Guardrail Call

The guardrail_call function validates input using a guardrail. This allows you to check content against various safety rules directly within a workflow.

• Function name: extensions.com.cloud-apim.llm-extension.guardrail_call

Parameters

Parameter	Type	Required	Description
kind	string	yes	The guardrail kind (see available guardrails below)
config	object	yes	The guardrail configuration
input	string/object/array	yes	The input to validate

Available guardrail kinds

• regex - Regular expression matching
• webhook - External webhook validation
• llm - LLM-based validation
• secrets_leakage - Detect secrets leakage
• auto_secrets_leakage - Automatic secrets leakage detection
• gibberish - Detect gibberish content
• pif - Detect personal information
• moderation - Language moderation
• moderation_model - Model-based moderation
• toxic_language - Detect toxic language
• racial_bias - Detect racial bias
• gender_bias - Detect gender bias
• personal_health_information - Detect personal health information
• prompt_injection - Detect prompt injection attempts
• faithfulness - Check response faithfulness
• sentences - Validate sentence count
• words - Validate word count
• characters - Validate character count
• contains - Check if text contains specific content
• semantic_contains - Semantic containment check
• quickjs - QuickJS scripted validation
• wasm - WASM-based validation

Output

{
  "pass": true
}

or

{
  "pass": false,
  "cause": "reason for denial"
}

Example

{
  "kind": "call",
  "function": "extensions.com.cloud-apim.llm-extension.guardrail_call",
  "args": {
    "kind": "regex",
    "config": {
      "deny": [".*credit card.*", ".*password.*"]
    },
    "input": "Hello, how are you?"
  },
  "result": "guardrail_result"
}