agent-policy

The AI Agent Access Control Standard

Define what AI agents are allowed to do on your site — before they do it.

---

The Problem

AI agents — crawlers, shopping bots, voice assistants, autonomous research tools — interact with websites in ways robots.txt was never designed to govern. They submit forms, extract PII, initiate purchases, and operate at machine speed. Site operators have no standard way to express intent, and agents have no standard way to ask for permission.

agent-policy.json and ai.txt fill that gap: a machine-readable contract between operators and AI systems, designed for the agentic web.

---

Two Complementary Formats

Format	Location	Purpose
agent-policy.json	/.well-known/agent-policy.json	Structured, machine-readable policy with granular per-capability controls
ai.txt	/ai.txt	Human-readable plain-text summary for quick inspection; mirrors robots.txt conventions

Use both. agent-policy.json is the authoritative source. ai.txt is the human-scannable summary.

---

Quick Start

Option 1 — Generate automatically

Visit agentpolicy.vercel.app, enter your URL and industry, and download a ready-made agent-policy.json + ai.txt pair in seconds.

Option 2 — Write manually

Create /.well-known/agent-policy.json on your server:

{
  "version": "1.0",
  "read_access": "allowed",
  "form_submission": "disallowed",
  "data_collection": "disallowed",
  "purchase_authority": "disallowed",
  "rate_limit": { "requests_per_minute": 30 }
}

---

Field Reference

Top-level fields

Field	Type	Values	Description
version	string	"1.0"	Spec version (required)
read_access	string	allowed | disallowed | conditional	Whether agents may read page content
form_submission	string	allowed | disallowed | conditional	Whether agents may submit forms
data_collection	string	allowed | disallowed | conditional	Whether agents may extract or store user data
purchase_authority	string	allowed | disallowed | conditional	Whether agents may initiate or complete purchases
rate_limit	object	—	Request frequency controls
contact	string	email or URL	Policy contact for questions
last_updated	string	ISO 8601 date	When this policy was last revised

rate_limit object

Field	Type	Description
requests_per_minute	integer	Max requests per minute per agent
requests_per_day	integer	Max requests per day per agent
crawl_delay_seconds	number	Minimum seconds between requests

conditional modifier

When a field is set to conditional, include an _conditions sibling key:

{
  "form_submission": "conditional",
  "form_submission_conditions": "Search and contact forms only. No checkout or account creation."
}

---

Examples

See the examples/ directory for industry-specific configurations:

• examples/ecommerce/agent-policy.json
• examples/saas/agent-policy.json
• examples/healthcare/agent-policy.json
• examples/media/agent-policy.json

---

Roadmap

• v1.0 core field set
• Generator tool at agentpolicy.vercel.app
• Industry presets (e-commerce, SaaS, healthcare, media, finance)
• agent-policy.json JSON Schema (for validation tooling)
• Browser extension for site scanning
• CLI validator: npx check-agent-policy https://example.com
• Per-path policy overrides (path_overrides)
• Agent identity assertions (agent_allowlist, agent_blocklist)
• Cryptographic signing support

---

Contributing

See CONTRIBUTING.md.

Short version: open an issue to discuss a change, then submit a PR against main. All spec changes require at least one real-world use case to justify them.

---

License

MIT — see LICENSE.