Search built into the database
PlanetScale has released TIN, a full-text search extension designed to keep search workloads inside Postgres while supporting the database features applications already depend on. The company made the extension generally available for its Postgres and Neki databases on September 16.
TIN, short for Text INdex, is intended for applications that need more than a simple keyword lookup. PlanetScale says it supports conjunction, disjunction and phrase queries, BM25 ranking, exact result counts and continuous inserts, updates and deletes. Search results are designed to reflect committed changes while preserving transaction visibility, and queries can be combined with joins and filters on other column types.
Those capabilities target several common product patterns. An online store might rank the ten products matching every term, while a legal discovery system may need every document containing any requested term. Photo services can use exact counts for tags. Keeping these operations in Postgres can reduce the need to synchronize records into a separate search service.
Vendor benchmarks show large gains
PlanetScale tested TIN 1.0.2 against ParadeDB 0.25.2, pg_textsearch 1.4.0 and Postgres 18.6's built-in GIN index. Its principal benchmark used an 85GB Stack Exchange export containing 150 million documents and 1,719 synthetic queries. Each engine ran in an isolated container with eight virtual CPUs and 32GB of memory on an AWS i7i.8xlarge instance.
In the company's mixed-query, top-10 ranking test without concurrent writes, TIN processed 25 times as many queries per second as ParadeDB and recorded 26 times lower 99th-percentile latency. In a conjunction-and-phrase workload, PlanetScale reported throughput ten times higher than ParadeDB and 541 times higher than GIN. Under a disjunction workload paired with a client targeting 1,000 updates per second, it reported TIN handling 36 times the query rate of pg_textsearch and 57 times that of ParadeDB.
These are vendor-run results, not an independent evaluation. PlanetScale said it used a fork of the open-source ParadeDB Benchmarker, pre-warmed systems before measurement and disclosed changes to several Postgres settings. The benchmark phases ran sequentially, so the engines did not compete for resources. It also noted that only ParadeDB completed every tested workload besides TIN; some rivals could not run particular query types or exhausted the configured memory.
TIN's release gives Postgres users another route to application search without a separate engine. The practical value will depend on independent testing, production behavior and workload-specific requirements, but the general availability release puts the extension into customers' hands now.
