-sentence abstract formula | [How to Write ML Papers](https://sebastianfarquhar.com/on-research/2024/11/04/how_to_write_ml_papers/) |
-| **Gopen & Swan** | 7 principles of reader expectations | [Science of Scientific Writing](https://cseweb.ucsd.edu/~swanson/papers/science-of-writing.pdf) |
-| **Zachary Lipton** | Word choice, eliminating hedging | [Heuristics for Scientific Writing](https://www.approximatelycorrect.com/2018/01/29/heuristics-technical-scientific-writing-machine-learning-perspective/) |
-| **Jacob Steinhardt** (UC Berkeley) | Precision, consistent terminology | [Writing Tips](https://bounded-regret.ghost.io/) |
-| **Ethan Perez** (Anthropic) | Micro-level clarity tips | [Easy Paper Writing Tips](https://ethanperez.net/easy-paper-writing-tips/) |
-| **Andrej Karpathy** | Single contribution focus | Various lectures |
-
-**For deeper dives into any of these, see:**
- [references/writing-guide.md](references/writing-guide.md) - Full explanations with examples
- [references/sources.md](references/sources.md) - Complete bibliography
-
-### Time Allocation (From Neel Nanda)
-
-Spend approximately **equal time** on each of:
-1. The abstract
-2. The introduction
-3. The figures
-4. Everything else combined
-
-**Why?** Most reviewers form judgments before reaching your methods. Readers encounter your paper as: **title → abstract → introduction → figures → maybe the rest.**
-
-### Writing Style Guidelines
-
-#### Sentence-Level Clarity (Gopen & Swan's 7 Principles)
-
-These principles are based on how readers actually process prose. Violating them forces readers to spend cognitive effort on structure rather than content.
-
-| Principle | Rule | Example |
-|-----------|------|---------|
-| **Subject-verb proximity** | Keep subject and verb close | ❌ "The model, which was trained on..., achieves" → ✅ "The model achieves... after training on..." |
-| **Stress position** | Place emphasis at sentence ends | ❌ "Accuracy improves by 15% when using attention" → ✅ "When using attention, accuracy improves by **15%**" |
-| **Topic position** | Put context first, new info after | ✅ "Given these constraints, we propose..." |
-| **Old before new** | Familiar info → unfamiliar info | Link backward, then introduce new |
-| **One unit, one function** | Each paragraph makes one point | Split multi-point paragraphs |
-| **Action in verb** | Use verbs, not nominalizations | ❌ "We performed an analysis" → ✅ "We analyzed" |
-| **Context before new** | Set stage before presenting | Explain before showing equation |
-
-**Full 7 principles with detailed examples:** See [references/writing-guide.md](references/writing-guide.md#the-7-principles-of-reader-expectations)
-
-#### Micro-Level Tips (Ethan Perez)
-
-These small changes accumulate into significantly clearer prose:
-
- **Minimize pronouns**: ❌ "This shows..." → ✅ "This result shows..."
- **Verbs early**: Position verbs near sentence start
- **Unfold apostrophes**: ❌ "X's Y" → ✅ "The Y of X" (when awkward)
- **Delete filler words**: "actually," "a bit," "very," "really," "basically," "quite," "essentially"
-
-**Full micro-tips with examples:** See [references/writing-guide.md](references/writing-guide.md#micro-level-writing-tips)
-
-#### Word Choice (Zachary Lipton)
-
- **Be specific**: ❌ "performance" → ✅ "accuracy" or "latency" (say what you mean)
- **Eliminate hedging**: Drop "may" and "can" unless genuinely uncertain
- **Avoid incremental vocabulary**: ❌ "combine," "modify," "expand" → ✅ "develop," "propose," "introduce"
- **Delete intensifiers**: ❌ "provides *very* tight approximation" → ✅ "provides tight approximation"
-
-#### Precision Over Brevity (Jacob Steinhardt)
-
- **Consistent terminology**: Different terms for same concept creates confusion. Pick one and stick with it.
- **State assumptions formally**: Before theorems, list all assumptions explicitly
- **Intuition + rigor**: Provide intuitive explanations alongside formal proofs
-
-### What Reviewers Actually Read
-
-Understanding reviewer behavior helps prioritize your effort:
-
-| Paper Section | % Reviewers Who Read | Implication |
-|---------------|---------------------|-------------|
-| Abstract | 100% | Must be perfect |
-| Introduction | 90%+ (skimmed) | Front-load contribution |
-| Figures | Examined before methods | Figure 1 is critical |
-| Methods | Only if interested | Don't bury the lede |
-| Appendix | Rarely | Put only supplementary details |
-
-**Bottom line**: If your abstract and intro don't hook reviewers, they may never read your brilliant methods section.
-
---
-
-## Conference Requirements Quick Reference
-
-| Conference | Page Limit | Extra for Camera-Ready | Key Requirement |
-|------------|------------|------------------------|-----------------|
-| **NeurIPS 2025** | 9 pages | +0 | Mandatory checklist, lay summary for accepted |
-| **ICML 2026** | 8 pages | +1 | Broader Impact Statement required |
-| **ICLR 2026** | 9 pages | +1 | LLM disclosure required, reciprocal reviewing |
-| **ACL 2025** | 8 pages (long) | varies | Limitations section mandatory |
-| **AAAI 2026** | 7 pages | +1 | Strict style file adherence |
-| **COLM 2025** | 9 pages | +1 | Focus on language models |
-
-**Universal Requirements:**
- Double-blind review (anonymize submissions)
- References don't count toward page limit
- Appendices unlimited but reviewers not required to read
- LaTeX required for all venues
-
-**LaTeX Templates:** See [templates/](templates/) directory for all conference templates.
-
---
-
-## Using LaTeX Templates Properly
-
-### Workflow 4: Starting a New Paper from Template
-
-**Always copy the entire template directory first, then write within it.**
-
-```
-Template Setup Checklist:
- [ ] Step 1: Copy entire template directory to new project
- [ ] Step 2: Verify template compiles as-is (before any changes)
- [ ] Step 3: Read the template's example content to understand structure
- [ ] Step 4: Replace example content section by section
- [ ] Step 5: Keep template comments/examples as reference until done
- [ ] Step 6: Clean up template artifacts only at the end
-```
-
-**Step 1: Copy the Full Template**
-
-```bash
-# Create your paper directory with the complete template
-cp -r templates/neurips2025/ ~/papers/my-new-paper/
-cd ~/papers/my-new-paper/
-
-# Verify structure is complete
-ls -la
-# Should see: main.tex, neurips.sty, Makefile, etc.
-```
-
-**⚠️ IMPORTANT**: Copy the ENTIRE directory, not just `main.tex`. Templates include:
- Style files (`.sty`) - required for compilation
- Bibliography styles (`.bst`) - required for references
- Example content - useful as reference
- Makefiles - for easy compilation
-
-**Step 2: Verify Template Compiles First**
-
-Before making ANY changes, compile the template as-is:
-
-```bash
-# Using latexmk (recommended)
-latexmk -pdf main.tex
-
-# Or manual compilation
-pdflatex main.tex
-bibtex main
-pdflatex main.tex
-pdflatex main.tex
-```
-
-If the unmodified template doesn't compile, fix that first. Common issues:
- Missing TeX packages → install via `tlmgr install <package>`
- Wrong TeX distribution → use TeX Live (recommended)
-
-**Step 3: Keep Template Content as Reference**
-
-Don't immediately delete all example content. Instead:
-
-```latex
-% KEEP template examples commented out as you write
-% This shows you the expected format
-
-% Template example (keep for reference):
-% \begin{figure}[t]
-%   \centering
-%   \includegraphics[width=0.8\linewidth]{example-image}
-%   \caption{Template shows caption style}
-% \end{figure}
-
-% Your actual figure:
-\begin{figure}[t]
-  \centering
-  \includegraphics[width=0.8\linewidth]{your-figure.pdf}
-  \caption{Your caption following the same style.}
-\end{figure}
-```
-
-**Step 4: Replace Content Section by Section**
-
-Work through the paper systematically:
-
-```
-Replacement Order:
-1. Title and authors (anonymize for submission)
-2. Abstract
-3. Introduction
-4. Methods
-5. Experiments
-6. Related Work
-7. Conclusion
-8. References (your .bib file)
-9. Appendix
-```
-
-For each section:
-1. Read the template's example content
-2. Note any special formatting or macros used
-3. Replace with your content following the same patterns
-4. Compile frequently to catch errors early
-
-**Step 5: Use Template Macros**
-
-Templates often define useful macros. Check the preamble for:
-
-```latex
-% Common template macros to use:
-\newcommand{\method}{YourMethodName}  % Consistent method naming
-\newcommand{\eg}{e.g.,\xspace}        % Proper abbreviations
-\newcommand{\ie}{i.e.,\xspace}
-\newcommand{\etal}{\textit{et al.}\xspace}
-```
-
-**Step 6: Clean Up Only at the End**
-
-Only remove template artifacts when paper is nearly complete:
-
-```latex
-% BEFORE SUBMISSION - remove these:
-% - Commented-out template examples
-% - Unused packages
-% - Template's example figures/tables
-% - Lorem ipsum or placeholder text
-
-% KEEP these:
-% - All style files (.sty)
-% - Bibliography style (.bst)
-% - Required packages from template
-% - Any custom macros you're using
-```
-
-### Template Pitfalls to Avoid
-
-| Pitfall | Problem | Solution |
-|---------|---------|----------|
-| Copying only `main.tex` | Missing `.sty`, won't compile | Copy entire directory |
-| Modifying `.sty` files | Breaks conference formatting | Never edit style files |
-| Adding random packages | Conflicts, breaks template | Only add if necessary |
-| Deleting template content too early | Lose formatting reference | Keep as comments until done |
-| Not compiling frequently | Errors accumulate | Compile after each section |
-
-### Quick Template Reference
-
-| Conference | Main File | Key Style File | Notes |
-|------------|-----------|----------------|-------|
-| NeurIPS 2025 | `main.tex` | `neurips.sty` | Has Makefile |
-| ICML 2026 | `example_paper.tex` | `icml2026.sty` | Includes algorithm packages |
-| ICLR 2026 | `iclr2026_conference.tex` | `iclr2026_conference.sty` | Has math_commands.tex |
-| ACL | `acl_latex.tex` | `acl.sty` | Strict formatting |
-| AAAI 2026 | `aaai2026-unified-template.tex` | `aaai2026.sty` | Very strict compliance |
-| COLM 2025 | `colm2025_conference.tex` | `colm2025_conference.sty` | Similar to ICLR |
-
---
-
-## Conference Resubmission & Format Conversion
-
-When a paper is rejected or withdrawn from one venue and resubmitted to another, format conversion is required. This is a common workflow in ML research.
