SDK — Enforce Mode

Enforce mode blocks policy-violating tool calls before they execute. A PolicyViolationError is raised so your agent can handle it gracefully.

Basic usage

from agentic_guard import AgenticGuard, PolicyViolationError

gate = AgenticGuard(
    api_key="ag_xxxxxxxxxxxx",
    base_url="https://api.agentic-guard.com",
    mode="enforce",
)

try:
    result = gate.call(
        tool="github",
        action="delete_repo",
        tool_url="https://api.github.com/repos/myorg/myrepo",
        method="DELETE",
        reasoning="Cleaning up old repo",
        task_id="cleanup_001",
    )
except PolicyViolationError as e:
    print(f"Blocked: {e.message}")
    # Handle gracefully — log, notify, skip

PolicyViolationError

class PolicyViolationError(Exception):
    message: str       # human-readable reason from the policy rule
    tool: str          # tool that was blocked
    action: str        # action that was blocked
    rule_id: str       # which rule matched (if named)

Example: require reasoning or block

# Policy YAML
rules:
  - tool: "*"
    action: POST
    require_reasoning: true
    message: "POST calls require a reasoning header"

  - tool: stripe
    action: "*"
    require_reasoning: true
    message: "All Stripe calls must include reasoning"

# This raises PolicyViolationError
gate.call(tool="stripe", action="charge", tool_url="...", method="POST")

# This succeeds
gate.call(
    tool="stripe",
    action="charge",
    tool_url="...",
    method="POST",
    reasoning="User confirmed purchase of plan Pro",
)

Building resilient agents

def safe_tool_call(gate, **kwargs):
    try:
        return gate.call(**kwargs)
    except PolicyViolationError as e:
        # Log to your observability stack
        logger.warning("Policy blocked: %s", e.message)
        # Return a structured signal to the LLM
        return {"error": "policy_violation", "message": e.message}

Pass the error back to the LLM so it can adapt its behavior — request user approval, skip the step, or choose an alternative tool.

Differences from observe mode

	Observe	Enforce
Allows all calls	✓
Blocks on policy match		✓
Logs every call	✓	✓
Raises exception		✓
Good for	Monitoring	Production safety