“QMD is an implementation of the best practices that I picked up in meetings with teams that work in search and retrieval. I tried to make it not overkill and keep things local. — Tobias Lütke, HN comment Jan 2026”
You know that feeling when you have hundreds of markdown notes scattered across folders, and Obsidian's search finds exact keyword matches but completely misses the conceptual connection you're looking for? You remember writing something about authentication flows, but searching 'auth' returns 50 files and none of them are what you need. QMD gives you semantic search that understands 'how do users log in' should surface your OAuth implementation notes, even if those exact words never appear together.
Think of QMD like having three librarians working together: one who's great at finding exact keyword matches (BM25), one who understands the meaning behind your question (vector embeddings), and one who's a subject expert who double-checks the results (LLM reranking). You point it at your folders, it indexes everything into a local SQLite database, generates vector embeddings for each chunk of text, and when you search, it runs all three approaches in parallel then blends the results. The clever bit: it chunks code files at function and class boundaries using tree-sitter, so a 500-line Python file becomes searchable by individual functions, not as one giant blob.
If you're a developer with hundreds of markdown notes in Obsidian, VS Code, or scattered across folders, and you've ever been frustrated that search only finds exact keyword matches — this is for you. Also valuable if you're building AI agents and need a local knowledge base they can query via MCP. Not for you if you need cloud sync, team collaboration, or a GUI — this is CLI-first, single-user, local-only.
Yes, worth trying if you have a local markdown collection and want semantic search. The project is in active development (v2.1.0 released April 5, 2026 with 25+ community PRs), has clear momentum (19k stars in 4 months), and the architecture is well-documented. The main caveats: macOS requires Homebrew SQLite, Windows has reported CUDA issues (Issue #519), and there's no cloud API option — it's local-only by design. If you're on macOS or Linux with a GPU, it's a solid weekend experiment.
Deep-dive insight, Easy and Pro modes, plus action playbooks — the full breakdown is one tap away.