pgvectorscale

// Overview

pgvectorscale is a PostgreSQL extension that complements pgvector, the open-source vector data extension for Postgres. It adds a StreamingDiskANN index type, inspired by Microsoft's DiskANN algorithm, along with Statistical Binary Quantization for compact storage and label-based filtered vector search.

It is meant for application developers and DBAs who already store embeddings in PostgreSQL with pgvector and want approximate nearest neighbor search to stay fast and affordable as data grows into the tens of millions of vectors. You keep your data in Postgres instead of moving it to a separate vector database.

Unlike pgvector, which is written in C, pgvectorscale is built in Rust using the PGRX framework. It installs as the `vectorscale` extension and works alongside pgvector, so you create indexes and run queries with familiar SQL and pgvector distance operators. In October 2026 Tiger Data said pgvectorscale would enter maintenance mode, with most new vector-search work moving to its partition-based pg_vectorsearch extension.

// What it does

• StreamingDiskANN index, a disk-friendly graph index inspired by Microsoft's DiskANN research
• Statistical Binary Quantization for compact storage and faster search than standard binary quantization
• Label-based filtered vector search that combines similarity search with label filtering
• Works as a complement to pgvector, reusing its VECTOR type and distance operators (<=>, <->, <#>)
• Self-hosted on your own PostgreSQL, available via Docker, source build, or Timescale Cloud
• Written in Rust with PGRX, giving the Postgres community another way to contribute to vector support

// Getting started

Install the extension, enable it in your database, then create a StreamingDiskANN index on an embedding column and query it with pgvector operators.

psql -d "postgres://<username>:<password>@<host>:<port>/<database-name>"

CREATE EXTENSION IF NOT EXISTS vectorscale CASCADE;

CREATE TABLE IF NOT EXISTS document_embedding (
    id BIGINT PRIMARY KEY GENERATED BY DEFAULT AS IDENTITY,
    metadata JSONB,
    contents TEXT,
    embedding VECTOR(1536)
);

CREATE INDEX document_embedding_idx ON document_embedding
USING diskann (embedding vector_cosine_ops);

SELECT *
FROM document_embedding
ORDER BY embedding <=> $1
LIMIT 10;

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Scaling an existing pgvector setup to tens of millions of embeddings while keeping low query latency
• Keeping vector search inside PostgreSQL instead of running a separate vector database
• Combining vector similarity with label filtering for more precise retrieval in a RAG pipeline
• Reducing storage cost for large embedding tables using binary quantization

// How pgvectorscale compares

pgvectorscale alongside other open-source vector databases tools AI/TLDR tracks, ranked by GitHub stars.