AGENT-POLICY v0.1 · Registry Entry № 001

robots.txt for the agentic economy.

Sites publish which AI agents are allowed, what they may do, and on what commercial terms — as a machine-readable policy at /.well-known/agent-policy.json.

Declaration · Not Enforcement

Fig. 0.1 — The standard, in three steps

How it works

Generate your policy with the free questionnaire — bank-grade strictness to wide-open, your call.
Publish it at /.well-known/agent-policy.json (or host it with us on Pro).
Validate any policy against the AGENT-POLICY v0.1 spec — conformance report in seconds.

Generate your policy — free Validate a policy

Honest scope

AgentGate is declaration + discovery, not enforcement. A published policy states the site owner's terms in a standard form — it cannot force any agent to obey it. A "valid" verdict means schema conformance only, never a legal opinion.

Why now

October 2026: major sites began blocking AI agents outright (reported: Yelp, Delta, JPMorgan Chase, Charles Schwab, Citi, Bank of America vs. agents like Meta's Muse and xAI's Grok). Blocking is a blunt instrument — AgentGate gives sites a standard way to say yes, but on these terms instead of just no.

A policy looks like this

{
  "policy_version": "0.1",
  "site": "https://brightlinegoods.example",
  "owner": {"name": "Brightline Goods", "contact": "[email protected]"},
  "default_policy": "conditional",
  "agents": [
    {
      "identifier": "openai/agent",
      "allowed": "yes",
      "permitted_actions": ["browse", "read", "shop"],
      "disallowed_actions": ["checkout", "submit-form"],
      "commercial_terms": "paid-tier",
      "rate_limit": {"requests_per_minute": 60},
      "data_use_terms": "Product data may be used for price comparison only."
    }
  ],
  "last_updated": "2026-10-18",
  "signature": null
}

See all 3 sample policies →