pgvector

pgvector is a PostgreSQL extension that lets you store embeddings in an ordinary table column, next to the rows they describe, and find the nearest ones with SQL. Instead of running a separate vector store and keeping IDs in sync between two systems, vectors live in the same database as your users, documents and permissions.

You enable it with CREATE EXTENSION vector, add a vector(n) column, and order a query by a distance operator (cosine, L2, inner product and others) with a LIMIT. Without an index Postgres compares every row and returns exact results; adding an HNSW or IVFFlat index switches to approximate search, which is faster but can return slightly different rows. Any language with a Postgres client works, and the project keeps helper libraries for Python, Node.js, Ruby, Go, Rust, Java and many more.

What a small team gets is Postgres itself: transactions, JOINs and WHERE clauses in the same query, plus WAL replication and point-in-time recovery. It also offers half-precision, binary and sparse vector types, and pairs with Postgres full-text search for hybrid queries. It runs anywhere Postgres 13+ runs (Docker, Homebrew, apt), and many hosted Postgres providers ship it preinstalled.

The limits: indexes cover up to 2,000 dimensions (4,000 with half precision); filters on an approximate index are applied after the index scan, so selective filters can return fewer rows unless you turn on iterative scans; and scaling beyond one machine means read replicas or a sharding tool such as Citus.