mrag

Step-by-step procedures for operating mrag — a lightweight, local-first RAG knowledge base CLI. Use this skill when the user wants to build, search, inspect, evaluate, or serve a small-scale RAG pipeline from local documents (PDF / Markdown / text) with hybrid keyword+vector retrieval, contextual augmentation, parent-child retrieval, reranking, or Dify external knowledge API integration. Common user intents include: building a local RAG knowledge base from individual files or recursively filtered directory trees, querying or tuning retrieval quality on an existing knowledge base, inspecting how documents were chunked and augmented, reversibly excluding or permanently removing indexed documents, aggregating multiple knowledge bases into a registry for Agentic RAG workflows, and serving the knowledge base over the HTTP API.