Example: Build and ship an iOS app

This walkthrough takes a model from Edge Studio to a signed iOS app generated with Edge Scaffold.

What this example builds

You will create a private on-device chat app with:

• A model optimized in Edge Studio.
• A generated SwiftUI app from Edge Scaffold.
• One configuration file for the app name, model category, and prompt.
• A release build tested on a real iPhone or iPad.

1. Optimize the model

Open Edge Studio and load the source model.

For a first app, use Simple mode:

1. Choose the target device class.
2. Select the model category.
3. Run the recommended optimization.
4. Validate a short prompt on the target device profile.

Use the Pro pipeline when you need a custom benchmark matrix, multiple export targets, or manual comparison across model variants.

2. Export an Edge Scaffold app

In Edge Studio, choose Export and select Edge Scaffold app.

Export settings to check:

Setting	Recommendation
App name	Use the product name you want to see in Xcode.
Model category	Match the exported model: LLM, VLM, TTS, or STT.
Model delivery	Bundle for small models; remote or on-demand delivery for larger ones.
Minimum OS	Match the devices you validated.

Edge Studio writes a ZIP containing the app template, configuration, and model metadata needed by Edge Kit.

3. Open in Xcode

Unzip the export and open the generated project in Xcode.

Before running:

1. Select your development team.
2. Set a unique bundle identifier.
3. Choose a real device as the run destination.
4. Confirm the deployment target is iOS 17 or later.

Use a Release build for performance validation. Debug builds are useful while editing UI, but they do not represent final load time or throughput.

4. Configure ScaffoldConfig.swift

ScaffoldConfig.swift controls the generated app behavior.

import EdgeInference

enum ScaffoldConfig {
    static let appName = "Pocket Research"
    static let appDescription = "A private on-device research assistant."
    static let defaultSystemPrompt = """
    You are a concise research assistant. Answer with citations when context is provided.
    """

    static let modelCategory: ModelCategory = .llm
    static let modelID = "qwen3.5-9b-4bit"
    static let modelDisplayName = "Qwen3.5 9B 4bit"
    static let modelSizeGB: Double = 5.4

    // Use this when the model is included in the app bundle.
    static let bundleModelName: String? = "Qwen3.5-9B-4bit"

    // Used only by TTS apps.
    static let defaultTTSSpeaker: String? = nil
}

ScaffoldConfig reference

Field	Controls
appName	Display name used in the app UI.
appDescription	Short onboarding and settings description.
defaultSystemPrompt	Initial system instruction for chat-style apps.
modelCategory	Which UI and engine path the app uses.
modelID	Stable identifier for logs, cache keys, and settings.
modelDisplayName	Human-readable model name.
modelSizeGB	Approximate size shown in device checks.
bundleModelName	Bundle folder name when shipping the model inside the app.
defaultTTSSpeaker	Optional default speaker for TTS models.

5. Configure permissions and entitlements

For larger models, enable the Increased Memory Limit entitlement on the iOS target.

Add usage strings for the capabilities your agent exposes:

<key>NSMicrophoneUsageDescription</key>
<string>This app records your voice for private on-device transcription.</string>

<key>NSPhotoLibraryUsageDescription</key>
<string>This app lets you choose photos for private on-device analysis.</string>

<key>NSLocalNetworkUsageDescription</key>
<string>This app discovers your nearby devices for private on-device AI.</string>

Only include permissions that the shipped app actually uses.

6. Test on device

Run this checklist before archiving:

Area	What to verify
First launch	App opens, model setup appears, no missing files.
Model load	Cold load succeeds on the minimum supported device.
Generation	First reply streams and completes.
Long session	Multi-turn use stays responsive.
Backgrounding	App handles background and foreground transitions.
Storage	Model cache and user data can be cleared.
Permissions	Only required prompts appear.

For VLM, TTS, STT, or mesh apps, run the feature-specific path end to end on device before submission.

7. Archive and submit

In Xcode:

1. Select Any iOS Device or your distribution destination.
2. Choose Product > Archive.
3. Open Organizer.
4. Validate the archive.
5. Upload to App Store Connect.

In App Store Connect, make sure the privacy nutrition labels match the app's actual behavior. If the app keeps all prompts, audio, images, and model data on device, say that clearly in the review notes and in user-facing copy.

Deployment checklist

• App name, icon, bundle ID, and signing team are final.
• Increased Memory Limit entitlement is enabled when needed.
• Model delivery strategy is tested: bundled, on-demand, or remote download.
• First launch works without developer-only paths.
• Minimum supported device has been tested with a Release build.
• Settings include a way to clear local model and personalization data.
• App Store privacy answers match the shipped app behavior.

Next steps

• See Edge Scaffold configuration.
• Review Platform requirements.