-
-### Workflow 3: Converting Between Conference Formats
-
-```
-Format Conversion Checklist:
- [ ] Step 1: Identify source and target template differences
- [ ] Step 2: Create new project with target template
- [ ] Step 3: Copy content sections (not preamble)
- [ ] Step 4: Adjust page limits and content
- [ ] Step 5: Update conference-specific requirements
- [ ] Step 6: Verify compilation and formatting
-```
-
-**Step 1: Key Template Differences**
-
-| From → To | Page Change | Key Adjustments |
-|-----------|-------------|-----------------|
-| NeurIPS → ICML | 9 → 8 pages | Cut 1 page, add Broader Impact if missing |
-| ICML → ICLR | 8 → 9 pages | Can expand experiments, add LLM disclosure |
-| NeurIPS → ACL | 9 → 8 pages | Restructure for NLP conventions, add Limitations |
-| ICLR → AAAI | 9 → 7 pages | Significant cuts needed, strict style adherence |
-| Any → COLM | varies → 9 | Reframe for language model focus |
-
-**Step 2: Content Migration (NOT Template Merge)**
-
-**Never copy LaTeX preambles between templates.** Instead:
-
-```bash
-# 1. Start fresh with target template
-cp -r templates/icml2026/ new_submission/
-
-# 2. Copy ONLY content sections from old paper
-# - Abstract text
-# - Section content (between \section{} commands)
-# - Figures and tables
-# - Bibliography entries
-
-# 3. Paste into target template structure
-```
-
-**Step 3: Adjusting for Page Limits**
-
-When cutting pages (e.g., NeurIPS 9 → AAAI 7):
- Move detailed proofs to appendix
- Condense related work (cite surveys instead of individual papers)
- Combine similar experiments into unified tables
- Use smaller figure sizes with subfigures
- Tighten writing: eliminate redundancy, use active voice
-
-When expanding (e.g., ICML 8 → ICLR 9):
- Add ablation studies reviewers requested
- Expand limitations discussion
- Include additional baselines
- Add qualitative examples
-
-**Step 4: Conference-Specific Adjustments**
-
-| Target Venue | Required Additions |
-|--------------|-------------------|
-| **ICML** | Broader Impact Statement (after conclusion) |
-| **ICLR** | LLM usage disclosure, reciprocal reviewing agreement |
-| **ACL/EMNLP** | Limitations section (mandatory), Ethics Statement |
-| **AAAI** | Strict adherence to style file (no modifications) |
-| **NeurIPS** | Paper checklist (appendix), lay summary if accepted |
-
-**Step 5: Update References**
-
-```latex
-% Remove self-citations that reveal identity (for blind review)
-% Update any "under review" citations to published versions
-% Add new relevant work published since last submission
-```
-
-**Step 6: Addressing Previous Reviews**
-
-When resubmitting after rejection:
- **Do** address reviewer concerns in the new version
- **Do** add experiments/clarifications reviewers requested
- **Don't** include a "changes from previous submission" section (blind review)
- **Don't** reference the previous submission or reviews
-
-**Common Conversion Pitfalls:**
- ❌ Copying `\usepackage` commands (causes conflicts)
- ❌ Keeping old conference header/footer commands
- ❌ Forgetting to update `\bibliography{}` path
- ❌ Missing conference-specific required sections
- ❌ Exceeding page limit after format change
-
---
-
-## Citation Workflow (Hallucination Prevention)
-
-**⚠️ CRITICAL**: AI-generated citations have ~40% error rate. **Never write BibTeX from memory.**
-
-### The Golden Rule
-
-```
-IF you cannot programmatically fetch a citation:
-    → Mark it as [CITATION NEEDED] or [PLACEHOLDER - VERIFY]
-    → Tell the scientist explicitly
-    → NEVER invent a plausible-sounding reference
-```
-
-### Workflow 2: Adding Citations
-
-```
-Citation Verification (MANDATORY for every citation):
- [ ] Step 1: Search using Exa MCP or Semantic Scholar API
- [ ] Step 2: Verify paper exists in 2+ sources (Semantic Scholar + arXiv/CrossRef)
- [ ] Step 3: Retrieve BibTeX via DOI (programmatically, not from memory)
- [ ] Step 4: Verify the claim you're citing actually appears in the paper
- [ ] Step 5: Add verified BibTeX to bibliography
- [ ] Step 6: If ANY step fails → mark as placeholder, inform scientist
-```
-
-**Step 0: Use Exa MCP for Initial Search (Recommended)**
-
-If Exa MCP is installed, use it to find relevant papers:
-```
-Search: "RLHF language model alignment 2023"
-Search: "sparse autoencoders interpretability"
-Search: "attention mechanism transformers Vaswani"
-```
-
-Then verify each result with Semantic Scholar and fetch BibTeX via DOI.
-
-**Step 1: Search Semantic Scholar**
-
-```python
-from semanticscholar import SemanticScholar
-
-sch = SemanticScholar()
-results = sch.search_paper("attention mechanism transformers", limit=5)
-for paper in results:
-    print(f"{paper.title} - {paper.paperId}")
-    print(f"  DOI: {paper.externalIds.get('DOI', 'N/A')}")
-```
-
-**Step 2: Verify Existence**
-
-Confirm paper appears in at least two sources (Semantic Scholar + CrossRef/arXiv).
-
-**Step 3: Retrieve BibTeX via DOI**
-
-```python
-import requests
-
-def doi_to_bibtex(doi: str) -> str:
-    """Get verified BibTeX from DOI via CrossRef."""
-    response = requests.get(
-        f"https://doi.org/{doi}",
-        headers={"Accept": "application/x-bibtex"}
-    )
-    response.raise_for_status()
-    return response.text
-
-# Example
-bibtex = doi_to_bibtex("10.48550/arXiv.1706.03762")
-print(bibtex)
-```
-
-**Step 4: Verify Claims**
-
-Before citing for a specific claim, access the paper and confirm the attributed claim actually appears.
-
-**Step 5: Handle Failures Explicitly**
-
-If you cannot verify a citation at ANY step:
-
-```latex
-% Option 1: Explicit placeholder
-\cite{PLACEHOLDER_smith2023_verify}  % TODO: Could not verify - scientist must confirm
-
-% Option 2: Note in text
-... as shown in prior work [CITATION NEEDED - could not verify Smith et al. 2023].
-```
-
-**Always inform the scientist:**
-> "I could not verify the following citations and have marked them as placeholders:
-> - Smith et al. 2023 on reward hacking - could not find in Semantic Scholar
-> - Jones 2022 on scaling laws - found similar paper but different authors
-> Please verify these before submission."
-
-### Summary: Citation Rules
-
-| Situation | Action |
-|-----------|--------|
-| Found paper, got DOI, fetched BibTeX | ✅ Use the citation |
-| Found paper, no DOI | ✅ Use arXiv BibTeX or manual entry from paper |
-| Paper exists but can't fetch BibTeX | ⚠️ Mark placeholder, inform scientist |
-| Uncertain if paper exists | ❌ Mark `[CITATION NEEDED]`, inform scientist |
-| "I think there's a paper about X" | ❌ **NEVER cite** - search first or mark placeholder |
-
-**🚨 NEVER generate BibTeX from memory—always fetch programmatically. 🚨**
-
-See [references/citation-workflow.md](references/citation-workflow.md) for complete API documentation.
-
---
-
-## Common Issues and Solutions
-
-**Issue: Abstract too generic**
-
-Delete first sentence if it could be prepended to any ML paper. Start with your specific contribution.
-
-**Issue: Introduction exceeds 1.5 pages**
-
-Split background into Related Work. Front-load contribution bullets. Methods should start by page 2-3.
-
-**Issue: Experiments lack explicit claims**
-
-Add sentence before each experiment: "This experiment tests whether [specific claim]..."
