pgvectorscale

Why we included this project

If you already store embeddings in PostgreSQL through pgvector and need similarity search to stay fast as your dataset grows, this extension is a natural upgrade path. It adds a StreamingDiskANN index that delivers much lower latency and higher throughput for approximate nearest-neighbor queries than the plain pgvector index, plus compression techniques that shrink storage for high-dimensional data. Label-based filtered search lets you combine vector similarity with metadata constraints in a single query, which matters for real-world retrieval where you rarely search on vectors alone. It installs as a standard Postgres extension and works alongside the existing pgvector tables you already have, so teams running production databases can adopt it incrementally rather than migrating to a separate vector service. The Rust-based implementation also gives Postgres users a more approachable codebase to inspect and extend than the C-based core.