Docs

Install the checkpoint. Passport the agent. Verify the action.

The integration model: free business checkpoints, $5/month passports, and one verifier decision your system can enforce.

For businesses

Add the free checkpoint to one risky action first.

Pick the protected action

The form, route, checkout step, or API write where unknown agents create risk.

Install the checkpoint

Place the verifier before that action. Enforce sensitive writes server-side.

Set policy

Decide what's allowed, denied, or routed to human approval.

Review decisions

Use the decision trail without turning logs into customer-data storage.

For agent owners

Create a passport for each agent that needs trust.

Create the passport

Attach it to the agent

Configure the agent to present its passport when a protected website asks.

Act within scope

Only request actions that match the passport scope; respect every decision.

Pause or revoke

If the agent changes roles or is compromised, revoke before the next action.

Verifier shape

One enforceable question: should this agent act right now?

POST /v1/verify

{
  "siteId": "site_example",
  "passportId": "ap_passport_123",
  "action": "checkout.submit",
  "origin": "https://store.example",
  "timestamp": "2026-06-10T20:00:00Z",
  "nonce": "unique-request-id",
  "signature": "agent-request-signature"
}

Response

{
  "decision": "needs_approval",
  "requestId": "ap_req_7QJ9",
  "passportStatus": "active",
  "matchedPolicy": "high-value-checkout",
  "approvalRequired": true,
  "expiresAt": "2026-06-10T20:12:00Z"
}

Decisions

Five outcomes. All enforceable.

allowedAgent, scope, policy, status, origin, and freshness checks pass.

needs_approvalThe action requires human review before it runs.

deniedThe request fails policy, scope, origin, signature, or risk checks.

expiredThe trusted window closed; a fresh decision is required.

revokedTrust was stopped; the protected system should refuse the action.

Keep the browser helpful and the server authoritative: enforce sensitive actions server-side, reject stale or reused requests, log hashed identifiers instead of raw data, and fail closed for risky actions. Production credentials and API access are issued directly by Gardien — don't wire sensitive workflows to placeholder examples.

Next step

Tell us the protected action and the agent. We'll map the first verifier flow.

Contact Gardien Review the flow