Step: Guardrail

• Key: guardrail
• Category: AI & Language Models
• Description: Validate content against guardrail policies for PII detection, prompt shielding, and moderation.

Inputs

• content (long-text, required): Content or template to evaluate. Flow variables are supported.
• policy (guardrail policy, optional): Policy configuration for PII detection, PromptShield, and moderation checks.

Outputs

• status (string): passed or blocked.
• passed (boolean): true when all configured checks pass.
• errors (array): Violations or execution errors, including category, severity, and action metadata when available.

Branching

If the step is configured with a failure branch internally, blocked results can include targetStepId so the Flow can continue on that branch.

Notes

• Place this step before sending user content to LLMs, tools, or external systems.
• Empty content is treated as blocked.
• Use the errors array for audit logs, explanations, or final responses.