pgvectorscale

What it does

Pgvectorscale is an open-source PostgreSQL extension, released under the PostgreSQL License, that builds on pgvector's data type and distance functions. It adds a StreamingDiskANN approximate nearest neighbor index and Statistical Binary Quantization, so existing embedding columns can be indexed with a single CREATE INDEX statement and queried inside a standard PostgreSQL database.

The index keeps part of its structure on disk instead of memory, which the announcement says lowers the cost of storing and searching large vector sets because SSDs are cheaper than RAM. Streaming filtering lets secondary metadata filters apply during similarity search without the fixed retrieval cutoff the announcement attributes to pgvector's HNSW index. This is useful for retrieval-augmented generation workloads that combine tags or other filters with vector similarity.

It suits teams already running PostgreSQL who want vector search without a separate vector database, and developers comfortable with SQL extensions. The extension requires pgvector, is written in Rust, and is documented on its GitHub repository. Performance and cost comparisons with Pinecone come from the vendor's own benchmark on one dataset, so validate them against your workload.

What it can do

• Adds StreamingDiskANN approximate nearest neighbor index on pgvector embedding columns
• Creates indexes with a CREATE INDEX ... USING diskann statement on existing pgvector data
• Supports streaming filtering so metadata filters apply during vector similarity search without fixed-cutoff accuracy loss
• Applies Statistical Binary Quantization to reduce vector storage space while improving accuracy over standard binary quantization
• Stores part of the index on disk to reduce memory costs for large vector workloads
• Runs inside PostgreSQL alongside relational, time-series, and other data in the same database

Works with

• pgvector
• PostgreSQL

Before you choose

• Benchmark results (28x lower p95 latency, 75-79% lower cost than Pinecone) are the vendor's own tests on a 50 million Cohere embedding dataset and should be validated on your workload.
• Pgvectorscale is available as a self-hosted extension and on Timescale's cloud PostgreSQL platform; vector-optimized databases were offered through a separate private beta sign-up.
• Tiger Cloud billing is based on consumption (compute hourly, storage by GB-month), and the pricing page does not itemize pgvectorscale-specific charges.

Listed under

• Data
• AI applications
• AI infrastructure
• Database extension
• Developers
• Embeddings
• Open source

Get started

Installation instructions are on the pgvectorscale GitHub repository.

1. Install pgvector and pgvectorscale on a PostgreSQL instance using the instructions in the pgvectorscale GitHub repository.
2. Run CREATE EXTENSION vectorscale; in the target database.
3. Create a StreamingDiskANN index on your existing embedding column with CREATE INDEX ... USING diskann(embedding);.

Pgvectorscale itself is open source. Hosting costs depend on where you run PostgreSQL: self-hosted infrastructure or Tiger Cloud, whose plans bill compute and storage by consumption. Check the Tiger Cloud pricing page for current platform rates.

We couldn’t find published prices for pgvectorscale. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Pgvector Is Now Faster than Pinecone at 75% Less Cost | Tiger Data ↗www.tigerdata.com
• Tiger Cloud Pricing. Scale Further. Spend Less. | Tiger Data ↗www.tigerdata.com
• Tiger Data Documentation | Tiger Data Docs ↗www.tigerdata.com

Company website

tigerdata.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.