Define the permission
Policy + checkpointThe owner bounds wallet, merchant, model, chain, asset, expiry and budget. Discovery joins aiplans.dev data with Toolkit machine offers. Human checkout is ineligible.
MONAD METROPOLIS / DEVELOPER PREVIEW
AgentPay
Permission. Procurement. Proof.
An agent purchases its next inference inside an owner's spending policy, then recovers its result and verifies settlement on Monad.
The owner is part of the loop.
Recorded browser interaction. The coding task is a scripted, locally executable fixture. The model usage and Monad payment are separate historical evidence. No new model call or payment is made in this video.
One request across three trust boundaries.
Buy the inference
Reservation + x402 + Permit2A purchase binds the request digest and token bounds. The merchant records provider output and usage before settlement. Retries must preserve the purchase ID.
Verify the settlement
Transfer + receipt + recoveryValidate the expected payer, payee, asset and amount. Unknown settlement stays pending. The agent can reconcile or recover purchase-scoped access to its result.
Task checkpoint → discovery → bounded authorization → metered settlement → result recovery → resume
AgentPay adds the procurement policy, request binding and recovery lifecycle. x402 and Permit2 remain the payment standards. Monad supplies EVM-compatible testnet settlement; this demo claims no measured speed or cost advantage over other chains.
Historical evidence, exact amounts.
0.100000 USDCPer-request authorization ceiling
0.000001 USDCActual September 7 testnet settlement
273 tokens17 input + 256 output, GLM-4.7-Flash
The Toolkit testnet rate is not a production model price. The amount is not a wage, task budget or gas fee. The recorded model output did not generate the interactive repair.