VectorChord is a PostgreSQL extension for scalable, high-performance, and disk-efficient vector similarity search, designed as a successor to pgvecto.rs. It targets developers and organizations needing to manage large vector datasets within Postgres, offering significant cost savings and performance improvements over existing solutions like pgvector.

How It Works

VectorChord leverages an IVF index with RaBitQ compression for efficient vector storage and retrieval. It supports external index building, potentially on GPUs, and includes autonomous reranking to maintain search quality. This approach allows for handling millions of vectors in modest memory footprints, with features like long vector support (up to 60,000 dimensions) and horizontal scaling capabilities.

Quick Start & Requirements

• Install: docker run -e POSTGRES_PASSWORD=mysecretpassword -p 5432:5432 -d ghcr.io/tensorchord/vchord-postgres:pg17-v0.3.0
• Prerequisites: PostgreSQL 17 (via Docker image), psql client.
• Setup: Minimal, especially with Docker.
• Docs: [Official Site][official-site-link], [Blog][blog-link]

Highlighted Details

• Up to 5x faster queries, 16x higher insert throughput, and 16x quicker index building compared to pgvector's HNSW.
• Stores 100M 768-dimensional vectors on a $250/month AWS instance (4 vCPUs, 32GB RAM).
• Supports vectors up to 60,000 dimensions.
• Production proven with 3B+ vectors.

Maintenance & Community

• Successor to pgvecto.rs, with a migration guide available.
• Active community support via Discord.
• Links to X (Twitter) and cloud integration.

Licensing & Compatibility

• Dual-licensed under AGPLv3 and Elastic License v2 (ELv2).
• AGPLv3 requires derivative works to be open-sourced. ELv2 has specific restrictions. Commercial collaboration is welcomed.

Limitations & Caveats

The AGPLv3 license may impose copyleft restrictions on commercial applications. The Elastic License v2 also contains specific usage limitations. Users should carefully review both licenses for compatibility with their deployment scenarios.