Kessa is an open-source system for verifiable AI agent delegation and attestation. Every consequential agent action carries a signed chain of custody back to the human who authorized it, built so that authority can only narrow at each step, never expand. The core differentiator: an independent verifier that checks the evidence trail offline, trusting nothing about the system that produced it, not even the enforcement point itself. Self-hostable, stdlib-only Go, open source.