BM25 full-text search for Postgres databases

AIThis description was written by a model from public signals. The facts behind it are listed under Evidence.

Problem

Postgres lacks built-in BM25 relevance-ranked full-text search capabilities. Teams building AI-centric applications need a scalable hybrid search solution combining keyword search with vector search, but existing options are either unavailable, proprietary, or locked behind restrictive licenses.

Solution concept

A performant, open-source Postgres extension delivering BM25 relevance-ranked full-text search with performance competitive with or exceeding mature systems, available under a permissive license and integrated into Postgres-compatible cloud platforms. Such a solution would let teams keep search capabilities within their primary database without lock-in or architectural complexity.

Who has it

Engineering teams at Postgres cloud vendors and companies running Postgres databases who need to support AI workloads and hybrid search (keyword + semantic search combined).

How they handle it today

Today, teams either build without BM25 search in Postgres, use closed-source or AGPL-licensed extensions like ParadeDB, migrate to separate search engines like Elasticsearch, or attempt to build their own extensions from scratch—a process estimated to take 6-12 months.

Why now

AI-centric workloads are driving demand for hybrid search (combining keyword relevance with semantic/vector search), and AI-assisted development tools (like Claude Code) have shortened development timelines for complex infrastructure projects from 6-12 months to 2 quarters, making it feasible to build competitive solutions quickly.

How it could charge

Inference: Postgres cloud vendors could monetize this as a differentiating feature in their managed Postgres offerings, or as an add-on service tier. Support, consulting, and optimization services around the extension could generate additional revenue.