Create an app with AI
Sign in, start a new app in the Calculator, and select Prompt your app. Describe what you want to calculate. Choose a generation mode and check its displayed price before submitting the prompt.
Install the domain packs you want the new app to use from Marketplace before starting the prompt. The builder uses your installed versions and your own domain packs.
Cheap and fast creates and test-runs an app directly. It uses the current complete API reference, relevant model-authoring guidance and three full examples when your account can use their modules: the complete household, property rental forecast with table input, and Swedish pension forecast. Other full examples are not added to this prompt. For financial transactions it can compose existing accounts, debit/credit flows and domain components; a simple arithmetic calculator can stay simple. Website embedding and themes belong to the host website, separate from the calculation model.
When a generated model fails a test, the next attempt receives its previous code, the actual calculation error and a matching short repair example. A bounded retry does not charge another prompt. If generation reaches its output or context limit, the app explains that limit and keeps your prompt available; it does not present incomplete code as a finished app.
Slow but better works through the app's purpose, editable inputs, calculation, and presentation. When a choice needs your input, it shows up to three questions together. Select an answer to every question and choose Continue. Each question includes a recommendation and an Other option with a text field for your own answer. During input planning, use Other to change the proposed pages, tabs, or fields.
The builder asks at most three question batches per stage, twelve in total. Clear stages can proceed without questions. All answers and internal test runs are included in the original generation price. The session remains available for up to 24 hours. Cancel, leaving the page, or signing out ends it; starting again creates a new paid prompt. Cancelled and failed generations still use the displayed prompt price.
The calculation runs with its default inputs before the builder designs its presentation. If the proposed presentation cannot pass its checks within three attempts, the app uses the default presentation with the validated calculation.
When the app is ready, select Run app to save it in Your Apps and open it. You can then change its inputs, calculate again, and edit the model or presentation through the Calculator's editors.
Investigate fast app failures as a superuser
In Settings, a superuser can open Fast app generation failures and select Refresh. This separate history shows the five most recent fast operations that encountered a real error, including operations that later succeeded after repair. Ordinary successful generations, normal refusals and ordinary cancellation do not add a failure entry. The help-chat failure history remains separate.
Each operation shows its time and identifier, model and reasoning, knowledge/context version, final outcome and individual attempts. Expand the request/context or an attempt to inspect captured guidance, a safe error category and reason, provider finish status and generated source. Credential values are masked. A truncation label identifies text shortened for storage; a captured excerpt is not the complete original source. Only an authenticated superuser can read this history, and a diagnostics-storage problem does not change the generation's result or credit charge.