You pay for two things: compute time while the agent is active, and model tokens — nothing else.
Compute is charged while the agent is running. It costs $0 while the agent is idle or parked. Model tokens are billed at up to 30% off the maker's list price. Tools and connectors add small per-call charges. If you connect your own Claude or ChatGPT plan, model tokens from that plan are $0.
Your balance is credits you add in advance. Credits are spent as you use agents. The cost of each chat is shown in the chat header and broken down in the receipt.
Where: Settings › Team › Billing — your balance, usage by chat, and payment history. Web and Mac.
Good to know: a card on file enables Auto top-up. You can export usage as a CSV from Settings › Team › Billing. Related: Add credits to your account, Use your Claude or ChatGPT plan, Set up automatic top-up.