Mnemosyne

A local, teaching-first RAG pipeline (Ollama + LangChain + FAISS) that turns any model into an instant expert on documents it has never seen, with no fine-tuning, no API keys, and nothing leaving the box.

Point Mnemosyne at a corpus; it embeds and indexes it, and a small local model then answers questions about it with inline citations. It ships a CLI, an MCP stdio server, and an HTTP server. The full story (how RAG works, the design choices, the knowledge-pack model) is in the GitHub README.

Install

pip install "mnemosyne-rag[cpu]"

FAISS is not a core dependency. The cpu extra pulls faiss-cpu from PyPI; the preferred path installs faiss-cpu or faiss-gpu from conda-forge (see environment.yml in the repo). The distribution is mnemosyne-rag; the import package and CLI are mnemosyne.

Quickstart

Requires Ollama running locally.

ollama pull bge-m3          # embeddings
ollama pull qwen2.5:1.5b    # tiny chat model

mnemosyne ingest ubiquiti                                 # build the index for a pack
mnemosyne ask ubiquiti "How do I adopt a UniFi switch?"   # grounded answer + citations
mnemosyne chat ubiquiti                                   # interactive, with history
mnemosyne packs                                           # list packs + index status

Serve

mnemosyne-mcp    # MCP stdio server (list_packs / ask / search) for coding agents
mnemosyne-http   # FastAPI server on 127.0.0.1:8088 (GET /health, POST /ask, POST /search)

Learn more

• Full README: the idea, how it works, technology choices
• RAG-101: RAG from first principles
• Knowledge packs: build your own expert
• Architecture: modules and data flow

License

Apache-2.0.