Real-time runtime guard that inspects and can veto AI agent actions just before execution to prevent erroneous refunds, hallucinated communications, and other costly agent mistakes. Designed for teams deploying agents who need improved correctness; lightweight SDK integration for rapid onboarding.