For agent builders
You're shipping an AI agent that browses, books, buys, posts, summarizes. Half the sites your agent visits block it because every bot looks the same to bot-defense vendors. Whitelisting individually doesn't scale — there are millions of sites. Per-vendor partnerships are slow. Pretending to be a human breaks the moment the site asks.
AgentPKI gives your agent an Ed25519-signed passport. Sites that have adopted the protocol verify it in milliseconds and either accept your declared intent or reject it cleanly. Sites that haven't aren't worse off — your agent falls back to whatever they did before.
Pick any of these and it justifies the integration on its own.
Sites with an AgentPKI intent policy that accepts your declared intent let your agent through with zero negotiation. The site operator wrote the rules once; your agent follows them on every request.
Your bot's actions are signed and traceable to your issuer. This is a feature, not a cost — sites trust accountable agents more than anonymous ones, and your reputation accrues over time.
Tell sites what you're here to do (purchase, monitor, read-public). They can rate-limit appropriately, give you the right API path, and skip CAPTCHA challenges intended for the wrong intent class.
EU AI Act and US state AI labeling laws are starting to require identification of AI agents. Cryptographic passport infrastructure makes compliance machine-checkable instead of paper-based.
You don't need permission to mint passports — the protocol is open, and you can stand up a tier-1 issuer at your own domain in minutes. Tier-2 verification (KYB-attested) and tier-3 (hardware-attested) add credibility for sites that require it.
Generate an Ed25519 keypair. Publish a directory document at /.well-known/agentpki-issuer.json. Sign your agent's passports with your private key. Tier 1, ready in 5 minutes.
dashboard.agentpki.dev walks you through: pick a domain, paste two DNS records, get listed in the public issuer directory within 5 minutes.
Your agent's framework (Vercel AI SDK, Anthropic SDK, MCP) calls your issuer for a short-lived passport per session. Verifiers check it; sites with policies accept or deny based on declared intent.
Hands-on
The demo issuer at demo.agentpki.dev mints a passport for you with no auth. Run through the three story-paths — including the AI-agent integration walkthrough with Anthropic SDK and Claude — to see the protocol fire end-to-end.
