AGENTPAY / MULTICHAIN WALKTHROUGH
Follow an agent buying one model task, losing its payment response, and recovering the result. Choose a network to see the same purchase lifecycle.
Illustrative USD budget; gas is excluded. A shared budget does not move funds between networks.
A research agent needs a model call to finish a vendor comparison. Its owner has set a $10 total budget and a $0.10 ceiling for each task.
Compare three vendors and produce a sourced buying recommendation.
The task, quotes and model output are scripted for this walkthrough.Sample JSON for explaining the lifecycle. Not a signed receipt, SDK response or transaction proof.
The design reserves against one total budget before execution. Adding a network must not create a second copy of the allowance. This demo visualizes that requirement; it does not enforce a production wallet policy.
The Monad scenario illustrates a capped, metered payment. Base and Solana illustrate a fixed quote. Their live authorization, settlement and recovery mechanisms still need integration and verification; Permit2 is not assumed on Solana.
When a response is lost, keep the purchase ID and its reservation. Check the existing payment before any retry. Switching chains could pay twice and is never an automatic recovery action.
The existing Monad implementation has a recorded testnet purchase. It is independent of the sample amounts and fictional task above. It is not evidence of a Base or Solana payment.
Historical Monad run: 0.100000 USDC authorized; 0.000001 USDC settled. September 7, 2026.
Watch the full story, then inspect the evidence boundary and the open source. This preview explains the multichain design; a competition submission still needs verified traces for each claimed live integration.