qmd

A local, on-device search engine for markdown files (github.com/tobi/qmd), recommended in llm-wiki-gist for searching an llm-wiki once it outgrows the plain index.md approach.

Why it’s relevant

• Hybrid BM25 + vector search with LLM re-ranking, all on-device.
• Exposes both a CLI (the LLM can shell out to it) and an model-context-protocol server (the LLM can use it as a native tool).
• The source frames it as optional: at small scale the index file is enough; qmd is for when the wiki grows to where proper search is needed.

vs. gbrain

qmd and gbrain occupy the same niche — local search over a markdown corpus, exposed via an MCP server — but at different scope. qmd is hybrid BM25/vector + rerank search; GBrain wraps comparable retrieval and adds a typed-edge knowledge-graph and a synthesis/answer layer on top. For the llm-wiki pattern, qmd is the focused “just add search” upgrade; GBrain is the “replace the whole substrate” option.

This wiki currently uses index-only navigation (no search engine), per its design. Related: obsidian, llm-wiki, model-context-protocol, gbrain.