mnemo – local-first AI memory layer for LLMs with persistent knowledge graph in SQLite and Rust

mnemo is a sidecar service that extracts named entities and relationships from conversations using an LLM, builds a persistent knowledge graph in SQLite, and injects relevant context back into future prompts automatically in under 50ms. It works with Ollama, OpenAI, Anthropic, or any OpenAI-compatible API, and ships as a single static binary with zero cloud dependency.

The system runs a 6-stage retrieval pipeline: full-text chunk search → entity name search → graph expansion (breadth-first search over the knowledge graph) → relation filter → score and rank → assemble context. Available as Docker + Ollama (free), static binary, CLI tool, or Python SDK.

Performance on Apple M2 in debug build: entity insert at ~0.12 ms (~8,300 ops/s), full-text chunk search at ~0.28 ms (~3,500 ops/s), and full retrieval pipeline at ~4.2 ms (~238 ops/s). Release build is 3–5× faster. Includes 122 Rust tests, 21 Python tests, and 12 benchmarks.