Writing a good AI generation prompt

The generator writes a single self-contained SwiftUI file targeting iOS 15.8, with no external frameworks beyond what ships with SwiftUI/Foundation. Be specific about the UI and behavior — vague prompts get vague apps.

Works well: single-screen utility apps — calculators, converters, timers, trackers, simple games without real-time 3D rendering, anything using CoreMotion/CoreImage/basic system frameworks.

Doesn't work well: 3D/Metal rendering, multiplayer/networking-heavy apps, anything needing more than one screen's worth of complex state, apps requiring external packages.

Why images aren't supported

This isn't a missing feature that'll get added later — it's a fundamental limit of how generation works here. DeepSeek, the model behind this pipeline, is text-only: it can write Swift code describing UI elements, but it has no way to produce actual image or icon pixel data. There's no image-generation model anywhere in this pipeline.

In practice, that means an AI-generated app can only use built-in SF Symbols (Apple's system icon set) for anything icon-like — never a custom picture, logo, or illustration. If your app needs real custom imagery, that has to be added manually after generation, not asked for in the prompt.

Worth noting even if that changed: image generators like the ones behind ChatGPT/DALL-E-style tools are genuinely impressive for illustrations and concept art, but they're not reliable for the specific, pixel-precise assets a real app needs — a crisp app icon at exact sizes, a clean monochrome symbol that matches a design system, a transparent-background asset. That's a much narrower, harder bar than "generate a nice-looking image," so even bolting one on wouldn't get you production-ready app assets for free.

Example prompts

A tip calculator with a bill amount field, a tip
percentage slider, and a big total shown at the bottom,
dark themed.

A sensors app that shows live gyroscope data using
CoreMotion. Display X, Y, Z rotation rate as three
progress bars updating in realtime.

A simple to-do list — add items, tap to mark done,
swipe to delete. No backend, just local state.

What happens after you submit

• The AI writes ContentView.swift
• It's pushed to a private build repo, triggering a real GitHub Actions build
• If it fails to compile, the actual compiler error is fed back to the AI automatically and it retries — up to 3 attempts
• On success, the built .ipa is auto-published to the catalog, no manual step

Installing on your device

Open LuiSearchStore on a jailbroken device with TrollStore installed. Tap any app's "Get" button — it hands off to TrollStore's own install URL scheme, since a regular app can't install another app directly.

API for developers

GET  /api/appstore/catalog
GET  /api/appstore/ipa/<id>
POST /api/appstore/generate      { prompt, name }
GET  /api/appstore/generate/status?job=<id>&name=<name>
POST /api/appstore/publish       (multipart: name, subtitle,
                                   version, icon, publisher, ipa)

All served from luisearch.pages.dev. See the wiki for more detail on the internals.