PlanetScale TIN builds text search around Postgres row locations

PlanetScale’s TIN text-search extension uses PostgreSQL’s physical row identifiers directly in its index, avoiding a separate mapping from search documents back to database rows. Compressed bitmaps let it combine matches with vector instructions. It supports ranked search, phrases and concurrent updates while checking which row versions are visible to each transaction.

PlanetScale · Sep 16, 2026

Why it matters

A useful design study for keeping search close to transactional data without adding a separate document-ID lookup.

More context

TIN is announced as generally available on PlanetScale Postgres and Neki databases. Try a representative corpus and concurrent writes before considering a migration. The engineering article explains how row locations also reduce work during index-segment merges.

Keep in mind

The performance comparisons are vendor benchmarks using synthetic queries and specified hardware. They are not 1min measurements or proof of superiority for every workload. Self-hosted availability is not established by this announcement.

Original read