tobi/qmd: On-Device Doc Search Engine with BM25 + Vector + LLM Re-Ranking and MCP Server — 18.5K Stars

GitHub·high signal

QMD is a local-first search engine that indexes markdown docs, meeting transcripts, and knowledge bases using a triple-layer approach: BM25 full-text, vector semantic search via node-llama-cpp with GGUF models, and LLM re-ranking — all running on-device. It ships with an MCP server for direct integration with AI agents and Claude Desktop. For anyone building agent-accessible documentation systems, qmd solves the local retrieval problem without any cloud dependency, and the MCP transport makes it immediately usable by coding agents.