Optional enabledWhether to run the background detector. Defaults to false.
Optional instructionsOverride the built-in extraction prompt.
Optional llmLLM used for extraction. An LLM instance, or a model string (e.g.
"google/gemini-3.5-flash") resolved via the inference gateway. Defaults to a
built-in detection model; the agent's own LLM is not used.
Optional maxCap on the confirmed (applied) detected keyterms if provided. Defaults to undefined.
Optional timeoutMilliseconds a single detection pass may run before it is dropped (no keyterm change).
Defaults to 10_000. Raise it if a slow detection llm needs longer.
Optional turnRun a pass once per N user turns. Defaults to 1.
Configuration for automatic keyterm detection.
Lives under the
keytermDetectionkey of KeytermsOptions. Absent or{ enabled: false }keeps detection off.