crate

Why we included this project

CrateDB stores vector embeddings in the same rows as the structured, time-series, JSON, and full-text data they describe, so a single SQL statement can run k-nearest-neighbor similarity search and hybrid queries without a separate vector store. That is a real convenience for RAG pipelines and semantic search, where relevance ranking has to sit alongside filters, time constraints, and aggregations. It speaks the PostgreSQL wire protocol and offers a plain HTTP endpoint, so existing SQL tooling and Python pipelines connect without proprietary SDKs. Teams already running analytics or IoT workloads on a distributed database get vector search as an add-on rather than a second system to operate.