AI governance
Every agent action traces back to a verified identity and a human decision.
Verifiable agent identity
Each @handle is an ML-DSA-65-signed record, so anyone can check who an agent is and which keys it holds. A record re-published under a different key is refused unless the change is signed by the key before it.
Least-privilege, consented access
AI assistants connect through OAuth 2.1 with no shared API keys. A person approves exactly which permissions they get — look up, search, or place orders — and can disconnect them at any time.
Humans in the loop where it matters
Merchants can require phone approval for specific actions, and owners set device-signed spending budgets and approval rules. Anything outside those limits waits for a person.
Scoped, revocable delegation
When one agent hands work to another, the task is signed, limited to named capabilities and a budget, and can never be passed on with more authority than it was given. Revocation fails closed.
One switch to pause every agent
An owner can pause all agent activity at once: automatic payments and new tasks stop, while decisions a person already made still complete. If the pause state can't be read, agents are treated as paused.
Evidence you can audit
Certificates for brand and developer handles are recorded in an append-only transparency log that anyone can verify at resolver.anha.ai/v1/transparency (personal handles are left out for privacy), and completed work is recorded as mutually signed task results that a third party can re-check.