pgvector

Open-source vector similarity search as a Postgres extension (PostgreSQL License, permissive MIT-like; ~22k stars). Adds a vector column type and three distance operators: <-> (L2/Euclidean), <=> (cosine), and <#> (negative inner product), with exact search plus approximate IVFFlat and HNSW indexes. The point is that it is just Postgres: the database you likely already run handles vector search over thousands to a few million rows without a second system to deploy, back up, and secure (this is what Supabase's vector offering is built on). Honest catch: the operator must match how your embedding model was trained, and the approximate indexes trade recall for speed, a dial you tune (ivfflat.probes / hnsw.ef_search) and must measure on your own data.