bigRAG Docs

bigRAG Docs is an open-source, self-hosted platform designed for retrieval-augmented generation (RAG) workflows centered on document ingestion, search, and retrieval. It enables users to upload documents in a wide range of formats, including PDF, DOCX, PPTX, HTML, Markdown, and images with OCR, utilizing Docling for universal document parsing. The system is built around Turbopuffer, which provides semantic, keyword, and hybrid search capabilities from a unified chunk store, and applies Reciprocal Rank Fusion for handling mixed queries.

The platform supports a variety of embedding models, including OpenAI, Cohere, Voyage, and other OpenAI-compatible models, with the flexibility to configure different providers for each collection and mix them across collections. Namespace isolation is implemented by mapping each collection to a Turbopuffer namespace, ensuring that vector writes, keyword indexes, exports, and deletes remain scoped to their respective collections. bigRAG Docs also offers HMAC-signed webhook payloads with automatic retries for both collection and connector events, facilitating integration with external systems.

Users can interact with bigRAG Docs through a zero-dependency TypeScript SDK, compatible with Node.js, browsers, Deno, and Bun, providing full type safety and automatic retries. Alternatively, a REST API is available, with comprehensive documentation accessible via Swagger. The platform is designed for straightforward deployment using Docker Compose; users are required to run the API, an admin UI, Postgres, and Redis, and then connect to Turbopuffer for managed vector and full-text retrieval.

bigRAG Docs is suitable for teams or individuals who require a self-hosted solution for document search and retrieval workflows on their own infrastructure. Its open-source nature allows for flexible deployment and customization, supporting a variety of document management and search applications.