If it fails because of us, you don't pay for it.

Paying for attempts

With AI, you pay for the work the model does, whether it goes well or not. If a task breaks because the service failed, the money is already spent and you have to ask again.

We think it's unfair for the person asking to pay for our mistakes. That's why Qwanter works the other way round.

How it works

• While it works: credits are deducted as they're used. No upfront estimates or holds.
• When it's done: the turn shows how many credits it actually cost.
• Out of credits: the task stops where it is and only what was used is charged. When you top up, Qwanter reminds you to continue it.
• If it fails because of us: what a task spent when it fails because of an error on our side or in the infrastructure goes back to your balance.

You don't pay for our mistakes.

Why we do it this way

Because it forces us to get it right. If a failure on our side earns us nothing, we want it to work the first time. That's why Build checks the result in a browser before delivering it, and Qwanter saves everything it can without lowering quality.

And because it lets you experiment without fear. Asking for something ambitious shouldn't cost you double because something broke along the way.

What it doesn't cover

A completed task isn't guaranteed to be exactly what you imagined. For that you have versions, the undo button and the conversation to adjust it.

All the details are in How credits work.