Guardrails IA

Validates an AI input or output against defined security rules. Detects inappropriate content, prompt injections, and policy violations.

Parameters

Parameter	Type	Required	Variable	Description
input	dynamic value	Yes	Yes	Content to validate (typically an LLM output or user text).
checks	array	No	No	List of checks to perform (pii_detection, toxicity, prompt_injection, hallucination, custom).
onViolation	choice (block, warn, sanitize)	No	No	Action on detected violation: block the workflow, warn, or sanitize the content. (Default: "warn")
sanitizeRules	array	No	No	Custom sanitization rules (regex pattern and replacement text).
groundingSources	dynamic value	No	Yes	Reference sources for hallucination detection (documents, factual data).
continueOnViolation	boolean	No	No	Continue the workflow even when a violation is detected. (Default: false)
anonymizePII	boolean	No	No	Anonymize detected personal data with reversible identifiers. (Default: false)
outputVariable	text	No	No	Output variable name containing the validated content.
violationsVariable	text	No	No	Variable name storing the list of detected violations.

Parameters marked Variable = Yes accept the {{blockName.field}} syntax.

Output

Output variable : guardrailResult

{
  "passed": false,
  "violations": []
}

Example

Check that an AI response complies with rules.

Input :

{"text": "Voici les resultats de l analyse..."}

Output :

{"passed": true, "violations": []}

Tip

{{guardrailResult.passed}} is true if no violations are detected. {{guardrailResult.violations}} lists violated rules if any.