-
-**Issue: Reviewers find paper hard to follow**
-
- Add explicit signposting: "In this section, we show X"
- Use consistent terminology throughout
- Include figure captions that stand alone
-
-**Issue: Missing statistical significance**
-
-Always include:
- Error bars (specify: std dev or std error)
- Number of runs
- Statistical tests if comparing methods
-
---
-
-## Reviewer Evaluation Criteria
-
-Reviewers assess papers on four dimensions:
-
-| Criterion | What Reviewers Look For |
-|-----------|------------------------|
-| **Quality** | Technical soundness, well-supported claims |
-| **Clarity** | Clear writing, reproducible by experts |
-| **Significance** | Community impact, advances understanding |
-| **Originality** | New insights (doesn't require new method) |
-
-**Scoring (NeurIPS 6-point scale):**
- 6: Strong Accept - Groundbreaking, flawless
- 5: Accept - Technically solid, high impact
- 4: Borderline Accept - Solid, limited evaluation
- 3: Borderline Reject - Solid but weaknesses outweigh
- 2: Reject - Technical flaws
- 1: Strong Reject - Known results or ethics issues
-
-See [references/reviewer-guidelines.md](references/reviewer-guidelines.md) for detailed reviewer instructions.
-
---
-
-## Tables and Figures
-
-### Tables
-
-Use `booktabs` LaTeX package for professional tables:
-
-```latex
-\usepackage{booktabs}
-\begin{tabular}{lcc}
-\toprule
-Method & Accuracy ↑ & Latency ↓ \\
-\midrule
-Baseline & 85.2 & 45ms \\
-\textbf{Ours} & \textbf{92.1} & 38ms \\
-\bottomrule
-\end{tabular}
-```
-
-**Rules:**
- Bold best value per metric
- Include direction symbols (↑ higher is better, ↓ lower is better)
- Right-align numerical columns
- Consistent decimal precision
-
-### Figures
-
- **Vector graphics** (PDF, EPS) for all plots and diagrams
- **Raster** (PNG 600 DPI) only for photographs
- Use **colorblind-safe palettes** (Okabe-Ito or Paul Tol)
- Verify **grayscale readability** (8% of men have color vision deficiency)
- **No title inside figure**—the caption serves this function
- **Self-contained captions**—reader should understand without main text
-
---
-
-## References & Resources
-
-### Reference Documents (Deep Dives)
-
-| Document | Contents |
-|----------|----------|
-| [writing-guide.md](references/writing-guide.md) | Gopen & Swan 7 principles, Ethan Perez micro-tips, word choice |
-| [citation-workflow.md](references/citation-workflow.md) | Citation APIs, Python code, BibTeX management |
-| [checklists.md](references/checklists.md) | NeurIPS 16-item, ICML, ICLR, ACL requirements |
-| [reviewer-guidelines.md](references/reviewer-guidelines.md) | Evaluation criteria, scoring, rebuttals |
-| [sources.md](references/sources.md) | Complete bibliography of all sources |
-
-### LaTeX Templates
-
-Templates in `templates/` directory: **ICML 2026**, **ICLR 2026**, **NeurIPS 2025**, **ACL/EMNLP**, **AAAI 2026**, **COLM 2025**.
-
-**Compiling to PDF:**
- **VS Code/Cursor**: Install LaTeX Workshop extension + TeX Live → Save to auto-compile
- **Command line**: `latexmk -pdf main.tex` or `pdflatex` + `bibtex` workflow
- **Online**: Upload to [Overleaf](https://overleaf.com)
-
-See [templates/README.md](templates/README.md) for detailed setup instructions.
-
-### Key External Sources
-
-**Writing Philosophy:**
- [Neel Nanda: How to Write ML Papers](https://www.alignmentforum.org/posts/eJGptPbbFPZGLpjsp/highly-opinionated-advice-on-how-to-write-ml-papers) - Narrative, "What/Why/So What"
- [Farquhar: How to Write ML Papers](https://sebastianfarquhar.com/on-research/2024/11/04/how_to_write_ml_papers/) - 5-sentence abstract
- [Gopen & Swan: Science of Scientific Writing](https://cseweb.ucsd.edu/~swanson/papers/science-of-writing.pdf) - 7 reader expectation principles
- [Lipton: Heuristics for Scientific Writing](https://www.approximatelycorrect.com/2018/01/29/heuristics-technical-scientific-writing-machine-learning-perspective/) - Word choice
- [Perez: Easy Paper Writing Tips](https://ethanperez.net/easy-paper-writing-tips/) - Micro-level clarity
-
-**APIs:** [Semantic Scholar](https://api.semanticscholar.org/api-docs/) | [CrossRef](https://www.crossref.org/documentation/retrieve-metadata/rest-api/) | [arXiv](https://info.arxiv.org/help/api/basics.html)
-
-**Venues:** [NeurIPS](https://neurips.cc/Conferences/2025/PaperInformation/StyleFiles) | [ICML](https://icml.cc/Conferences/2025/AuthorInstructions) | [ICLR](https://iclr.cc/Conferences/2026/AuthorGuide) | [ACL](https://github.com/acl-org/acl-style-files)
-
@@ -0,0 +1,394 @@
+# Autoreason: Iterative Refinement Methodology
+
+Complete reference for the autoreason iterative refinement method, derived from experimental results across subjective writing tasks, competitive programming, and four model tiers. Use this when any output (paper draft, experiment script, analysis, task definition) needs iterative improvement.
+
+**Source**: [NousResearch/autoreason](https://github.com/NousResearch/autoreason) — "Autoreason: When Iterative LLM Refinement Works and Why It Fails"
+
+---
+
+## Strategy Selection Guide
+
+### Decision Tree
+
+```
+Is the task objectively verifiable (code, math, factual)?
+├── YES → Does the model solve it on the first attempt?
+│   ├── YES → Use single pass (no refinement needed)
+│   └── NO → Use autoreason (structured analysis → reason-informed revision)
+│
+└── NO (subjective) → What model tier are you using?
+    ├── Weak (Llama 8B, small models)
+    │   → Single pass. Model too weak for refinement to help.
+    │     Invest in generation quality, not iteration.
+    │
+    ├── Mid-tier (Haiku 3.5, Gemini Flash)
+    │   → Autoreason with stronger judges. This is the sweet spot.
+    │     Self-refinement DESTROYS weak model outputs — autoreason prevents this.
+    │
+    ├── Strong (Sonnet 4)
+    │   → Autoreason for open-ended tasks. Wins 3/5.
+    │     Critique-and-revise for concrete technical tasks (2/5).
+    │
+    └── Frontier (Sonnet 4.6, Opus)
+        ├── Constrained scope? → Autoreason. Wins 2/3 constrained tasks.
+        └── Unconstrained? → Critique-and-revise or single pass.
+            Autoreason FAILS on unconstrained frontier tasks (comes last).
+```
+
+### Strategy Comparison Table
+
+| Strategy | Best For | Avoid When | Compute (per iteration) |
+|----------|----------|------------|------------------------|
+| **Single pass** | Frontier models, template tasks, tight budgets | Mid-tier models where quality ceiling is low | 1 call |
+| **Critique-and-revise** | Concrete technical requirements (system design, specifications) | Weak models (degrades output), unconstrained subjective tasks | 2 calls |
+| **Autoreason** | Mid-tier models, constrained scope, tasks with genuine tradeoffs | Weak models (Llama 8B), frontier + unconstrained | ~6 calls |
+| **Best-of-N** | Almost never recommended | Weak models especially — worse than single pass | N calls |
+
+### Why Each Strategy Fails
+
+| Strategy | Failure Mode | Mechanism |
+|----------|-------------|-----------|
+| **Single pass** | Quality ceiling | No mechanism to improve beyond first attempt |
+| **Critique-and-revise** | Progressive degradation | Model hallucinates problems (sycophancy), scope creeps each pass, never declines to change |
+| **Best-of-N** | Random selection | Without good ranking signal, more samples = more mediocre options |
+| **Autoreason (unconstrained)** | Synthesis drift | Stronger models produce syntheses so consistently preferred that incumbent never stabilizes |
+
+---
+
+## The Autoreason Loop
+
+### Architecture
+
+```
+┌──────────────────────────────────────────────────────────┐
+│                    ITERATION LOOP                         │
+│                                                           │
+│   Incumbent A ──► Critic ──► Author B ──► Synthesizer     │
+│       │                                      │            │
+│       │              ┌───────────────────────┘            │
+│       ▼              ▼                                    │
+│      [A]           [AB]          [B]                      │
+│       │              │            │                       │
+│       └──────────────┼────────────┘                       │
+│                      ▼                                    │
+│              Judge Panel (blind)                          │
+│                      │                                    │
+│                      ▼                                    │
+│                   Winner                                  │
+│                      │                                    │
+│              ┌───────┴───────┐                            │
+│              ▼               ▼                            │
+│         A wins k=2      B or AB wins                      │
+│         consecutive?    → new incumbent                   │
+│              │                                            │
+│              ▼                                            │
+│           CONVERGED                                       │
+└──────────────────────────────────────────────────────────┘
+```
+
+### Roles
+
+Every role is a **fresh, isolated agent** with no shared context:
+
+| Role | Input | Output | Key Rule |
+|------|-------|--------|----------|
+| **Critic** | Task + Incumbent A | List of problems | Find problems ONLY. No fixes. No suggestions. |
+| **Author B** | Task + A + Critique | Revised version B | Address each criticism. State which problem each change fixes. |
+| **Synthesizer** | Task + X + Y (randomized labels) | Synthesis AB | Take strongest elements of each. Not a compromise. |
+| **Judge Panel** | Task + A, AB, B (randomized labels + order) | Ranking | Rank best to worst. No authorship stake. |
+
+### Configuration
+
+| Parameter | Value | Rationale |
+|-----------|-------|-----------|
+| **Convergence k** | 2 | k=1 premature (94% displaced later). k=2 converges 100%, quality plateaus. k=3 fails 24%, 2x cost, no quality gain. |
+| **Author temperature** | 0.7-0.8 | Encourages diverse revisions |
+| **Judge temperature** | 0.3 | Encourages consistent evaluation |
+| **In-loop judges** | 3 | Balance per-pass cost vs evaluation stability |
+| **Final evaluation judges** | 7 | Higher statistical power for final comparison |
+| **Max tokens** | 4096 | Standard; 8192 for long-form (papers) |
+| **Judge type** | Chain-of-thought | 3x faster convergence on some tasks. Always use. |
+| **Tiebreak** | Conservative (incumbent wins) | Prevents false positives — A must be genuinely beaten |
+| **Max passes** | 25 (constrained), 50 (remedy) | Safety cap; most converge by pass 10-15 |
+
+### Prompts
+
+#### Critic
+```
+System: You are a critical reviewer. Your only job is to find real problems. 
