← AgentProof

NO-SIGNUP QUICKSTART

Guard one dangerous function.

Do not rebuild your agent. Put one deterministic policy call immediately before the function that spends, pays, books, procures, changes credentials or makes another irreversible mutation.

1. Prove the decision in one command

curl -s https://agentproof.hatchable.site/api/pilot/demo \
  -H "content-type: application/json" \
  -d '{
    "policy_max": 500,
    "amount": 1200,
    "allowed_counterparty": "demo:supplier",
    "counterparty": "demo:supplier",
    "human_above": 5000
  }'

Expected result: DENY. This endpoint is a public test and never authorizes a production action.

2. Put the same pattern at your commit point

const decision = await agentproof.resolve(action);

if (!["ALLOW","ALLOW_WITH_LIMITS"].includes(decision.decision)) {
  throw new Error("AgentProof blocked commit");
}

// only now call the real payment / booking / procurement function
await existingRail.execute(action);

3. Choose how far to take it

Free test tenant

24 hours, isolated test environment, up to 1,000 resolutions, scoped key, no production permit.

Launch it →

One production check

Use the public x402 resolver at the exact crossing. The core admission check is $0.01 per successful call.

Make the 1¢ call →

One real production action

Use the 30-day founding pilot to integrate a single consequential workflow end to end before expanding scope.

Apply →

Need an integration hand?

OpenAPI · MCP · Compatibility map · Readiness · Conformance

Built on Hatchable