January 26, 2026
Vault wars heat up
QMD - Quick Markdown Search
Local note search goes turbo — Obsidian crowd chants “plugin when”
TLDR: QMD promises fast, private, on-device search that understands your notes beyond exact keywords. The community loves the idea but debates performance on huge Obsidian vaults and begs for a plugin, while the creator insists it’s practical, not overkill—making local AI search feel finally within reach.
QMD is the new “search everything on your own laptop” tool, and the community is buzzing. The creator popped into the thread with a casual “hi orange page” and insisted it’s not overkill—just a smart blend of old-school keyword matching, meaning-aware search, and AI ranking that all runs locally. Fans of Obsidian (the popular note app) swooned, saying their hundreds of markdown files finally have a chance at finding “conceptual connections,” not just exact words. One big question: Can it handle giant note vaults? The crowd wants receipts, and they want them fast. The hottest debate: is this the note-search glow-up we’ve been waiting for, or yet another power tool that belongs in a lab? The author says it’s built from real-world search best practices and is already working on better query expansion and reranking—translation: it’s learning to guess what you meant, not just what you typed. Meanwhile, a chorus is chanting “Obsidian plugin when?” and dropping jokes about their chaotic docs named “finalFINAL2.md.” Privacy lovers are cheering the local-only angle, while skeptics worry it’s doing too much. The vibe? Excited, a little chaotic, and very ready to point QMD at notes, meeting transcripts, and the digital junk drawer. Check the repo: QMD.
Key Points
- •QMD is a local search engine for markdown notes, transcripts, and documentation, supporting keyword and natural language queries.
- •It combines BM25, semantic vector search, and LLM reranking, running on-device via node-llama-cpp with GGUF models.
- •Command-line workflow includes creating collections, adding context, generating embeddings (`qmd embed`), and multiple search modes (BM25, vector, hybrid).
- •QMD offers agent-friendly outputs (`--json`, `--files`) and an MCP server exposing search and retrieval tools, with example configs for Claude Desktop and Claude Code.
- •The hybrid architecture uses Qwen3-0.6B for query expansion, parallel BM25/Vector retrieval, RRF fusion with bonuses, and final reranking.