+Be specific and concrete. Do not suggest fixes.
+
+User: Find real problems with this proposal. Focus on:
+- Things that won't work as described
+- Complexity that doesn't pay for itself
+- Assumptions that are wrong
+- Missing pieces
+Do NOT propose fixes. Just the problems.
+```
+
+#### Author B
+```
+System: You are a senior consultant revising a proposal based on specific 
+criticisms. Address each valid criticism directly. Do not make changes not 
+motivated by an identified problem.
+
+User: [TASK] + [VERSION A] + [CRITIC OUTPUT]
+Revise to address these problems. For each change, state which problem it fixes.
+```
+
+#### Synthesizer
+```
+System: You are given two versions as equal inputs. Take the strongest elements 
+from each and produce a coherent synthesis. This is not a compromise.
+
+User: [TASK] + [VERSION X] + [VERSION Y]
+(labels randomized — synthesizer doesn't know which is incumbent)
+```
+
+#### Judge (Chain-of-Thought) — ALWAYS USE THIS VERSION
+```
+System: You are an independent evaluator. Think carefully before deciding.
+
+User: [TASK] + Three proposals. For each, think step by step:
+1. What does it get right?
+2. What does it get wrong or miss?
+3. Are numbers and claims defensible?
+4. Is detail appropriate or bloated?
+After reasoning, rank all three.
+RANKING: [best], [second], [worst]
+```
+
+#### Baseline Prompts (for comparison experiments)
+
+| Baseline | Prompt |
+|----------|--------|
+| **Conservative** | "Make minimal improvements while preserving what works. Do not add new sections or significantly expand scope." |
+| **Improve this** | "Improve this document." (no further guidance) |
+| **Harsh critic** | "Critically evaluate and rewrite, fixing all weaknesses you identify." |
+| **Critique & revise** | Step 1: "Produce a structured critique. List specific weaknesses." Step 2: "Revise to address each criticism." |
+
+---
+
+## Scoring: Borda Count
+
+Judges rank candidates. Points awarded by rank position:
+
+| Rank | Points (3 candidates) |
+|------|----------------------|
+| 1st | 3 |
+| 2nd | 2 |
+| 3rd | 1 |
+
+**Aggregation**: Sum across all judges. Winner = highest total.
+**Tiebreak**: Incumbent (A) wins any tie.
+
+**Example** (3 judges):
+- Judge 1: AB > A > B → AB gets 3, A gets 2, B gets 1
+- Judge 2: A > AB > B → A gets 3, AB gets 2, B gets 1
+- Judge 3: AB > B > A → AB gets 3, B gets 2, A gets 1
+- Totals: AB=8, A=6, B=4 → AB wins, becomes new incumbent
+
+**Randomization per judge**:
+- Candidate labels randomized (A might be called "Proposal X" for one judge, "Proposal Z" for another)
+- Presentation order randomized (AB might appear first or last)
+- This prevents position bias and label bias
+
+---
+
+## Model Selection Guide
+
+### Empirical Results by Model Tier
+
+| Model | Autoreason Wins | Autoreason Avg Borda | Best Baseline | Margin | Recommendation |
+|-------|----------------|---------------------|---------------|--------|----------------|
+| **Llama 3.1 8B** | 1/3 | 23.7 | 25.0 (single) | -1.3 | Skip autoreason. Model too weak for diverse candidates. |
+| **Gemini 2.0 Flash** | 2/3 | 25.0 | 20.0 (single) | +5.0 | Good candidate. Moderate gains. |
+| **Haiku 3.5** | 3/3 | **42.0** | 33.7 (single) | **+8.3** | **Best candidate.** Perfect scores. Baselines actively destroy quality. |
+| **Sonnet 4** | 3/5 | 27.8 | 22.4 (C&R) | +5.4 | Good candidate for open tasks. C&R better for technical tasks. |
+| **Sonnet 4.6 (unconstrained)** | 0/1 | 7.0 | 31.0 (C&R) | -24.0 | Do NOT use autoreason without constraints. |
+| **Sonnet 4.6 (constrained)** | 2/3 | 29.0 | 27.0 (improve) | +2.0 | Use only with scope constraints. |
+
+### The Generation-Evaluation Gap
+
+The core insight: **autoreason's value depends on the gap between a model's generation capability and its self-evaluation capability.**
+
+```
+Weak models (Llama 8B):
+  Generation: Poor  |  Self-evaluation: Poor
+  Gap: Small (both bad) → Autoreason can't help, no diverse candidates
+
+Mid-tier models (Haiku, Flash):
+  Generation: Decent  |  Self-evaluation: Poor
+  Gap: LARGE → Autoreason's sweet spot. External eval bridges the gap.
+
+Strong models (Sonnet 4):
+  Generation: Good  |  Self-evaluation: Decent
+  Gap: Moderate → Autoreason helps on 3/5 tasks
+
+Frontier models (Sonnet 4.6):
+  Generation: Excellent  |  Self-evaluation: Good
+  Gap: Small → Simple methods suffice. Autoreason hurts on unconstrained tasks.
+```
+
+**Practical rule**: As model costs drop and capabilities improve, today's frontier becomes tomorrow's mid-tier. The generation-evaluation gap is structural, not temporary. Match refinement architecture to the model's position on the capability curve.
+
+### Judge Selection
+
+| Author Model | Recommended Judge | Rationale |
+|-------------|------------------|-----------|
+| Llama 8B | Don't use autoreason | Model too weak |
+| Gemini Flash | Sonnet 4 | Cross-model evaluation works |
+| Haiku 3.5 | Sonnet 4 | Strong external eval is the mechanism |
+| Haiku 3.5 | Haiku 3.5 (same) | Still works — tournament structure provides value even without strong judges (20.7 vs 18.3 avg Borda) |
+| Sonnet 4 | Sonnet 4 (same) | Same-model judges work at this tier |
+| Sonnet 4.6 | Sonnet 4.6 (same) | Only with scope constraints |
+
+---
+
+## Scope Constraint Design
+
+### What Makes Autoreason Work on Constrained Tasks
+
+The same model (Sonnet 4.6) goes from **last place** (unconstrained) to **first place** (constrained) with scope constraints. The constraints bound the improvement space so synthesis drift can't accumulate.
+
+### Effective Constraints
+
+| Constraint Type | Example | Why It Works |
+|----------------|---------|-------------|
+| **Fixed facts** | "Use only these 8 data points, add nothing else" | Bounds information space |
+| **Fixed deliverable** | "500-word startup pitch" (not "improve this") | Defines done condition |
+| **Fixed structure** | "Exactly 4 sections, each with 3 numbered items" | Prevents structural drift |
+| **Fixed change items** | "Address exactly these 3 reviewer concerns" | Bounds modification scope |
+
+### Ineffective Constraints
+
+| Constraint | Why It Fails | What Happens |
+|-----------|-------------|-------------|
+| Word count alone | Not a scope constraint | False convergence — rejected for length, not quality |
+| "Be concise" | Too vague | Ignored after 2-3 passes |
+| "Be comprehensive" | Anti-constraint | Invites scope creep |
+| No constraints at all | Unbounded improvement space | Synthesis dominates, no convergence |
+
+### Task Categories
+
+| Task Type | Autoreason Works? | Why |
+|-----------|-------------------|-----|
+| Tasks with genuine tradeoffs (strategy, policy) | Yes | Multiple valid approaches for tournament to select between |
+| Constrained writing (pitch, memo, postmortem) | Mostly (2/3) | Bounded scope, clear evaluation criteria |
+| Template-filling (incident postmortem) | No | One correct structure, minimal decision space |
+| Competitive programming | Yes | Naturally scoped, test suite provides external verification |
+| Open-ended unconstrained + frontier model | No | Synthesis drift, no convergence |
+
+---
+
+## Failure Taxonomy
+
+| Failure Mode | Condition | Detection | Evidence |
+|-------------|-----------|-----------|----------|
+| **Self-correction unreliable** | No external evaluation signal | Baselines degrade below single pass | Haiku baselines: 16.3 avg vs 33.7 single pass |
+| **Drift / synthesis dominance** | Unconstrained scope | A wins <15%, AB dominates | Sonnet 4.6 unconstrained: A wins 12%, AB wins 60%+ |
+| **Overfitting to visible feedback** | Shallow revision loop (C&R) | High public/private divergence | C&R overfits 32% on hard code problems |
+| **No convergence** | Broken judge pipeline | Parsing failures, <3 valid judges | Mixed panel parser failure: 11+ passes |
+| **Model too weak** | Insufficient generation diversity | All candidates look similar | Llama 8B wins only 1/3 tasks |
+
+### Recovery Patterns
+
+| Failure | Recovery |
+|---------|----------|
+| No convergence (drift) | Add scope constraints to the task |
+| No convergence (broken judges) | Fix parser, ensure 3 valid judges before continuing |
+| Quality degrades with iteration | Switch to single pass or add constraints |
+| Model too weak | Use a stronger model for generation, keep weak model for cheap roles |
+| Overfitting (code) | Use structured analysis step, not just test feedback |
+
+---
+
+## Code Domain Adaptation
+
+The autoreason method adapts differently for code vs writing:
+
+### Writing Domain
+```
+Call 1: Critic (find problems in incumbent)
+Call 2: Author B (revise based on critique)
+Call 3: Synthesizer (merge A and B)
+Calls 4-6: Judge Panel (3 blind judges rank A, B, AB)
+```
+
+### Code Domain (6-call budget)
+```
+Call 1: Initial generation
+Call 2: Structured analysis (5 points — NO CODE):
+  - Problem analysis: what does the problem actually require?
