VectorChord is the second-generation Postgres vector extension from TensorChord, built as the production successor to pgvecto.rs. Where pgvector showed that Postgres can do vector search and pgvecto.rs proved Rust can make it fast, VectorChord is engineered for the case where vectors are the dominant workload — billions of embeddings, low-latency hybrid queries, and the team still wants one database to back up, observe, and reason about.
The headline technology is RaBitQ quantization paired with IVF indexes. RaBitQ is a 2024 quantization technique that gives near-full-precision recall at a fraction of the memory cost, which is exactly the bottleneck most pgvector deployments hit before they migrate to a dedicated vector DB. By bringing it into Postgres, VectorChord lets teams scale past the point where pgvector typically forces a rewrite. Filtered search and hybrid retrieval with full-text indexes both work the way Postgres users expect.
VectorChord runs as a Postgres extension, which means it inherits Postgres replication, backup, point-in-time recovery, and the entire ecosystem of operational tooling. It supports pre-built Docker images, Kubernetes operators, and Aurora-style cloud Postgres deployments. The licensing is open source, and Tensorchord offers commercial support and managed deployments for teams that want a paid escape hatch without giving up Postgres-native operations.
