I built RAGGuard because most RAG systems have a security gap: they filter documents after retrieval, meaning unauthorized data has already hit the retrieval layer. RAGGuard filters during vector search at the database level.
- Works with 14 vector DBs (Qdrant, ChromaDB, Pinecone, pgvector, etc.) - Integrates with existing auth (OPA, Cerbos, OpenFGA, custom RBAC) - Drop-in replacement for LangChain/LlamaIndex retrievers
GitHub: https://github.com/maximus242/ragguard
Would appreciate any feedback on the approach or API design.