On March 31, 2026, TJ Green announced pg_textsearch v1.0 is now generally available on GitHub under the Postgres license. The project is a full BM25 search engine—including tokenization, indexing, compression, and query execution—built in C directly inside Postgres, without any sidecar or separate system. Indexes live in Postgres pages and use the buffer pool, making it a true native Postgres extension.
The project demonstrates the power of AI-assisted development in accelerating complex database systems work. TJ Green, who has 25 years of experience in database systems internals, told his CTO at Tiger Data (Timescale's parent company) that he could solo the project in one quarter using Claude Code and Opus. While it ultimately took two quarters with community help after open-sourcing the pre-release, this represents a massive acceleration compared to traditional estimates of 6-12 months with a small team.
Performance Benchmarks
Testing on the MS-MARCO dataset revealed impressive performance advantages over existing solutions:
- 4.7x faster query throughput than ParadeDB/Tantivy at scale
- Block-Max WAND optimization: delivers up to 4x faster top-k queries
- Parallel index builds: reduce indexing times by 4x or more for large tables
- Advanced compression: using delta encoding and bitpacking reduces index sizes by 41% while improving query performance by 10-20% for shorter queries
Technical Implementation
What sets pg_textsearch apart from alternatives like ParadeDB is its deep integration with Postgres:
- Indexes live in Postgres pages and use the buffer pool
- Support for 29+ languages
- Works with partitioned tables
- The entire search engine is built from scratch on top of Postgres's storage layer, not wrapping an external library like Tantivy
- Released under the Postgres license (not AGPL like ParadeDB)
Community Response and AI-Assisted Development
The Hacker News Show HN post received 91 points and 30 comments, with developers particularly interested in the AI-assisted development story. Green's success in building a production-grade search engine largely solo using Claude Code highlights how AI tools are enabling experienced developers to tackle projects that previously required entire teams.
The project is now available on Tiger Database Cloud and as open source on GitHub, offering developers a high-performance alternative to external search engines that maintains full integration with Postgres.
Key Takeaways
- pg_textsearch 1.0 is a complete BM25 search engine built natively inside Postgres using C, requiring no external systems
- The project was developed primarily solo in two quarters using AI-assisted development (Claude Code + Opus), compared to traditional 6-12 month estimates with a team
- Benchmarks show 4.7x faster query throughput than ParadeDB/Tantivy at scale, with 41% smaller index sizes
- The extension is released under the Postgres license and supports 29+ languages with full partitioned table support
- It's now available on Tiger Database Cloud and as open source on GitHub