OptionalallowLiteral values that never count as PII, e.g. a shared support address.
OptionalclassifierModel performing the screening. Defaults to DEFAULT_PII_CLASSIFIER_MODEL.
OptionalcustomApp-specific PII in plain language, e.g. 'internal case numbers like CASE-12345'.
OptionalentitiesEntity kinds to screen for. Defaults to DEFAULT_PII_ENTITY_TYPES.
reject refuses a prompt carrying PII; off (default) disables screening entirely.
Rejects a prompt before it reaches the agent's model when it carries PII, so the prompt is never sent to the model provider, never answered, and never written to chat memory.
Screening is one classifierModel call per prompt, covering entities and customRules together. Because a model does the reading rather than a pattern, a value is judged by what the prompt says it is, and app-specific identifiers can be described in plain language instead of expressed as a format.
This is the inverse of GuardrailsOptions.disablePii, which asks the agent's own model not to emit PII in its answer.