VectorChord

Why we included this project

VectorChord is a Postgres extension that brings approximate nearest-neighbor search to an existing database, no separate service required. It compresses vectors with RaBitQ quantization and does automatic reranking, which keeps storage costs down while maintaining recall. The project is the successor to pgvecto.rs and stays compatible with pgvector's types and syntax, so switching often means no query rewrites. The README's numbers are compelling: about 400,000 vectors per dollar of storage, and a hundred million vectors on a single i4i.xlarge instance, with index builds taking around 20 minutes. If you're weighing whether to consolidate vector workloads into your current Postgres or add a dedicated vector store, the benchmarks and disk-usage figures are worth a look.