pgvector

pgvector is a PostgreSQL extension that adds native vector support and vector operations, including nearest neighbors search and cosine similarity calculation. In Supabase, pgvector comes installed by default, so one database handles both relational data (URLs, metadata, titles) and vector embeddings; no separate vector database needed.

In semantic audit pipelines, pgvector stores embeddings for page titles, content chunks, and keywords, enabling fast similarity search with SQL queries using the <=> operator (cosine distance). The <=> operator returns cosine distance (0 = identical, 2 = opposite), so ORDER BY ... <=> ... ASC gives nearest neighbors.

For example: SELECT title, embedding <=> query_embedding AS distance FROM pages ORDER BY distance LIMIT 10. This returns the 10 most similar pages.

In practice, with less than 100k vectors, pgvector in Supabase is sufficient — you need Qdrant only with millions of vectors or ultra-low latency requirements.