VectorChord

Use it when

• Your pgvector index build and query latency have become unworkable now that the table holds tens or hundreds of millions of embeddings
• You want to keep vector search inside your existing Postgres instead of paying for a dedicated vector database as the corpus grows toward billion scale

What it solves

Not the fit when

• Non-PostgreSQL stacks
• Assuming the repository's 100M/1B scale, cost, build-time, or recall figures apply to different dimensions, datasets, hardware, tuning, or workloads
• Using the AGPLv3 option without meeting its source-sharing obligations, especially for network services
• Using the ELv2 option for prohibited managed-service scenarios
• Tiny datasets where plain pgvector already meets latency, memory, and operational needs
• generating or improving the embeddings themselves
• chunking and document ingestion pipelines
• keyword or BM25 full text search on its own
• managed hosting of the Postgres instance

Install

docker run --name vectorchord-demo -e POSTGRES_PASSWORD=mysecretpassword -p 5432:5432 -d ghcr.io/tensorchord/vchord-postgres:pg18-v1.1.1

Invoke

psql in, then CREATE EXTENSION IF NOT EXISTS vchord CASCADE; CREATE INDEX ON items USING vchordrq (embedding vector_l2_ops); SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;

Alternatives

No reviewed alternatives recorded yet.