LEANN
Compact local vector index for private RAG over personal data
What it is
LEANN is a vector database and RAG toolkit that recomputes embeddings on demand from a pruned graph instead of storing them, cutting index size by around 97% with little accuracy loss. It can index files, email, browser history, chat logs from apps like WeChat and iMessage, and code, then answer questions using local models through Ollama or other backends. It also ships an MCP server that gives coding agents such as Claude Code semantic code search. Developers and privacy-minded users run it on a laptop to search large personal datasets without sending anything to the cloud.