agentic.do shows which AI agents act on your site, whether they proved who they are, and whether they finish what they came to do — so you can serve them like a channel, not block them like fraud.
One script tag. No page content, form values or keystrokes ever leave the browser.
Who the agent claims to be, proven cryptographically — and how it actually behaves, measured in the browser.
Agents sign each request with Web Bot Auth. Valid signatures from trusted directories rise from Unknown to Signed, Registered and Delegated.
Timing, pointer paths, Fitts' law and keystroke rhythm score how agent-like a session is — and split hybrid sessions where a person hands off to an agent.
Every session, human or agent, gets a channel, source and campaign, including visits from ChatGPT, Perplexity and other AI assistants, with conversions per channel.
Screen readers and voice control can look agent-like. Only failed signatures are challenged.
IPs stored only as a daily-rotating salted hash. Raw behaviour kept 30 days.
Hold each agent to the rate it declared, instead of one blunt limit for all bots.
<head>
…
<script defer src="https://agentic.do/aa.js"
data-site="aa_pk_…"></script>
</head>
<!-- optional: track a goal -->
<script>
window.aa?.task("checkout", "complete");
</script>Register an agent, sign its requests, and get the acceptance data no single site can give you.
Ed25519 keys are generated in your browser. Only the public key is published, with rotation and instant revocation.
A standard /.well-known signatures directory, per-agent JWKS and a Signature Agent Card any site can trust.
Requests, acceptance rate, verification failures and task success — aggregated, never naming individual sites.