KeywordClassifier

Buckets free text into one of a fixed set of categories by keyword hits — no model call, no network, no on-device inference. The "bucket this free text" case doesn't always need a full StructuredOutput round trip; when the categories are known in advance (intent routing, ticket triage), this is a shared, tested classifier instead of every caller hand-rolling its own .contains() chain that breaks on a synonym.

Matching is case-insensitive substring containment. Each category's score is how many of its keywords appear anywhere in the text; the highest-scoring category wins, ties broken by categories' declaration order (the same tiebreak a when chain would give). Returns null when no category's keywords appear at all.