AIWebApp

AI app builder that keeps your idea on your device

Start with a focus timer: the planner sets 25-minute sessions and a dark theme from the description below. Change the sentence and generate again.

Focus TimerTimerExample — not saved until you change it

Runs on this device in under a second. Ctrl+Enter also generates.

What the planner understood

Confidence100%
  • timerTimer template
  • pomodoroTimer template
  • focusTimer template
  • darkdark colour mode
  • pomodoropomodoro timer, 25 min
Feasibility, templates and prompt history

Feasibility

  • NATIVELight and dark theme
  • NATIVEResponsive layout

Start from a template

Picking a template keeps the name, colours and fields from your description.

Prompt history

Off by default. Prompts are only kept in this browser if you turn this on.

index.html
Sandboxed · network off · fits 772px

App type

Timer — Pomodoro, countdown or stopwatch with laps.

Pages and sections

  • index.html
    1. App (Timer)

Drag sections to reorder, or use the arrow buttons. Navigation links are generated for every page.

Wireframe

Component library (8)

Native <dialog> with focus return, Escape to close and a labelled title.

<button class="btn btn-primary" type="button" data-open-dialog="demo-dialog">Open dialog</button>
<dialog class="c-modal" id="demo-dialog" aria-labelledby="demo-dialog-title">
  <h2 id="demo-dialog-title">Delete this project?</h2>
  <p>This removes it from this device. You can export a copy first.</p>
  <div class="c-modal-actions">
    <button class="btn" type="button" data-close-dialog>Cancel</button>
    <button class="btn btn-primary" type="button" data-close-dialog value="confirm">Delete</button>
  </div>
</dialog>

Two engines, one honest result

Most “AI app builders” send your prompt to a hosted model. This one starts with a rule-based planner that recognises app types, names, pages, fields, colours and durations, then composes the app from tested templates. That path is instant, works offline once loaded, and gives the same output for the same input.

The on-device model, when you enable it, improves the plan rather than writing unchecked code: it can rename things, write better copy, generate quiz questions or flashcards on your topic and propose data fields. Its JSON is validated against a schema; invalid output is repaired if possible and otherwise rolled back, and you see which happened.

Good first apps

Timers, to-do lists, notes, quizzes, flashcards, calculators, expense trackers, kanban boards, forms, admin tables and dashboards. Each one saves data with localStorage, supports light and dark themes, and is checked for labels, focus states and contrast.

Use the Checks tab to see the generated acceptance checklist — concrete things to try, like “Pause holds the time; Reset returns to the full duration” — and any runtime errors the preview reported.

Frequently asked questions

Is the AI required to build an app?

No. Every template and the planner work without it. The model is an optional refinement step.

Which models are available?

Small open models compiled for WebLLM, such as Qwen2.5 Coder 0.5B and 1.5B and Llama 3.2 1B. The AI tab shows each one’s download size, GPU memory need, licence and source before anything downloads.

Can the AI break my project?

Model output never applies silently. Plans are schema-validated; code edits are shown as a diff and only applied when you approve them.