pgvectorscale
Tiger Datapgvectorscale is a Postgres extension that complements pgvector with a StreamingDiskANN index, inspired by Microsoft's DiskANN research, and a statistical binary quantization scheme for cost-efficient storage at scale. It also adds label-based filtered vector search, combining similarity search with metadata filtering in a single index. Written in Rust using the PGRX framework and developed by Tiger Data (formerly Timescale).
Openness
5 high confidence- license
- PostgreSQL(OSI)
- source
- public
- core features withheld
- no
pgvectorscale uses the PostgreSQL License, the same OSI-approved terms as pgvector, and the public, unarchived repository builds the whole extension — StreamingDiskANN, statistical binary quantization and label-based filtered search — with no separate paid tier gating any of it. Tiger Data's hosted Timescale Cloud service sits beside the extension rather than withholding any part of it.
- https://api.github.com/repos/timescale/pgvectorscale recorded 2026-09-03
Repo metadata - license spdx_id PostgreSQL, private false, archived false, default branch main - for timescale/pgvectorscale.
- https://raw.githubusercontent.com/timescale/pgvectorscale/main/LICENSE recorded 2026-10-07
The PostgreSQL License - permissive terms carrying a Tiger Data disclaimer of warranty and liability, structurally identical to the license pgvector records.
- https://raw.githubusercontent.com/timescale/pgvectorscale/main/README.md recorded 2026-10-07
README describes StreamingDiskANN indexing, statistical binary quantization and label-based filtered vector search shipped in the one repository, installed via `CREATE EXTENSION vectorscale`, with a separate paid "private beta" Timescale Cloud signup offered alongside the extension rather than gating any part of what is in the tree.
Adoption
2 low confidencepgvectorscale installs as a Postgres extension via Docker or from source rather than through a PyPI or npm package, and no distribution channel publishes a comparable download count, so GitHub stars are the only adoption signal available. A star reflects interest rather than confirmed use, so it cannot show how many databases are actually running pgvectorscale.
- https://api.github.com/repos/timescale/pgvectorscale recorded 2026-09-03
Repo metadata - stargazers_count = 3,119 - for timescale/pgvectorscale.
Capability
2 medium confidencepgvectorscale ships a faster index type, StreamingDiskANN, plus a quantization scheme, over vector columns that pgvector already defines and stores; it holds no storage of its own. That is a narrower scope than pgvector, which is a working single-purpose store and index in its own right, which is why pgvectorscale sits one step behind it.
- https://raw.githubusercontent.com/timescale/pgvectorscale/main/README.md recorded 2026-10-07
README states pgvectorscale "complements pgvector... and introduces the following key innovations for pgvector data", requires `CREATE EXTENSION vectorscale CASCADE` to automatically install pgvector, and indexes the vector columns pgvector defines rather than defining or storing data itself.
Verified 2026-09-27