The agent's work cost more than you thought
#The build quality decides the price of each step. In Choose build quality, each level shows its model's cost, from "Lowest cost" to "Highest cost", and Ultra is marked "Priciest". A long chat also costs more per message.
- Use Economy for small edits, and switch to Standard or Ultra only for the hard parts. Choose build quality opens from the chat's toolbar.
- Start a new chat for a new topic. In the same menu, Context sets how much of the chat the agent keeps: "Larger context can cost more, and on some models context above ~200K tokens costs roughly 2x."
- Ask for one change at a time, and check it in the preview before the next one.
See Spend fewer credits.
One message cost a lot
#Open the message in Recent Activity and look at its Cost breakdown. When most of it is Cache write, the app explains: "Most of this is a one-time cache write of your prompt and tools. Follow-up messages reuse it at up to ~90% off." So for small follow-ups on the same topic, continue the chat instead of starting a new one.
The agent went in circles
#When the agent repeats itself, it stops: "I got stuck in a loop and stopped to save your credits." Choose the stop button (Stop generation) as soon as a run goes the wrong way. Then undo, describe the goal in other words, or switch to a higher build quality. See How building with the agent works.
A background task spent a lot
#The agent can start background tasks for heavy work. Each one may spend up to its task budget, 150 credits unless you change it, and then pauses: "This task reached its credit budget. All work so far is saved." Only you can let it continue. On Pro or Max, Settings, Billing & plans shows the budget; choose Change beside it to set a smaller one.
Images, video, music or voice
#Each one the agent generates costs credits. Ask for one version, check it, then ask for changes.
A Cloud app uses credits when nobody is building
#A Cloud app pays for the hours its computers are awake, for the disk it stores, and for traffic past the plan's included amount. A sleeping computer costs only its disk.
- Settings, Computer shows the app's prices. Always on keeps the production computer awake around the clock; turn it off unless the app needs it.
- A shorter Sleep timer means fewer awake minutes after each visit. A schedule that runs a command keeps the computer awake while the command runs.
- A busy site uses more traffic. Settings, Usage shows the Traffic row.
See Run a Cloud app and Bandwidth and limits.
Your app's AI features spent credits
#Every AI request your app sends through Playcode AI Gateway costs credits. A loop in the app's code, or a page that calls AI on every visit, can spend fast. Look at the project's row marked AI Gateway in Settings, Usage, then ask the agent to find where the app calls AI and to limit it.
Your app's email spent credits
#Pro and Max include a number of email recipients per billing cycle; past it, each recipient costs credits. Emails to your own team are free. The Email row in Settings, Usage shows how many were sent and how many were included.
Search research spent credits
#Keyword, competitor and backlink research charges for each fresh lookup. The same lookup within 24 hours is free.
You want a warning next time
#On Pro or Max, open Settings, Billing & plans, choose Spending controls, and set "Email me when I've used ... cr in a month". The app says: "A heads-up, never a stop - your AI keeps working and your computers never pause because of it." That holds for the agent and your computers. AI requests that your own app sends with its app key are refused once the month's spending passes that amount, until the next month or until you raise it.
When the balance reaches zero, the agent stops with "You're out of credits. Add more to keep building." A Cloud app's computers stop 72 hours later, after an email to the workspace owner. See Plans and credits on one page.
If you think a charge is wrong, get help from a person with the date and the project. The Terms of Service say AI credits are non-refundable.