Tiger Data · 2026-10-13 · notable
pg_vectorsearch — Tiger Data's new Postgres vector index beats HNSW at 100M
pg_vectorsearch is a new open-source PostgreSQL extension from Tiger Data. Its PRISM index served 100M vectors at 95% recall in 4 ms and built in 49 minutes, where pgvector's HNSW took 42 hours.
A partition-based vector index for Postgres that builds in minutes where HNSW takes days.
Key specs
| GitHub stars | 245 |
|---|---|
| Index build (100 m) | 49 min vs 42 h |
What is it?
pg_vectorsearch adds a new PRISM index type to PostgreSQL 18 for approximate nearest neighbor search. Tiger Data (the company behind Timescale) released it in alpha on 13 October 2026 under the PostgreSQL License. PRISM can index existing pgvector `vector` and `halfvec` columns, so tables and queries stay the same.
How does it work?
PRISM groups similar vectors into k-means partitions and stores compressed RaBitQ codes, about one bit per dimension, packed into database pages. A query walks a small tree of cluster centroids, then scans the nearest partitions with SIMD. Boundary vectors are stored in more than one partition to raise recall, and one setting, `prism.nprobe`, trades latency for recall (0 targets about 95%).
Why does it matter?
Teams that keep embeddings in Postgres often hit HNSW's slow builds and large memory needs. In Tiger Data's 100M-vector test, an r8i.4xlarge (128 GiB) machine served 99% recall in 21 ms with PRISM, while HNSW could not finish building on that machine. Tiger Data says its older pgvectorscale extension now moves to maintenance mode. The project is still alpha: deletes do not free space until REINDEX, and filtered search is planned for GA.
Who is it for?
developers running RAG and vector search on PostgreSQL
Try it
CREATE EXTENSION pg_vectorsearch; CREATE INDEX ON documents USING prism (embedding vec32_l2_ops);