Agent Action Runtime

Your AI agent called the tool successfully. But did the action actually happen? Agent Action Runtime is an open-source execution layer for agents that take real-world actions. It adds policy, human approval, retries, post-action verification, recovery, and execution receipts between the agent and the side effect. Agent → Policy → Approval → Execute → Verify → Recover → Receipt Because API success ≠ action success.