Search markdown knowledge bases using BM25, vector, and hyde query types.
npx skills add https://github.com/thoroc/thoroc.github.io --skill qmd-thoroc
Please help me install this Agent Skill. Skill: qmd Source: https://github.com/thoroc/thoroc.github.io/tree/main/.agents/skills/qmd Command: npx skills add https://github.com/thoroc/thoroc.github.io --skill qmd-thoroc
lex (BM25 keywords), vec (natural language questions), and hyde (hypothetical answer passages) in a single search, with reciprocal rank fusion and reranking. - Multiple Interfaces: Query via MCP tools (query, get, multi_get, status), the qmd CLI, or an HTTP API on port 8181. - Collection Filtering & Intent Steering: Restrict searches to named collections and disambiguate vague terms with an intent signal. - Use Case: Ask "how does the rate limiter handle burst traffic" and retrieve the relevant chunk from your engineering docs, combining a keyword query for rate limiter with a semantic question for best recall. ## Quick Start Ask the assistant to search your notes for how the payment service processes refunds using qmd.High-intent search queries and answers about installing and using this skill.
Run qmd query followed by your question, or use the MCP query tool with structured searches. Combine lex for exact keywords, vec for natural language questions, and hyde for hypothetical answer passages to maximize recall.
lex uses BM25 keyword matching for exact terms and code identifiers, vec uses vector embeddings for natural language questions, and hyde matches a 50-100 word hypothetical answer passage. Combining all three gives the best recall on complex topics.
Install qmd globally with npm install -g @tobilu/qmd, add a collection with qmd collection add, run qmd embed, then register "qmd mcp" as an MCP server command in your Claude Code or Claude Desktop configuration file.
No results usually means collections are not registered or embeddings are missing. Run qmd collection list to verify collections exist and qmd embed to build the vector index before searching.
Yes, qmd works as a standalone CLI with commands like qmd query, qmd search, qmd get, and qmd multi-get. It also exposes an HTTP API on port 8181 via qmd mcp --http for integration with other tools.
pedronauckSearch markdown knowledge bases using BM25, vector, and hyde query types via QMD.
tocadaoncamodExtract insights, summaries, and cited answers from PDF documents via RAG.
tc9011Search personal markdown knowledge bases with hybrid BM25 and vector retrieval.
massgenLocate text and structural code patterns across large codebases using ripgrep and ast-grep.