AI LocalBase
A local-first, self-hosted retrieval-augmented question-answering system that indexes local documents into a vector database and answers questions through Ollama or an OpenAI-compatible model, with a built-in MCP server.
AI LocalBase is a local-first retrieval-augmented generation (RAG) system for connecting local documents to vector search and a large-model chat workflow. It provides a web interface for knowledge-base management, document upload, retrieval-augmented question answering, persisted chat history, and model configuration for either Ollama or an OpenAI-compatible endpoint. Supported document formats include text, Markdown, PDF, Excel, and CSV.
Retrieval features include automatic text chunking and batch embedding, dynamic candidate recall, keyword-coverage reranking, redundancy removal, a second recall pass for low-confidence queries, embedding and optional semantic caching, and optional hybrid search, semantic reranking, query rewriting, and context compression. It can also generate a RAG evaluation dataset from an existing knowledge base's documents for small-sample testing, and exposes an MCP server so external agents and tools can query it.
It is built with a Go and Gin backend, a React and TypeScript frontend, and Qdrant as its vector database, and deploys through Docker Compose.