Index markdown and text documents for local hybrid search with reranking.
npx skills add https://github.com/MarbleSodas/Mavis --skill qmd-marblesodas
Please help me install this Agent Skill. Skill: qmd Source: https://github.com/MarbleSodas/Mavis/tree/main/optional-skills/research/qmd Command: npx skills add https://github.com/MarbleSodas/Mavis --skill qmd-marblesodas
Finding the right note, doc, or meeting transcript across a growing personal knowledge base is slow and unreliable when you only have keyword search or vague memory.
Ask: "Search my qmd collection for the decision we made about the API redesign, and return the most relevant excerpts."
High-intent search queries and answers about installing and using this skill.
You can search markdown notes and meeting transcripts locally by indexing text documents into a collection and using hybrid retrieval modes that combine BM25 keyword matching with vector semantic search and LLM reranking.
Hybrid RAG improves local knowledge base search by combining BM25 keyword matching with vector semantic search and LLM reranking, returning highly relevant excerpts for fast keyword lookup or semantic question answering on personal notes.
Yes, you can use the integrated MCP server to let an AI agent automatically call search and retrieval tools to query your indexed markdown and text document collections directly.
Yes, local hybrid search across personal knowledge bases requires Node.js 22 or higher, along with local collection and index setup, and optional SQLite extension support on macOS.
The best way to find specific decisions in meeting transcripts is using the query mode, which applies hybrid BM25 and vector search with reranking to return the most relevant excerpts from your indexed documents.
Limitations include requiring local collection and index setup for your documents, needing Node.js 22+ on your machine, and configuring optional SQLite extension support if you are running the search on macOS.
zeronx798Index and retrieve local markdown files using hybrid BM25, vector, and LLM reranking search.
zilliztechDecompose questions, retrieve across documents, and synthesize cited answers.
ybondaStore and retrieve cross-project memories via local semantic search.
icyzhIndex and search local markdown notes using BM25 and semantic embeddings.