+  - Approach analysis: what approach did we use, is it correct?
+  - Failure analysis: why did tests fail?
+  - Alternative approaches: what else could work?
+  - Edge cases: what inputs might break the solution?
+Calls 3-6: Reason-informed revisions
+  - Each revision must explain WHY it fixes the issue
+  - Sees test results from public (visible) test cases
+```
+
+**Key difference**: The code strategy replaces the judge panel with test-suite evaluation (objective ground truth). The structured analysis step (Call 2) is what drives recovery — it forces reasoning about *why* the approach failed before attempting fixes.
+
+**Results**: Recovery is the mechanism. Among problems where both autoreason and single-pass failed initially, autoreason recovered 62% vs single-pass's 43% (McNemar p=0.041, Cohen's h=0.32).
+
+---
+
+## Applying Autoreason to Paper Writing
+
+The paper itself was refined using autoreason (Section 8 of the paper):
+
+### Setup
+- Model: claude-opus-4
+- Judges: 3 Opus judges
+- Enhancement: Ground-truth critic (access to actual experimental data)
+- Result: Converged in 9 passes
+
+### Key Findings for Paper Refinement
+
+1. **Ground-truth critic is essential**: Without ground-truth access, Opus hallucinated a fabricated ablation study, fake confidence intervals, wrong model names, and incorrect role descriptions. With ground-truth access, the critic caught all four on pass 1.
+
+2. **Judge panel integrity matters**: A broken parser in one judge (Gemini output format mismatch) reduced the panel from 3 to 2 judges. This prevented convergence for 11+ passes. Fixing to 3 working judges, the same incumbent converged in 2 passes. A broken judge doesn't add noise — it prevents equilibrium.
+
+### Recommended Setup for Paper Refinement
+
+```
+Critic prompt: "You are reviewing a research paper draft. You have access to the 
+actual experimental results [GROUND TRUTH DATA]. Find factual errors, unsupported 
+claims, hallucinated results, and structural problems. Do not suggest fixes."
+
+Author B prompt: "Revise this paper draft to fix the identified problems. For each 
+change, cite the specific problem it addresses. Do not add claims not supported by 
+the provided experimental data."
+
+Judge prompt (CoT): "Compare three versions of this paper. For each, evaluate:
+1. Factual accuracy against the provided results
+2. Clarity of the narrative and contribution
+3. Whether claims are properly hedged and supported
+4. Writing quality (concision, precision, no filler)
+After reasoning, rank all three. RANKING: [best], [second], [worst]"
+```
+
+### What to Provide as Ground Truth
+- All experimental result JSON files
+- Statistical test outputs
+- Raw numbers for every table and figure
+- Configuration files showing exact hyperparameters
+- Code that generated the results (for method description accuracy)
+
+---
+
+## Compute Budget Reference
+
+| Method | Calls per Pass | Typical Passes | Total Calls | Relative Cost |
+|--------|---------------|----------------|-------------|---------------|
+| Single pass | 1 | 1 | 1 | 1x |
+| Best-of-N | N | 1 | N | Nx |
+| Critique & revise | 2 | 15 | 30 | 30x |
+| Autoreason (in-loop) | ~6 | 10-15 | 60-90 | 60-90x |
+| Autoreason (with final eval) | ~6 + 7 | 10-15 + 1 | 67-97 | ~80x |
+
+**Cost-quality tradeoff**: Autoreason uses ~6x more compute per pass and typically runs more passes. This is a real tradeoff. The method trades compute for evaluation quality. On constrained tasks with mid-tier models, this tradeoff is strongly positive. On unconstrained tasks with frontier models, it's negative.
+
+**CoT judges reduce cost**: 1 CoT judge provides evaluation quality comparable to 3 standard judges, at ~40% cost savings. Always use CoT judges.
@@ -10,6 +10,8 @@ This reference documents the mandatory checklist requirements for major ML/AI co
 - [ICML Paper Checklist](#icml-paper-checklist)
 - [ICLR Requirements](#iclr-requirements)
 - [ACL Requirements](#acl-requirements)
+- [AAAI Requirements](#aaai-requirements)
+- [COLM Requirements](#colm-requirements)
 - [Universal Pre-Submission Checklist](#universal-pre-submission-checklist)

 ---
@@ -280,6 +282,77 @@ If applicable:

 ---

+## AAAI Requirements
+
+### Formatting (Strictest of All Venues)
+
+AAAI enforces formatting rules more strictly than any other major venue. Papers that deviate from the template are desk-rejected.
+
+- [ ] Use the **exact** AAAI style file without modification — no `\setlength`, no `\vspace` hacks, no font overrides
+- [ ] 7 pages main content (8 for camera-ready with author info)
+- [ ] Two-column format, Times font (set by template)
+- [ ] References and appendices do not count toward page limit
+- [ ] Abstract must be a single paragraph
+- [ ] Do not modify margins, column widths, or font sizes
+
+### Required Sections
+
+- [ ] Abstract (single paragraph, no math or citations)
+- [ ] Introduction with clear contribution statement
+- [ ] References in AAAI format (uses `aaai2026.bst`)
+- [ ] Appendix (optional, unlimited)
+
+### Ethics and Reproducibility
+
+- [ ] Broader impact statement (encouraged but not always mandatory — check current year's CFP)
+- [ ] Reproducibility details (datasets, code availability)
+- [ ] Acknowledge use of AI writing tools if applicable
+
+### Key Differences from Other Venues
+
+- **No separate limitations section required** (unlike ACL), but discussing limitations is recommended
+- **Strictest formatting enforcement** — the style checker will reject non-compliant PDFs
+- **No paper checklist** like NeurIPS has, but the universal checklist below still applies
+- **Unified template** covers main paper and supplementary in the same file
+
+---
+
+## COLM Requirements
+
+### Overview
+
+COLM (Conference on Language Modeling) focuses specifically on language model research. Framing must target this community.
+
+### Formatting
+
+- [ ] 9 pages main content (10 for camera-ready)
+- [ ] Use COLM template (based on ICLR template with modifications)
+- [ ] Double-blind review
+- [ ] References and appendices unlimited
+
+### Required Sections
+
+- [ ] Abstract
+- [ ] Introduction framed for language modeling community
+- [ ] Conclusion
+- [ ] References
+
+### Content Expectations
+
+- [ ] Contribution must be relevant to language models (broadly interpreted: training, evaluation, applications, theory, alignment, safety)
+- [ ] If the method is general, frame with language model examples
+- [ ] Baselines should include recent LM-specific methods where applicable
+
+### Key Differences from Other Venues
+
+- **Narrower scope** than NeurIPS/ICML — must frame for LM community
+- **Template derived from ICLR** — similar formatting rules
+- **Newer venue** — reviewer norms are still establishing; err on the side of thorough evaluation
+- **No mandatory checklist** like NeurIPS, but broader impact discussion is expected
+- **LLM disclosure**: If LLMs were used in research (code generation, data annotation, writing assistance), disclose this
+
+---
+
 ## Universal Pre-Submission Checklist

 ### Before Every Submission
@@ -289,7 +289,7 @@ class CitationManager:
                )
                if resp.status_code == 200:
                    sources.append("CrossRef")
-            except:
+            except Exception:
                pass

        # Check arXiv if ID available
@@ -301,7 +301,7 @@ class CitationManager:
                )
                if "<entry>" in resp.text and "<title>" in resp.text:
                    sources.append("arXiv")
-            except:
+            except Exception:
                pass

        return len(sources) >= 2, sources
@@ -318,7 +318,7 @@ class CitationManager:
                )
                if resp.status_code == 200:
                    return resp.text
-            except:
+            except Exception:
                pass

        # Fallback: generate from paper data
@@ -419,7 +419,7 @@ def batch_cite(queries: List[str], output_file: str = "references.bib"):
 | Customization | Limited | Highly flexible |
 | Backend | bibtex | Biber (recommended) |

-**Recommendation**: Use BibLaTeX with Biber for new papers.
+**Recommendation**: Use natbib with BibTeX for conference submissions — all major venue templates (NeurIPS, ICML, ICLR, ACL, AAAI, COLM) ship with natbib and `.bst` files. BibLaTeX with Biber is an option for journals or personal projects where you control the template.

