[HN]score: 0.30
Mnemo: Local-First LLM Memory Layer in Rust With Sub-50ms Retrieval
June 3, 2026
Mnemo is a self-hosted sidecar service that builds a persistent knowledge graph from conversation history using SQLite and petgraph, with entity extraction via any OpenAI-compatible LLM and context injection under 50ms. It ships as a single static binary, works with Ollama for fully local operation, and exposes a simple POST /ingest and /retrieve HTTP API.
HOW THIS AFFECTS YOU
●
builderYou can drop this into any LLM-backed app to add persistent memory without cloud dependencies — the single binary and OpenAI-compatible API make integration low-friction.
●
researcherThe entity deduplication and alias-merging approach over SQLite plus in-memory graph traversal is a concrete architecture worth benchmarking against vector-only memory systems.