R2R

R2R packages a full retrieval-augmented generation backend behind a RESTful API, aimed at teams that want production features rather than a notebook demo. Ingestion handles text, PDFs, images, audio, and JSON, feeding hybrid search that combines semantic vectors with keyword matching, and the engine can automatically build knowledge graphs with entity and relationship extraction to support graph-augmented retrieval. A Deep Research API runs multi-step agentic reasoning that fetches evidence before answering, and the server includes authentication plus user and collection management, so multi-tenant deployments need no extra plumbing. A lightweight mode installs with pip install r2r, while the full stack deploys through Docker Compose with PostgreSQL and pgvector handling both relational data and vectors; Python and JavaScript SDKs cover both modes. LLM and embedding providers are pluggable, including locally hosted models, which keeps the whole system self-hostable on private infrastructure. The code is MIT licensed and the SciPhi-AI repository holds around 7,900 stars, though commit activity has slowed since late 2025.