No approval, no valid action. Every agent action produces a signed, independently verifiable receipt — before it executes.
How it works
Agent asks permission for a specific action
Human explicitly grants or denies it
Ed25519-signed receipt generated and bound to user, action, and nonce
Any system independently verifies — no trust required
Install the SDK
Why AT1C
AT1C never holds signing keys. Agent keys and user keys stay on their respective devices.
Every receipt is cryptographically signed. Any party can verify independently — no trust required.
Nonce-based single-use receipts. An approval for one action can never be reused for another.
MIT licensed. Inspect the protocol, fork it, wrap your existing system. No lock-in.
EU AI Act compliance
The EU AI Act requires accountability for automated decisions, human oversight of high-risk AI actions, and auditable records of AI behaviour. AT1C gives you all three out of the box. Enforcement begins August 2026.