PostgreSQL + pgvector

pgvector is an open-source extension that adds vector similarity search capabilities to PostgreSQL. It enables storing embeddings and performing similarity queries directly in your existing PostgreSQL database, without needing a separate vector database.

Key features include: support for exact and approximate nearest neighbor search (IVFFlat, HNSW indexes), integration with standard SQL queries, L2 distance, inner product, and cosine distance metrics, and seamless combination with relational data and filtering.

For AI engineers, pgvector offers: reduced infrastructure complexity (one database instead of two), familiar SQL interface, transaction support across vector and relational data, and cost savings for moderate scale. Trade-offs include: performance at very large scale compared to dedicated vector databases, and fewer specialized features. pgvector is excellent for many production RAG applications, especially when teams already use PostgreSQL.