LEANN

Why we included this project

LEANN is a good fit for developers who want semantic search and retrieval over very large personal collections while keeping everything on their own machine. It stores graph-based indexes instead of a full embedding per chunk and recomputes embeddings on demand, which is how it fits millions of documents in a few gigabytes and cuts storage by about 97%. Alongside the engine, it ships ready-made workflows for searching your file system, email, chat history, or browser history, and it supports MCP natively, so tools like Claude Code can use it as a semantic search service. It also plays well with LangChain, LlamaIndex, and Ollama, so it slots into stacks people already run. If your constraints are storage cost and keeping data off the cloud, this gives you a concrete, offline-first route to real RAG rather than a library you'd have to wire together yourself.