PostgreSQL ships a full-text search engine inside the database — no separate service, no syncing job, no eventual-consistency window. Instead of LIKE '%term%' (which is substring matching that can't use an index, ignores word boundaries, and has no notion of relevance), FTS understands words: it tokenizes text into normalized lexemes, strips stop words, applies stemming, and matches a parsed query against that. "running" matches a document containing "ran" or "runs" because all three stem to the same lexeme.
Two data types carry the whole feature. A tsvector is the preprocessed document — a sorted list of distinct lexemes with their positions. A tsquery is a parsed search expression with boolean operators. The @@ operator asks "does this vector satisfy this query?"
SELECT to_tsvector('english', 'The fox was running fast')
@@ to_tsquery('english', 'run & fox');
-- true: "running" stems to "run", "the"/"was" are stop wordsYou make this fast with a GIN index — either on the expression to_tsvector('english', body), or (better) on a generated tsvector column that you index directly. GIN stores an entry per lexeme pointing at the rows that contain it, so a search probes only the matching lexemes rather than scanning every row.
Results come back via ts_rank / ts_rank_cd, which score how well each document matches so you can ORDER BY relevance. The whole pipeline is driven by a text search configuration (english, simple, etc.) that bundles a parser and a chain of dictionaries deciding how each token is normalized.
The honest trade-off: this is excellent for search within your existing relational data — transactional, no extra infrastructure, joins to your other tables for free. It is not Elasticsearch: no distributed sharding across a cluster, far less relevance tuning, no fuzzy/typo tolerance out of the box, weaker analytics. For most apps that's a fine deal; for a search-first product at scale it isn't.
The questions below cover tsvector/tsquery, building the GIN index, the query-construction functions, ranking, language configurations, and when to graduate to a dedicated engine.