hermes-config/skills/devops/hermes-config-management/references/memory-provider-comparison.md

Memory Provider Comparison

Comparison of all 8 external Hermes memory providers and the built-in memory. Focused on free/self-hosted options.

Built-In (always active alongside any external provider)

Property	Value
Storage	Two flat markdown files (~/.hermes/memories/)
Limits	MEMORY.md: 2,200 chars, USER.md: 1,375 chars
Cost	Free — zero dependencies
Retrieval	Full scan via system prompt injection
Search	None (agent reads everything every session)
Notes	Always active. Use for always-in-context essentials only when using an external provider.

External Providers (only one active at a time)

Holographic ✅ Recommended for homelab

Property	Value
Storage	Local SQLite with FTS5
Hosting	Local only
API Key Required	No
LLM Costs	None — all operations are local DB queries
Dependencies	None (SQLite always available)
Setup	hermes config set memory.provider holographic
Tools	fact_store (9 actions), fact_feedback
Features	FTS5 search, trust scoring, entity resolution, HRR vector algebra
Config	plugins.hermes-memory-store.db_path (default: ~/.hermes/memory_store.db)
Notes	Zero ongoing cost. DB auto-creates on first use.

Hindsight (Local Embedded)

Property	Value
Storage	PostgreSQL (built-in daemon)
Hosting	Local daemon
API Key Required	Yes — LLM API key for memory extraction
LLM Costs	Yes — each retain/recall round burns tokens
Dependencies	LLM API key (OpenAI, Anthropic, OpenRouter, Ollama, etc.)
Setup	hermes memory setup → select hindsight → local_embedded
Features	Knowledge graph, entity resolution, multi-strategy retrieval, auto-retain/recall
Notes	Most powerful option but burns tokens. Free if pointed at local Ollama. Daemon auto-starts/auto-stops.

Hindsight (Cloud)

Property	Value
Storage	Managed cloud
Hosting	vectorize.io
API Key Required	Yes — from ui.hindsight.vectorize.io
Cost	Paid (free tier may exist)
Features	Same as local embedded but no daemon to manage

Honcho (Self-Hosted)

Property	Value
Storage	PostgreSQL + pgvector
Hosting	Docker/Python self-hosted
API Key Required	Yes — LLM API key for its own inference
LLM Costs	Yes — dialectic reasoning, dream consolidation, summarization
Dependencies	PostgreSQL, pgvector, LLM API key
Notes	Self-hosting doesn't eliminate LLM costs — Honcho runs its own inference for memory operations. Complex setup.

OpenViking

Property	Value
Storage	Context database (Volcengine/ByteDance)
Hosting	Local server process
API Key Required	Optional
LLM Costs	Varies — needs embedding + VLM models
Dependencies	pip install openviking, running server, embedding models
Setup	hermes config set memory.provider openviking, set OPENVIKING_ENDPOINT
Notes	Filesystem-style knowledge hierarchy. Requires running a server daemon.

RetainDB

Property	Value
Storage	Cloud API
Hosting	retaindb.com
API Key Required	Yes
Cost	$20/month subscription
Notes	Paid only. Hybrid search (Vector + BM25 + Reranking), 7 memory types.

ByteRover

Property	Value
Cost	API key required
Notes	Paid.

Supermemory

Property	Value
Cost	API key required
Notes	Paid.

Mem0

Property	Value
Cost	API key required or self-hosted
Notes	Self-hosted is complex. Managed tier is paid.

Decision Flow

1. Just want more space with zero cost/fuss? → Holographic (SQLite, local, free)
2. Want knowledge graphs and deep retrieval? → Hindsight local embedded (needs LLM API key, tokens cost)
3. Want to run a full server? → OpenViking or Honcho self-hosted
4. Happy to pay? → Hindsight Cloud, Mem0, RetainDB

Configuration Reference

memory:
  memory_enabled: true
  user_profile_enabled: true
  provider: holographic  # or builtin, honcho, mem0, openviking, hindsight, etc.

hermes memory setup         # interactive picker
hermes memory status        # check active provider
hermes memory off           # disable external provider (reverts to built-in only)