Intent Guard

Intent Guards block the agent from executing actions that match specific forbidden intents. They intercept user requests and provide custom responses, making them ideal for security guardrails and policy enforcement.

Overview

Intent Guards are triggered when user input matches configured patterns. When triggered, they can:

• Block the action completely
• Provide a custom response explaining why the action is blocked
• Allow override (optional) for certain scenarios

Usage

await agent.policies.add_intent_guard(
    name="Block Deletions",
    description="Prevent data deletion",
    keywords=["delete", "remove", "drop"],
    response="I cannot perform deletion operations."
)

{
  "name": "Block Deletions",
  "type": "intent_guard",
  "triggers": [
      { "type": "keyword", "value": ["delete", "remove"], "operator": "or" }
  ],
  "response": {
      "content": "I cannot perform deletion operations."
  }
}

Intent Guard triggers automatically target "intent" (the user's current intent/input), ensuring guards are evaluated based on what the user is trying to accomplish.

Advanced Example with Natural Language Trigger

For more flexible intent matching, use natural language triggers:

await agent.policies.add_intent_guard(
    name="Block Financial Operations",
    description="Prevent unauthorized financial transactions",
    intent_examples=[
        "transfer money",
        "make a payment",
        "send funds",
        "wire transfer"
    ],
    response="I cannot perform financial transactions without proper authorization."
)

Response Configuration

Intent Guards support different response types:

• Natural Language: Plain text response (default)
• JSON: Structured JSON response
• Template: Template-based response with variables

For detailed trigger configuration, see Triggers.

Use Cases

• Security Guardrails: Block dangerous operations (deletions, modifications)
• Compliance: Enforce regulatory requirements
• Access Control: Restrict actions based on user roles
• Data Protection: Prevent unauthorized data access