wtf is pgcontext?

pgContext is a PostgreSQL extension, written in Rust, that turns the database into a full AI search engine. Instead of running a separate search service alongside your database and copying your data into it, pgContext keeps the search inside Postgres. Your existing tables remain the single source of truth for vectors, metadata, permissions, backups, and replication. The extension covers dense vector search using HNSW indexes, metadata-filtered approximate search, and hybrid retrieval that combines vector search with PostgreSQL's full-text ranking. It supports several distance metrics: L2, inner-product, cosine, and L1. Filters can be applied over registered columns and JSONB paths using a style similar to Qdrant, with must, should, and must_not conditions, plus range and equality checks. Results go through exact re-scoring against live rows before being returned, so approximate matches are verified under PostgreSQL's transaction visibility and row-level security rules. On the standard GloVe benchmark with 1.18 million vectors, pgContext matches pgvector's recall at every search setting while answering queries 3.8 to 5.3 times faster. At roughly 2.5 milliseconds per query, pgContext reaches 0.91 recall where pgvector reaches 0.75. The extension also ships with hybrid retrieval built in, fusing dense vector results with full-text ranking using reciprocal-rank fusion, rather than leaving that as application-level glue. The project is Apache-2.0 licensed and targets PostgreSQL 17 and 18. It is built by a company called Evokoa, which also offers a managed hosting version through a service called Polygres. Two live demos are available: a Wikipedia-scale hybrid search and an interactive memory demo with adjustable fusion weights. The README includes benchmark comparisons against pgvector and Qdrant, along with links to full reports and documentation. The full README is longer than what was shown.