# bigRAG

An open-source, self-hostable retrieval-augmented-generation platform that ingests documents, embeds and indexes them with Turbopuffer, and serves semantic, keyword and hybrid search behind one REST API.

bigRAG is a self-hostable RAG platform: documents are uploaded in formats such as PDF, DOCX, PPTX, HTML, Markdown and images, automatically chunked and embedded, and made retrievable through semantic, keyword and hybrid search backed by Turbopuffer, which stores vectors, chunk text and metadata filters per collection, with each collection mapped to its own namespace for isolated writes, exports and deletion.

Embedding providers include OpenAI, OpenAI-compatible gateways, Cohere and Voyage, with named, reusable embedding presets, and Cohere reranking can improve result relevance; a query can also span multiple collections at once. Other features include a stateless, backend-grounded chat playground with citations, bulk upload and batch operations, a connector that mirrors an S3 or R2 bucket into the platform, admin accounts with scoped API keys and full audit logs, a retrieval-evaluation runner that tracks recall, mean reciprocal rank and normalized discounted cumulative gain against a labeled set, and HMAC-signed webhooks with retries.

It ships as a single docker-compose deployment with an API, worker, admin UI, Postgres and Redis, plus TypeScript and Python client SDKs and an MCP server for clients such as Claude Desktop and Cursor. It is MIT licensed.

Website: https://rag.computer