A self-hosted MCP server (plus REST API) that adds persistent, personalized long-term memory to any MCP-compatible assistant (Claude Code, ChatGPT, Cursor, Open WebUI, and more). A single unified LLM call performs fact extraction, metadata classification, deduplication, and contradiction resolution at once. Two-tier design: fast searchable summaries in a vector store, plus a detailed artifact store retrieved on demand.
- Storage
- Qdrant for vectors (searchable fact summaries) and S3/MinIO for the artifact store (reports, code, research). Per-user isolation with API-key or JWT auth. Memories carry configurable TTL (context ~7 days, episodic ~90 days) with reinforcement so frequently accessed memories stay alive; a three-phase health checker (fsck) detects duplicates, contradictions, quality issues, and prompt injection.
- Retrieval
- AI-powered multi-query semantic search: generates multiple queries covering different angles, runs them in parallel, and reranks by relevance, with temporal awareness ('what did I decide last week?'). At conversation start the agent loads pinned core memories plus recent context; relevant memories surface automatically during the session.
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Open-source Apache-2.0, free and self-hosted (`uvx mnemory`); needs an OpenAI-compatible LLM/embedding key, with Qdrant + S3/MinIO for production. No paid tier. · Free / OSS
- GitHub stars
- 182
- Last release
- 2026-05-07
- Last commit
- 2026-06-09
- First catalogued
- 2026-06-28
Strengths
- Single-LLM-call extraction + classification + dedup + contradiction resolution keeps writes cheap and self-consistent
- Two-tier memory (vector summaries + on-demand artifact store) plus a built-in fsck health checker that also flags prompt injection
- Self-hosted and private by default: dual MCP + REST interfaces, per-user isolation, Prometheus metrics, management UI, 10+ client integrations
Watch out
- README shows a self-reported LoCoMo table (73.2 overall) — record via harvest-benchmarks with selfReported:true, not here
- GitHub classifies the license 'NOASSERTION', but the LICENSE file is verbatim Apache-2.0 (confirmed) — no added rider found
- Small project (~180 stars); production setup expects Qdrant + S3/MinIO infrastructure
Best for
- Self-hosters wanting a private, MCP-native memory server with automatic fact extraction, dedup, and contradiction handling
Benchmark results
No sourced results yet.
Sources
- mnemory README (vendor)
- GitHub API repo metadata (stars, v1.12.1 release) (third-party)
- LICENSE (verbatim Apache-2.0 text; GitHub auto-classifies NOASSERTION) (vendor)
Last verified 2026-06-28 · updated by discover-frameworks