Agent Passport is a cryptographic identity and accountability layer for AI agents, built on Avalanche Fuji. Users sign in with email, spin up an autonomous agent in a three-field wizard, and the platform provisions it with its own on-chain passport, a dedicated EOA wallet, and a budget of AVAX that acts as operational fuel — when it's spent, the agent stops, giving the human owner a hard ceiling on autonomous action.

Every task the agent runs produces a signed session receipt: actions are batched into a single Merkle root and anchored on-chain in one transaction, so third parties can independently verify what an agent did without trusting our servers. The same identity primitive powers a trust protocol: agents sign outbound HTTP requests with three custom headers, letting any website verify in roughly thirty lines of middleware that the request comes from a registered, owner-accountable agent — a working alternative to CAPTCHA-based gatekeeping for the post-bot internet.

The system makes deliberately modest claims: we don't prove an AI is "real," we anchor it to a known human owner who can revoke it on-chain at any time, which is the same accountability model as a driver's license. Built with Next.js 14, viem, Foundry, Thirdweb, Vercel AI SDK, and Anthropic Claude, deployed on Vercel against live Avalanche Fuji contracts.