pgvector for PostgreSQL
A PostgreSQL extension that adds vector columns, similarity operators, and approximate search indexes.
Published by pgvector contributors
How to use it
Use a vector column for each passage embedding and retain the existing SQL columns that identify the document, section, and access scope. That makes it possible to constrain retrieval using your normal database rules.
Start with exact nearest-neighbor queries and compare them to HNSW or IVFFlat results at a fixed cutoff. Keep the index operator class consistent with the similarity metric your embedding model expects.
- Compatibility
- PostgreSQL, SQL
- Components
- Vector database extension, SQL filtering, Approximate nearest neighbors
- Access
- Local PostgreSQL extension. Requires a PostgreSQL installation that supports the extension.