Proof layer for AI agents. Every action signed, chained, verifiable.
This page drives a real demo stack on this machine: a verifier that checks signed agent credentials, and the “Smallville Permit Office” site serving three lanes — verified agents sail through, humans pass unchanged, everything else gets the challenge path. Press a button: every number below comes from a live call, never a mock.
Reset mints fresh demo keys and reboots the verifier and the site. The backend binds 127.0.0.1 only; nothing leaves this machine.
A legitimate agent presents a signed credential plus its delegation chain. The verifier checks signatures, chain integrity, scope, replay, and rate limits — then the request sails through, and the decision lands in the transparency log as a signed, hash-chained record.
A plain browser form post, no credential. It passes through unchanged — the verifier is never consulted.
A tampered credential — the scope was widened after issuance, so the issuer’s signature no longer matches. Denied with the reason shown, and sent to the challenge path instead of reaching the form.
Credential-less bots slam the agent endpoint. Each one gets the status-quo challenge instead of reaching the form.
Every allow and deny above becomes a signed, hash-chained record. Re-run the chain verification against the live log.