Combine BM25 keyword matching, vector search, and LLM reranking for hybrid local retrieval.
npx skills add https://github.com/JamesFincher/gengar --skill qmd-jamesfincher
Please help me install this Agent Skill. Skill: qmd Source: https://github.com/JamesFincher/gengar/tree/main/optional-skills/research/qmd Command: npx skills add https://github.com/JamesFincher/gengar --skill qmd-jamesfincher
Finding the right note, transcript, or doc inside a large personal knowledge base is slow and error-prone, especially when you do not remember exact keywords.
Real-world use: you ask for decisions made about a project’s API redesign, and qmd finds relevant meeting transcripts and docs even if your phrasing differs from what was written.
Run qmd query "what decisions were made about the API redesign" on your indexed collections to get the most relevant documents with reranked results.
High-intent search queries and answers about installing and using this skill.
Local semantic search across markdown notes and meeting transcripts is performed by combining BM25 keyword matching, vector embeddings, and LLM-based reranking to retrieve relevant documents from indexed collections via CLI or MCP.
Yes, you can use an MCP server to let your agent query a local knowledge base. The MCP integration allows agents to perform on-demand retrieval across indexed markdown and text collections using hybrid search.
The best way to find information without exact keywords is hybrid local retrieval, which matches semantic meaning using vectors alongside BM25 keyword matching and then applies LLM reranking to return the most relevant documents.
Yes, you need Node.js version 22 or higher to index markdown files for local document retrieval. The system installs models locally and uses an embedded SQLite index to store and query your collections.
LLM-based reranking improves knowledge base search results by reordering the initial BM25 and vector matches using a language model, ensuring the returned documents best align with the specific query intent and contextual relevance.
Yes, you can index markdown and text files from multiple project documentation directories as named collections, optionally adding context descriptions to improve search relevance across different knowledge bases.
zeronx798Index and retrieve local markdown files using hybrid BM25, vector, and LLM reranking search.
thecloudtipsSynthesize sourced answers with inline citations and confidence ratings.
zerwizSearch local knowledge bases with hybrid BM25 and vector retrieval.
melikhanmutluOptimize embedding, indexing, and retrieval strategies for RAG systems.
