haiku.rag

haiku.rag is an open-source, local-first RAG system by ggozad that answers questions about your own documents with citations to page numbers and section headings. Its MCP server lets your agent search and read the knowledge base you've built with it. What you can do with it - Search your documents with hybrid vector and full-text search, or search by image - List documents, read a document's outline and pull out a specific section - Run sandboxed Python over the documents for counting and aggregation - Work across several databases at once, or limit a search to some of them How it connects - It's a local MCP server, started with uvx haiku-rag mcp --stdio. You need Python 3.12 or later. - Build the knowledge base first with the haiku-rag CLI: haiku-rag init, then haiku-rag add-src for each file, URL or folder. It ingests PDFs, office documents, HTML, Markdown and images with Docling. - By default it uses Ollama for embeddings, so documents stay on your machine; OpenAI-compatible and other providers are optional. - In Claude Code you can also install the repo's plugin, which registers the server and a skill: claude plugin marketplace add ggozad/haiku.rag, then claude plugin install haiku-rag. Good to know - The MCP server opens the database read-only. Adding or deleting documents happens through the CLI or the haiku-ingester service. - It's open source under the MIT licence.