VectorChord

Postgres can hold a billion vectors with VectorChord, TensorChord's extension for scalable vector search and the designated successor to pgvecto.rs. Its index pairs clustering with RaBitQ quantization, trading full-precision storage for compact bit representations that make search disk-friendly: the project reports indexing 100 million 768-dimension vectors in about 20 minutes, serving that scale from a 247 dollar per month AWS i4i.xlarge instance, and fitting 6x more vectors per dollar than Pinecone, with native 4-bit and 8-bit vector types available. Queries stay inside PostgreSQL, so filtering, joins, and transactions combine with similarity search in ordinary SQL. Deployment is easiest through the prebuilt Docker images at ghcr.io/tensorchord/vchord-postgres, with the current release at v1.1.1 and documentation at docs.vectorchord.ai. The code is dual licensed under AGPL-3.0 and the Elastic License 2.0. Around 1,800 GitHub stars and steady adoption among teams scaling past pgvector's comfort zone make it one of the notable Postgres vector options.