tobi/qmd

QMD (Query Markup Documents) is a command-line search engine that runs entirely on your own computer and lets you search through your personal collection of Markdown files, notes, meeting transcripts, documentation, knowledge bases, or any text files. The core problem it solves is that when you have hundreds of notes scattered across folders, finding what you're looking for by scrolling through files becomes impractical. It combines three different search techniques to give the most relevant results. BM25 is classic keyword matching, similar to how search engines worked before AI, fast and precise for exact phrases. Vector search is semantic search, meaning it understands meaning rather than just matching words, so searching "how to deploy" also finds documents about "shipping to production." LLM re-ranking uses a language model to reorder the top results by relevance after the initial search. All three run locally using GGUF model files (a compact format for running AI models on a laptop), so none of your data ever leaves your machine. You organize your files into named collections (like "notes" or "meetings"), generate search embeddings once, and then query from the command line. QMD is also designed to integrate with AI coding assistants and agents, it exposes an MCP (Model Context Protocol) server so tools like Claude can use it to look up relevant documents during conversations. A developer, researcher, or knowledge worker who takes many notes and wants a fast, private, AI-enhanced way to search them would use QMD. It is written in TypeScript and runs on Node.js or Bun.