Let AI agents change real systems without holding the keys.
The agent proposes a structured change. A gateway checks it, applies it, verifies it and writes a receipt, with rollback to the recorded state. It is proven in software on a live game server and a robot simulator; production adapters are being built with design partners.
Agents are only useful when they can change things, and nobody wants to hand them credentials. Governed writes sits in between, so the agent never holds write credentials. It proposes a structured change pinned to the exact state it read. A gateway validates the change in a staging twin (feasibility-proven on disposable clusters) and admits or rejects it; it never silently repairs it. Then it applies the change, verifies the fields it owns and writes a machine-checkable receipt.
Byte-exact rollback and receipts are proven on the exercised paths. The full signed chain and an independent auditor verifier are in build, designed to support SOC 2 change-management evidence rather than establish compliance. Patent pending.
90-second demo in a controlled procedural test world: a pinned proposal admitted through five checks, a direct edit refused with state unchanged, and a verified reverse restore to the recorded pre-change state. Presentation pacing is disclosed on screen; no live AI authorship is depicted.