 ### LaTeX Setup

@@ -0,0 +1,728 @@
+# Experiment Design Patterns
+
+Patterns and best practices distilled from running research experiments at scale with the Hermes agent. These cover experiment infrastructure, evaluation protocols, monitoring, and failure recovery.
+
+---
+
+## Experiment Infrastructure
+
+### Directory Structure
+
+Organize experiments with a consistent structure:
+
+```
+workspace/
+  experiments/
+    run_main.py                # Core experiment runner
+    run_baselines.py           # Baseline comparison
+    run_ablation.py            # Ablation studies
+    strategies.py              # Method implementations
+    config.yaml                # Shared configuration
+  results/
+    <experiment_name>/
+      <task_or_problem>/
+        <strategy>/
+          result.json          # Final metrics
+          final_output.md      # Final output artifact
+          history.json         # Full trajectory/log
+          pass_01/             # Per-iteration artifacts (if iterative)
+            intermediate.md
+  analysis/
+    analyze_results.py         # Statistical analysis
+    compute_stats.py           # Significance tests
+    make_charts.py             # Visualization
+  paper/
+    paper.tex                  # LaTeX source
+    fig_*.pdf                  # Generated figures
+```
+
+### Script Design Principles
+
+**1. Incremental Saving (Crash Recovery)**
+
+Every experiment script should save results after each unit of work, and skip already-completed work on restart:
+
+```python
+import json, os
+from pathlib import Path
+
+def run_experiment(problems, strategies, output_dir):
+    for problem in problems:
+        for strategy in strategies:
+            result_path = Path(output_dir) / problem["id"] / strategy / "result.json"
+            if result_path.exists():
+                print(f"Skipping {problem['id']}/{strategy} (already done)")
+                continue
+            
+            # Run the experiment
+            result = execute_strategy(problem, strategy)
+            
+            # Save immediately
+            result_path.parent.mkdir(parents=True, exist_ok=True)
+            with open(result_path, 'w') as f:
+                json.dump(result, f, indent=2)
+```
+
+This pattern makes re-runs safe and efficient. If a process crashes at problem 47/150, restarting skips the first 46.
+
+**2. Artifact Preservation**
+
+Save all intermediate outputs, not just final results. This enables post-hoc analysis without re-running:
+
+```python
+def save_pass_artifacts(output_dir, pass_num, artifacts):
+    """Save all artifacts from a single pass of an iterative method."""
+    pass_dir = Path(output_dir) / f"pass_{pass_num:02d}"
+    pass_dir.mkdir(parents=True, exist_ok=True)
+    
+    for name, content in artifacts.items():
+        with open(pass_dir / f"{name}.md", 'w') as f:
+            f.write(content)
+```
+
+**3. Configuration Management**
+
+Use YAML configs for reproducibility:
+
+```yaml
+# config.yaml
+model: anthropic/claude-sonnet-4-20250514
+author_temperature: 0.8
+judge_temperature: 0.3
+max_tokens: 4096
+num_judges: 3
+max_passes: 15
+convergence_k: 2
+```
+
+```python
+import yaml
+
+with open("config.yaml") as f:
+    config = yaml.safe_load(f)
+```
+
+**4. Separation of Concerns**
+
+Keep generation, evaluation, and visualization in separate scripts:
+
+| Script | Purpose |
+|--------|---------|
+| `run_experiment.py` | Core method execution |
+| `run_baselines.py` | Baseline comparisons at same compute |
+| `run_eval.py` | Blind evaluation / judge panels |
+| `analyze_results.py` | Statistical analysis |
+| `make_charts.py` | Figure generation |
+
+This lets you re-run evaluation without re-running expensive generation, and regenerate figures without re-running analysis.
+
+---
+
+## Evaluation Protocols
+
+### Blind Judge Panels (for Subjective Tasks)
+
+When evaluating subjective outputs (writing, analysis, recommendations), use a blind judge panel:
+
+```python
+import random
+
+def run_blind_evaluation(outputs: dict, task_prompt: str, num_judges: int = 7):
+    """
+    Run blind evaluation of multiple method outputs.
+    
+    Args:
+        outputs: {"method_name": "output_text", ...}
+        task_prompt: The original task description
+        num_judges: Number of independent judge evaluations
+    """
+    rankings = []
+    
+    for judge_i in range(num_judges):
+        # Randomize labels and presentation order per judge
+        methods = list(outputs.keys())
+        random.shuffle(methods)
+        labels = {m: chr(65 + i) for i, m in enumerate(methods)}  # A, B, C...
+        
+        # Present to judge with randomized labels
+        prompt = f"Task: {task_prompt}\n\n"
+        for method in methods:
+            prompt += f"--- Proposal {labels[method]} ---\n{outputs[method]}\n\n"
+        prompt += "Rank all proposals from best to worst. Format: RANKING: [best], [second], [worst]"
+        
+        ranking = call_judge(prompt)
+        rankings.append({"labels": labels, "ranking": ranking})
+    
+    # Aggregate via Borda count
+    return compute_borda(rankings)
+
+def compute_borda(rankings, n_methods=3):
+    """Borda count: 3/2/1 points for 1st/2nd/3rd."""
+    scores = {}
+    points = {0: n_methods, 1: n_methods - 1, 2: n_methods - 2}  # Adjust for n_methods
+    
+    for r in rankings:
+        for position, method in enumerate(r["ranking"]):
+            scores[method] = scores.get(method, 0) + points.get(position, 0)
+    
+    return scores
+```
+
+Key design decisions:
+- **Randomize both labels AND order** per judge to prevent position bias
+- **Use odd number of judges** (3, 5, 7) to break ties
+- **Conservative tiebreak**: Incumbent/baseline wins ties (prevents false positives)
+- **CoT judges** match non-CoT quality at ~40% cost (1 CoT judge ≈ 3 standard judges)
+
+### Code/Objective Evaluation
+
+For tasks with ground-truth evaluation (code, math, factual):
+
+```python
+import subprocess
+
+def evaluate_code(solution: str, test_cases: list, timeout: int = 30):
+    """Run code solution against test cases with sandboxed execution."""
+    results = {"public": [], "private": []}
+    
+    for test in test_cases:
+        try:
+            proc = subprocess.run(
+                ["python3", "-c", solution],
+                input=test["input"],
+                capture_output=True,
+                timeout=timeout,
+                text=True
+            )
+            actual = proc.stdout.strip()
+            expected = test["expected"].strip()
+            passed = actual == expected
+        except subprocess.TimeoutExpired:
+            passed = False
+        
+        category = "public" if test.get("public") else "private"
+        results[category].append(passed)
+    
+    return {
+        "public_pass_rate": sum(results["public"]) / max(len(results["public"]), 1),
+        "private_pass_rate": sum(results["private"]) / max(len(results["private"]), 1),
+    }
+```
+
+### Compute-Matched Comparison
+
+Always compare methods at equal compute budget. If your method uses N API calls, baselines get N calls too:
+
+| Method | Call Budget | Allocation |
+|--------|-----------|------------|
+| Single pass | 6 calls | 6 independent generations |
+| Critique & revise | 6 calls | 1 generate + 5 revise rounds |
+| Autoreason | 6 calls | 1 generate + 1 analysis + 4 revisions |
+| Best-of-N | 6 calls | 6 independent, pick best on public test |
+
+### Human Evaluation Design
+
+Many ML/NLP papers require human evaluation, especially for subjective tasks (text generation, summarization, dialogue, creative writing). Poorly designed human evals are a common rejection reason.
+
+#### When Human Evaluation Is Required
+
+| Task Type | Required? | Notes |
+|-----------|-----------|-------|
+| Text generation (open-ended) | Yes | LLM-as-judge alone is insufficient for acceptance at ACL/EMNLP |
+| Summarization | Usually | At minimum for a subset of outputs |
+| Dialogue systems | Yes | User studies or annotation |
+| Code generation | No | Test suites are objective ground truth |
+| Classification | No | Standard metrics suffice |
+| Any task with subjective quality | Strongly recommended | Strengthens the paper significantly |
+
+#### Annotation Protocol Design
+
+```
+Human Evaluation Protocol:
+1. Define the evaluation dimensions (fluency, relevance, factual accuracy, etc.)
+2. Create annotation guidelines with examples of each score level
+3. Run a pilot with 2-3 annotators on 20-30 examples
+4. Compute pilot inter-annotator agreement — if low, revise guidelines
+5. Run full evaluation
+6. Report: annotator count, agreement metrics, compensation, time per item
+```
+
+**Evaluation dimensions** (pick relevant subset):
+
+| Dimension | Definition | Scale |
+|-----------|-----------|-------|
+| Fluency | Grammaticality and naturalness | 1-5 Likert |
+| Relevance | Does it address the task? | 1-5 Likert |
+| Factual accuracy | Are stated facts correct? | Binary or 1-5 |
+| Coherence | Logical flow and consistency | 1-5 Likert |
+| Informativeness | Does it provide useful information? | 1-5 Likert |
+| Overall preference | Which output is better? | A/B/Tie (pairwise) |
+
+**Pairwise comparison** (preferred over absolute scoring — more reliable):
+- Present two outputs side-by-side (randomize left/right position)
+- Ask: "Which is better? A / B / Tie"
+- More discriminative and less susceptible to annotator calibration drift
+
+#### Inter-Annotator Agreement
+
+Always report agreement metrics. Without them, reviewers assume your annotations are unreliable.
