What Happens When Credits Run Out in an AI Builder?

Generation stops. Nothing else does. Your published project keeps serving at its address, your custom domain keeps pointing at it, your Supabase data and Stripe payments carry on, and the code is still yours to export. Credits pay for the builder to produce work, so running out pauses new builds and edits, not the thing you already made.

That distinction is the whole answer, and it is worth being precise about, because "hitting a limit" in a hosted tool usually means something goes dark. Here it means the meter that pays for generation reached zero. This post covers what that changes, what it does not, and how far an allowance actually goes.

What stops the moment you reach zero?

The actions that consume credits, which are the ones where the builder generates something. A fresh build costs 6, an edit to an existing build costs 3, and a multi-page platform costs 12 for the first three pages plus 3 for each page after that. Hi-Fi mode doubles the build and the edit.

Those are the levers, and they are published in advance rather than metered by tokens, which means you can plan a week of work on paper before you spend anything. What you cannot do at zero is ask for more generation until the allowance renews or you move up a plan.

Everything else is unaffected because it was never priced per action. Hosting a published project, pointing a domain at it, reading your own code, and the connected accounts you set up yourself all continue. The connectors are yours: a Supabase project lives in your Supabase account, and Stripe keeps processing payments through your Stripe account regardless of the state of a builder balance.

Does the published site go offline?

No. A project published to <slug>.zugo.run stays served, and a custom domain attached to it keeps resolving. Visitors have no way to tell what your credit balance is, and the site does not know either.

This matters most for the people who use a builder for client work or for a live offer. A landing page running an ad campaign does not have a metered heartbeat behind it. Running out of credits mid-month means you cannot change the page that week, not that the campaign stops.

If you want a stronger guarantee than any hosting promise, take the code. GitHub export puts the full source in a repository under your account, and after that the project's survival has nothing to do with a subscription at all. The mechanics are in exporting the code.

How far do 200 credits actually go?

Pro is $25 per month and includes 200 credits. The arithmetic below is the honest ceiling for each action taken alone, which is not how anyone actually works, but it frames the budget.

Real projects mix them. A typical month might be one platform, a handful of extra pages, and a long tail of edits as the copy settles. Business is $99 per month for teams that generate at a higher rate.

The rate of spend also depends on how you work. Ten separate one-line edits cost more than one edit that lists ten changes, so batching is the single largest saving available. More on that in editing after generation.

What about the Free plan's 5 credits?

Here is the uncomfortable arithmetic, stated rather than buried: Free grants 5 credits and a fresh build costs 6. A new account cannot complete a full build on the grant alone. That is a real constraint and pretending otherwise would waste your evening.

What the free tier is genuinely good for is looking around. You can see the interface, read the 25 templates including the 5 game ones, understand how the credit ledger works, and decide whether the shape of the tool suits you before any money changes hands.

If you want to evaluate the actual output quality, the honest advice is to budget one month of Pro rather than trying to squeeze a verdict out of 5 credits. A month at $25 with 200 credits is enough to build the real thing, break it, fix it, and know whether it belongs in your workflow. The wider cost picture is in AI app builder pricing.

How do you avoid burning credits on the wrong thing?

Five habits, in rough order of how much they save.

Write the whole brief before the first build. A build produced from a vague sentence costs the same 6 credits as one produced from a clear paragraph, and the vague one needs more edits afterwards.

Batch your changes. Collect a list of corrections, then spend one edit on all of them. Three edits at 3 credits each to fix three typos is the most common avoidable spend.

Start from a template when one fits. There are 25 of them, 5 for games, and starting closer to the destination means fewer corrective edits.

Use Hi-Fi deliberately. It doubles the price of the action, which is worth it for a page that has to look finished and wasteful for a rough draft nobody will see.

Decide the page count before building a platform. Twelve credits cover the first three pages and each one after costs 3, so knowing you need six pages up front is cheaper than discovering it in three separate rounds.

What should you check rather than assume?

Two things, because they are billing details that change and we would rather point you at the source than restate a number that drifts.

Whether unused credits carry into the next cycle, and exactly when a monthly allowance renews, both belong on the pricing page and in your account rather than in a blog post. Read the current terms there before you plan a large build around a renewal date.

The same applies to what happens to a failed attempt. If a build does not complete, check the balance in your account rather than trusting a general claim, and if the number looks wrong, that is a support conversation with a specific case attached. Why builds fail at all is covered in why did my build fail.

What are the honest limits here?

Credits are a real constraint, not a soft one. A month where the plan runs dry in week two is a month where the project waits, and no amount of planning removes that if the scope was underestimated.

The mitigation is the export path rather than a bigger number. Because the source can leave for a repository you own, a credit ceiling limits how fast you generate, not whether you keep what you generated. That is a meaningfully different kind of limit from one that holds your work hostage.

If you want to see where your own project lands on that table, the cheapest experiment is to build one real page and count what it took. Start at zugo.dev, note the balance before and after, and plan the month from a number you measured instead of one you guessed.