01
Hybrid Search
BM25 full-text (FTS5) combined with vector semantic similarity (sqlite-vec). Fused ranking for precise memory retrieval.
Give Claude Code long-term memory that survives across sessions. Hybrid BM25 + vector search, auto-clustering, and confidence scoring — all stored locally.
A complete memory lifecycle — from structured storage to intelligent retrieval, clustering, and time decay.
BM25 full-text (FTS5) combined with vector semantic similarity (sqlite-vec). Fused ranking for precise memory retrieval.
Pull via MCP tools on demand, plus auto-push through hooks on user prompt, pre-tool, and post-tool events.
Memories auto-structured into what / when / do / warn XML format by LLM for consistent retrieval quality.
Similar memories grouped automatically. Mature clusters promoted to reusable skill memories that never decay.
Each memory has a confidence score adjusted through validation feedback and usage, with configurable time decay.
All data in local SQLite. Your memories never leave your machine. Zero cloud dependency for storage.
On-demand MCP tools paired with automatic hook injection across the full Claude Code lifecycle.
From save to structure, embed, search, validate, cluster, promote, and decay.
Clone, configure, and start. Then add MCP + hooks config to your Claude Code project.
git clone https://github.com/ MIMI180306/claude-persistent-memory.git cd claude-persistent-memory npm install
cp config.default.js config.js # Set Azure OpenAI credentials export AZURE_OPENAI_ENDPOINT=... export AZURE_OPENAI_KEY=... export AZURE_OPENAI_DEPLOYMENT=...
# Terminal 1: Embeddings (~2GB RAM) npm run embedding-server # Terminal 2: LLM proxy npm run llm-server
Each type has its own half-life. Skills promoted from mature clusters never expire.