+
+```python
+# Krippendorff's alpha (preferred — handles missing data, any scale)
+# pip install krippendorffs-alpha
+import krippendorff
+
+# Ratings: rows = annotators, columns = items, values = scores
+ratings = [
+    [3, 4, 1, 2, 5, None, 3],  # Annotator 1
+    [3, 5, 1, 3, 5, 2, 3],     # Annotator 2
+    [4, 4, 2, 2, 4, 2, None],  # Annotator 3
+]
+alpha = krippendorff.alpha(reliability_data=ratings, level_of_measurement="ordinal")
+print(f"Krippendorff's alpha: {alpha:.3f}")
+# Interpretation: >0.80 good, 0.67-0.80 acceptable, <0.67 questionable
+```
+
+```python
+# Cohen's kappa (for exactly 2 annotators, categorical data)
+from sklearn.metrics import cohen_kappa_score
+
+annotator_1 = [1, 2, 3, 1, 2, 3, 2]
+annotator_2 = [1, 2, 2, 1, 3, 3, 2]
+kappa = cohen_kappa_score(annotator_1, annotator_2)
+print(f"Cohen's kappa: {kappa:.3f}")
+# Interpretation: >0.80 excellent, 0.60-0.80 substantial, 0.40-0.60 moderate
+```
+
+| Metric | When to Use | Annotators | Scale |
+|--------|------------|-----------|-------|
+| Krippendorff's alpha | Default choice | Any number | Any (ordinal, nominal, ratio) |
+| Cohen's kappa | 2 annotators, categorical | Exactly 2 | Nominal/ordinal |
+| Fleiss' kappa | 3+ annotators, categorical | 3+ | Nominal |
+| Pearson/Spearman | Continuous scores | 2 | Interval/ratio |
+
+#### Crowdsourcing Platforms
+
+| Platform | Best For | Cost | Quality |
+|----------|----------|------|---------|
+| **Prolific** | Academic research, higher quality | $8-15/hr | High — academic participant pool |
+| **MTurk** | Large-scale, fast turnaround | $2-10/hr | Variable — use qualifications |
+| **Surge AI** | NLP-specific annotations | Premium | High — trained annotators |
+| **Expert annotators** | Domain-specific (medical, legal) | Highest | Highest — but slow |
+
+**Ethics requirements**:
+- Report compensation rate (must be at minimum local minimum wage)
+- Describe annotator demographics if relevant
+- Obtain IRB/ethics approval if required by your institution
+- ACL venues explicitly require compensation documentation
+
+#### What to Report in the Paper
+
+```
+Human Evaluation Section Checklist:
+- [ ] Number of annotators
+- [ ] Annotator qualifications / recruitment method
+- [ ] Number of items evaluated
+- [ ] Evaluation dimensions with definitions
+- [ ] Scale used (Likert, pairwise, binary)
+- [ ] Inter-annotator agreement (Krippendorff's alpha or Cohen's kappa)
+- [ ] Compensation rate
+- [ ] Time per annotation item
+- [ ] Whether annotators saw model identities (should be blind)
+- [ ] Randomization of presentation order
+```
+
+---
+
+## Statistical Analysis
+
+### Required Tests
+
+| Test | When to Use | Python |
+|------|------------|--------|
+| McNemar's test | Comparing two methods on same problems | `scipy.stats.binomtest` for small n |
+| Two-proportion z-test | Comparing success rates | Custom or `statsmodels` |
+| Fisher's exact test | Small sample pairwise comparison | `scipy.stats.fisher_exact` |
+| Bootstrapped CI | Confidence intervals for any metric | Custom bootstrap |
+| Cohen's h | Effect size for proportions | Manual calculation |
+
+### Standard Analysis Script
+
+```python
+import numpy as np
+from scipy import stats
+from pathlib import Path
+import json
+
+def load_all_results(results_dir):
+    """Load all results into a structured format."""
+    results = {}
+    for result_file in Path(results_dir).rglob("result.json"):
+        parts = result_file.relative_to(results_dir).parts
+        if len(parts) >= 3:
+            experiment, task, strategy = parts[0], parts[1], parts[2]
+            data = json.loads(result_file.read_text())
+            results.setdefault(experiment, {}).setdefault(strategy, {})[task] = data
+    return results
+
+def pairwise_mcnemar(method_a_results, method_b_results):
+    """McNemar's test for paired binary outcomes."""
+    a_win_b_lose = sum(1 for a, b in zip(method_a_results, method_b_results) if a and not b)
+    b_win_a_lose = sum(1 for a, b in zip(method_a_results, method_b_results) if b and not a)
+    
+    n = a_win_b_lose + b_win_a_lose
+    if n < 25:
+        # Use exact binomial for small samples
+        result = stats.binomtest(a_win_b_lose, n, 0.5)
+        p_value = result.pvalue
+    else:
+        # Chi-squared approximation
+        chi2 = (abs(a_win_b_lose - b_win_a_lose) - 1)**2 / (a_win_b_lose + b_win_a_lose)
+        p_value = 1 - stats.chi2.cdf(chi2, df=1)
+    
+    return {
+        "a_wins": a_win_b_lose,
+        "b_wins": b_win_a_lose,
+        "n_discordant": n,
+        "p_value": p_value,
+        "significant": p_value < 0.05
+    }
+
+def bootstrap_ci(data, n_bootstrap=10000, ci=0.95):
+    """Bootstrap confidence interval for mean."""
+    means = []
+    for _ in range(n_bootstrap):
+        sample = np.random.choice(data, size=len(data), replace=True)
+        means.append(np.mean(sample))
+    lower = np.percentile(means, (1 - ci) / 2 * 100)
+    upper = np.percentile(means, (1 + ci) / 2 * 100)
+    return {"mean": np.mean(data), "ci_lower": lower, "ci_upper": upper}
+
+def cohens_h(p1, p2):
+    """Cohen's h effect size for two proportions."""
+    return 2 * np.arcsin(np.sqrt(p1)) - 2 * np.arcsin(np.sqrt(p2))
+```
+
+### Reporting Standards
+
+Always include in the paper:
+- **Sample sizes**: n=X problems/tasks
+- **Number of runs**: K independent runs if applicable
+- **Error bars**: Specify standard deviation or standard error
+- **Confidence intervals**: 95% CI for key results
+- **Significance tests**: p-values for key comparisons
+- **Effect sizes**: Cohen's d or h for practical significance
+
+---
+
+## Monitoring (Cron Pattern)
+
+### Cron Prompt Template
+
+For each experiment batch, create a monitoring prompt:
+
+```
+Check the status of the [EXPERIMENT_NAME] experiment:
+
+1. Process check: ps aux | grep [PROCESS_PATTERN]
+2. Log check: tail -30 [LOG_FILE]
+3. Results check: ls [RESULT_DIR]/eval/ (or appropriate result location)
+4. If results are available:
+   - Read the result JSON files
+   - Report metrics in a table (Borda scores, accuracy, etc.)
+   - Compute key comparisons between methods
+5. If all experiments in this batch are complete:
+   - git add -A && git commit -m "[COMMIT_MESSAGE]" && git push
+   - Report final summary
+6. Key question: [SPECIFIC ANALYTICAL QUESTION]
+
+If nothing has changed since the last check, respond with [SILENT].
+```
+
+### Monitoring Best Practices
+
+1. **Check processes first** — don't read results if the experiment is still running and results are incomplete
+2. **Read the log tail** — look for errors, progress indicators, completion messages
+3. **Count completed vs expected** — "45/150 problems done" is more useful than "some results exist"
+4. **Report in structured tables** — always include key metrics in a table
+5. **Answer the key question** — each experiment should have a specific analytical question to answer when done
+6. **[SILENT] for no-news** — suppress notifications when nothing has changed
+7. **Commit on completion** — every completed batch gets committed with a descriptive message
+
+### Example Monitoring Report
+
+```
+## Code Experiments (Haiku 3.5) - COMPLETE
+
+| Strategy | Pass Rate (150 problems) | vs Single |
+|----------|------------------------|-----------|
+| single_pass | 38.0% | — |
+| critique_revise | 35.2% | -2.8pp |
+| **autoreason** | **40.0%** | **+2.0pp** |
+| best_of_6 | 31.0% | -7.0pp |
+
+Key finding: Autoreason shows +2pp improvement over single pass, while 
+best-of-6 collapses due to single-public-test selection issue.
+
+Committed: `git commit -m "Add Haiku code results (150 problems, 4 strategies)"`
+Next: Run significance tests on these results.
