qmd · levineam — Claude Code skill: search your markdown notes locally

What qmd does

qmd is a local search tool for markdown notes, docs and knowledge bases, and this skill teaches an agent how to drive it. You register a folder as a collection, run the indexer, and searches stay on your machine.

Keyword lookups go through qmd search, which is built on BM25 and usually returns right away. When keywords miss, qmd vsearch falls back to vector similarity — expect roughly a minute on a cold start, since a local model such as Qwen3-1.7B has to load first. The hybrid qmd query mode adds LLM reranking on top and the skill advises skipping it for interactive work: it runs longer and can time out. Agents get machine-readable output from --json and --files, and qmd get or qmd multi-get return whole documents.

Bun 1.0.0 or newer and the qmd binary are required, and $HOME/.bun/bin has to sit in PATH or the commands simply won't resolve. On macOS you also need brew install sqlite for the SQLite extensions. Vector and hybrid search stay off until you run qmd embed once, and you repeat it after edits. Models land in ~/.cache/qmd/models/, which XDG_CACHE_HOME can move. Note the scope: this covers notes, not source trees — for code you want a code search tool.