Pryxor is an open-source runtime security gateway that sits between AI agents and external systems, evaluating and controlling each action before execution. It enforces explicit policies to approve safe actions, block forbidden ones, or pause risky actions for human review, addressing the gap between identity authorization and action-level security for autonomous agents.