+```
+
+---
+
+## Failure Recovery
+
+### Common Failures and Recovery
+
+| Failure | Detection | Recovery |
+|---------|-----------|----------|
+| **API credit exhaustion** | 402 errors in logs, incomplete results | Top up credits, re-run (skips completed work automatically) |
+| **Rate limiting** | 429 errors, slow progress | Add retry logic with exponential backoff |
+| **Process crash** | PID gone, log stops mid-problem | Re-run script (resumes from last checkpoint) |
+| **Wrong model ID** | Model not found errors | Fix ID (e.g., `claude-opus-4-6` not `claude-opus-4.6`) |
+| **Parallel slowdown** | Each experiment taking 2x longer | Reduce parallel experiments to 2-3 max |
+| **Security scan blocks** | Commands blocked by security | Use `execute_code` instead of piped `terminal` commands |
+| **Delegation failures** | `delegate_task` returns errors | Fall back to doing work directly |
+| **Timeout on hard problems** | Process stuck, no log progress | Kill, skip problem, note in results |
+| **Dataset path mismatch** | File not found errors | Verify paths before launching |
+
+### Retry Naming Convention
+
+When re-running failed experiments, use a suffix to track rounds:
+
+```
+logs/experiment_haiku_0_50.log       # Round 1
+logs/experiment_haiku_0_50_r2.log    # Round 2 (after credit exhaustion)
+logs/experiment_haiku_0_50_r3.log    # Round 3 (after bug fix)
+```
+
+### Pre-Flight Checklist
+
+Before launching any experiment batch:
+
+```
+Pre-Flight:
+- [ ] API credits sufficient for estimated calls
+- [ ] Model IDs correct (test with 1 problem first)
+- [ ] Output directory exists and is writable
+- [ ] Resume logic works (re-run won't overwrite existing results)
+- [ ] Log file path is unique (won't overwrite previous logs)
+- [ ] Dataset/task files are accessible
+- [ ] Config matches intended experiment
+```
+
+---
+
+## Task/Benchmark Design
+
+### Open-Ended Tasks (Subjective Evaluation)
+
+Design tasks that have clear objectives but subjective quality:
+
+```markdown
+# Task: [Title]
+
+## Context
+[Specific scenario with concrete details: company size, constraints, timeline]
+
+## Deliverable
+[Exact format and structure required]
+
+## Requirements
+- [Specific, measurable requirements]
+- [Not vague — "be comprehensive" is bad, "include exactly 6 sections" is good]
+```
+
+### Constrained Tasks (for Testing Scope Effects)
+
+Constrained tasks test whether methods respect scope boundaries. Design with:
+
+- **Fixed facts**: "Use only these N data points, add nothing else"
+- **Fixed deliverable**: Specific format (pitch, postmortem, memo — not "improve this")
+- **Fixed structure**: "These sections in this order, do not add/remove"
+- **Fixed change items**: "Address exactly these N points, nothing else"
+
+**Do NOT use word count as a scope constraint.** Word limits cause false convergence — outputs get rejected for length, not quality. Constrain scope (what to include) not length.
+
+### Example: Good vs Bad Constraints
+
+| Bad Constraint | Why | Good Constraint |
+|---------------|-----|-----------------|
+| "Max 500 words" | Judges reject for length | "Exactly 4 sections, each with 3 numbered items" |
+| "Be concise" | Too vague | "Each prohibition must reference a specific base fact" |
+| "Improve this" | Unbounded scope | "Write a 600-word incident postmortem with this exact structure" |
+| "Make it better" | No clear criterion | "Address exactly these 3 reviewer concerns" |
+
+---
+
+## Visualization Best Practices
+
+### Setup: SciencePlots + matplotlib
+
+Install SciencePlots for publication-ready defaults:
+
+```bash
+pip install SciencePlots matplotlib numpy
+```
+
+**Option A: SciencePlots styles** (recommended — handles most defaults automatically):
+
+```python
+import matplotlib.pyplot as plt
+import scienceplots  # registers the styles
+
+# Pick a style:
+# 'science'        — clean, serif fonts, suitable for most venues
+# 'science+ieee'   — IEEE-style (good for two-column papers)
+# 'science+nature' — Nature-style
+# Add 'no-latex' if LaTeX is not installed on the machine generating plots
+
+with plt.style.context(['science', 'no-latex']):
+    fig, ax = plt.subplots(figsize=(3.5, 2.5))  # single-column width
+    # ... plot ...
+    fig.savefig('paper/fig_results.pdf', bbox_inches='tight')
+```
+
+**Option B: Manual rcParams** (when you need full control):
+
+```python
+import matplotlib.pyplot as plt
+
+plt.rcParams.update({
+    'font.size': 10,
+    'font.family': 'serif',
+    'axes.labelsize': 11,
+    'axes.titlesize': 11,
+    'xtick.labelsize': 9,
+    'ytick.labelsize': 9,
+    'legend.fontsize': 9,
+    'figure.figsize': (3.5, 2.5),    # single-column default
+    'figure.dpi': 300,
+    'savefig.dpi': 300,
+    'savefig.bbox': 'tight',
+    'savefig.pad_inches': 0.05,
+    'axes.linewidth': 0.8,
+    'lines.linewidth': 1.5,
+    'lines.markersize': 5,
+    'axes.grid': True,
+    'grid.alpha': 0.3,
+    'grid.linewidth': 0.5,
+})
+```
+
+### Standard Figure Sizes (Two-Column Format)
+
+| Use Case | figsize | Notes |
+|----------|---------|-------|
+| Single column | `(3.5, 2.5)` | Fits in one column of two-column layout |
+| Double column | `(7.0, 3.0)` | Spans full page width |
+| Square (heatmap, confusion matrix) | `(3.5, 3.5)` | Single column |
+| Tall single (many rows) | `(3.5, 5.0)` | Use sparingly |
+
+### Colorblind-Safe Palette (Okabe-Ito)
+
+Use this palette for all paper figures. It is distinguishable by people with all common forms of color vision deficiency:
+
+```python
+COLORS = {
+    'blue':    '#0072B2',
+    'orange':  '#E69F00',
+    'green':   '#009E73',
+    'red':     '#D55E00',
+    'purple':  '#CC79A7',
+    'cyan':    '#56B4E9',
+    'yellow':  '#F0E442',
+    'black':   '#000000',
+}
+
+# As a list for cycling:
+COLOR_CYCLE = ['#0072B2', '#D55E00', '#009E73', '#E69F00', '#CC79A7', '#56B4E9']
+```
+
+Also differentiate lines by **marker and linestyle**, not just color:
+```python
+STYLES = [
+    {'color': '#0072B2', 'marker': 'o', 'linestyle': '-'},
+    {'color': '#D55E00', 'marker': 's', 'linestyle': '--'},
+    {'color': '#009E73', 'marker': '^', 'linestyle': '-.'},
+    {'color': '#E69F00', 'marker': 'D', 'linestyle': ':'},
+]
+```
+
+### Complete Example: Method Comparison Bar Chart
+
+```python
+import matplotlib.pyplot as plt
+import numpy as np
+
+try:
+    import scienceplots
+    style = ['science', 'no-latex']
+except ImportError:
+    style = 'default'
+
+with plt.style.context(style):
+    methods = ['Single Pass', 'Critique+Revise', 'Best-of-N', 'Ours']
+    scores = [73.2, 74.1, 68.5, 77.0]
+    errors = [2.1, 1.8, 3.2, 1.5]
+    colors = ['#56B4E9', '#E69F00', '#CC79A7', '#0072B2']
+    
+    fig, ax = plt.subplots(figsize=(3.5, 2.5))
+    bars = ax.bar(methods, scores, yerr=errors, capsize=3,
+                  color=colors, edgecolor='black', linewidth=0.5)
+    
+    # Highlight "Ours"
+    bars[-1].set_edgecolor('#0072B2')
+    bars[-1].set_linewidth(1.5)
+    
+    ax.set_ylabel('Pass Rate (%)')
+    ax.set_ylim(60, 85)
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+    
+    fig.savefig('paper/fig_comparison.pdf', bbox_inches='tight')
+```
+
+### Complete Example: Convergence/Trajectory Line Chart
+
+```python
+with plt.style.context(style):
+    fig, ax = plt.subplots(figsize=(3.5, 2.5))
+    
+    passes = np.arange(1, 16)
+    ours = [65, 72, 78, 82, 85, 87, 88, 89, 89.5, 90, 90, 90, 90, 90, 90]
+    baseline = [65, 68, 70, 71, 69, 67, 66, 65, 64, 63, 62, 61, 60, 59, 58]
+    
+    ax.plot(passes, ours, **STYLES[0], label='Ours', markersize=4)
+    ax.plot(passes, baseline, **STYLES[1], label='Critique+Revise', markersize=4)
+    
+    # Mark convergence point
+    ax.axvline(x=10, color='gray', linestyle=':', alpha=0.5, linewidth=0.8)
+    ax.annotate('Converged', xy=(10, 90), fontsize=8, ha='center',
+                xytext=(10, 93), arrowprops=dict(arrowstyle='->', color='gray'))
+    
+    ax.set_xlabel('Iteration')
+    ax.set_ylabel('Quality Score')
+    ax.legend(loc='lower right')
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+    
+    fig.savefig('paper/fig_trajectory.pdf', bbox_inches='tight')
+```
+
+### Output Rules
+
+- **Always save as PDF**: `fig.savefig('fig.pdf')` — vector graphics, sharp at any zoom
+- **Never save as PNG** for paper figures — raster PNGs look blurry when printed/zoomed
+- **Exception**: Screenshots, photographs, or pixel-art visualizations → PNG at 600 DPI
+- **Verify grayscale**: Print to grayscale PDF and check all information is still visible
+
+### Chart Types for Common Comparisons
+
+| Comparison Type | Chart | Notes |
+|----------------|-------|-------|
+| Method vs method | Grouped bar chart | Include error bars |
+| Across model sizes | Line chart with CI bands | Log scale for model size axis |
+| Ablation study | Stacked/grouped bar | Highlight removed component |
+| Trajectory/convergence | Line chart over iterations | Show winner per iteration |
+| Per-task breakdown | Heatmap or grouped bar | Show variance across tasks |
--- a/Show More
+++ b/Show More