Set up your first Agent
Complete the four setup steps shown in the AI Agents dashboard. Once setup is complete, you can optionally try a small payment to verify the full flow.
Sign in to the dashboard. To enable test payments, use a wallet you control on Arbitrum Sepolia. That wallet needs Paxos Test USDG for the payment and native testnet gas for blockchain transactions. Never paste its private key into the dashboard or an AI chat.
Create Agent
Open AI Agents, enter a name under Create AI Agent, and create it. This name helps you recognize which Agent requested a payment.
Done when: the Agent appears in the list with its own details page.
Connect Payment Wallet
Open that Agent, choose Connect Payment Wallet, and select an existing verified wallet or connect a new one. Sign the ownership challenge with that wallet and bind it to the Agent. One verified wallet can serve more than one Agent; each Agent still has separate rules.
Done when: the Agent shows the wallet's address and network.
Set Spending Policy
Choose a daily limit and a limit per payment, then confirm the onchain Policy with your wallet. The dashboard asks for a separate, clearly explained long-term wallet confirmation for the current Router. An earlier limited confirmation may need a one-time upgrade.
Done when: the Agent says Spending Policy ready.
Read the wallet request.Your onchain daily budget renews automatically at 00:00 UTC. The wallet's long-term Router permission does not renew daily; it remains until you revoke it. Your wallet may describe this as an unlimited USDG approval. Every Router payment still requires an exact signed authorization and must pass the onchain daily and per-payment Policy limits. Check the Router address in your wallet before approving. Understand the boundary →
Connect Agent
Select Connect Agent in the Agent details. Create or update its connection, then copy the Client ID and Client Secret into the Agent Runtime's environment file, secret store or settings panel. Keep them out of chat, screenshots, source code and Git. A read-only connection can inspect or preview; payment requests require payment authorization and remain subject to the onchain Policy. See how the Runtime uses these credentials →
Done when: the Agent has a payment-capable connection. The last-request time separately shows whether Spendkit has received a request from the Runtime; it does not prove the Agent is online now. Trying a payment is a separate optional check.
Start your Agent Runtime with its connection settings and the matching local Payment Wallet. In the reference local demo, the Agent previews a payment, requests authorization, shows the details, and asks you to confirm before sending it. Then open Payment Activity to review the final record; use Refresh payment status if confirmation is still pending. A test payment is not required to complete setup.