Kastra decides what agents, models, and AI tools are allowed to do — before they do it. Sub-millisecond policy decisions, cryptographic audit trails, zero implicit trust.