Core AI Language Model Test

A multiplatform Apple app for experimenting with on-device AI language models. Built with SwiftUI and SwiftData, targeting iOS, macOS, and visionOS from a single codebase.

Platforms

Platform	Minimum Version
iOS / iPadOS	26.5
macOS	26.3
visionOS (xrOS)	supported

Device families: iPhone, iPad, Apple Vision Pro (TARGETED_DEVICE_FAMILY = 1,2,7).

Project Structure

Core AI Language Model Test/
├── Core AI Language Model Test/            # App target source
│   ├── Core_AI_Language_Model_TestApp.swift  # @main entry, SwiftData ModelContainer setup
│   ├── ContentView.swift                     # SwiftData item list with platform-adaptive navigation
│   ├── Item.swift                            # SwiftData @Model (timestamped item)
│   ├── LanguageModelDemo/                    # Foundation Models framework demos
│   │   ├── RootTabView.swift                 #   Tab navigation (Chat / Guided / Items)
│   │   ├── ChatDemoView.swift                #   Streaming chat with the on-device model
│   │   └── GuidedGenerationDemoView.swift    #   Structured output via @Generable
│   ├── Info.plist                            # Background modes (remote notifications)
│   ├── Core_AI_Language_Model_Test.entitlements  # CloudKit, push notifications
│   └── Assets.xcassets/                      # App icon, accent color
├── Core AI Language Model TestTests/       # Unit tests (Swift Testing framework)
├── Core AI Language Model TestUITests/     # UI tests (XCTest)
└── Core AI Language Model Test.xcodeproj/  # Xcode project

Language Model Demos

The app showcases Apple's on-device Foundation Models framework:

• Reference: Adding intelligent app features with generative models

• Chat (ChatDemoView) — streaming conversation with LanguageModelSession.streamResponse(to:), session prewarming, availability handling (Apple Intelligence eligibility, model download state), and session reset.

• Guided Generation (GuidedGenerationDemoView) — structured output using @Generable/@Guide: the model fills a typed TripPlan struct (title, destination, days, activities) streamed as PartiallyGenerated snapshots.

Both demos degrade gracefully when the model is unavailable. Running them requires a device/OS with Apple Intelligence enabled.

Architecture

• UI: SwiftUI. RootTabView hosts three tabs: the two language-model demos and the SwiftData item list. ContentView renders a List of items; a NavigationViewWrapper adapts navigation per platform (NavigationSplitView on macOS, plain content elsewhere). Platform-specific toolbar items via #if os(...) conditionals.
• AI: FoundationModels. SystemLanguageModel.default with availability checks; LanguageModelSession for streaming chat and guided generation of @Generable types.
• Persistence: SwiftData. Item is the single @Model; the app configures a shared ModelContainer at launch (persistent, not in-memory).
• Capabilities: CloudKit (iCloud services) and remote push notifications (aps-environment: development, remote-notification background mode) — groundwork for cloud sync.

Targets & Schemes

Target	Purpose
Core AI Language Model Test	Main app
Core AI Language Model TestTests	Unit tests — uses the Swift Testing framework (@Test, #expect)
Core AI Language Model TestUITests	UI tests — uses XCTest

Single shared scheme: Core AI Language Model Test.

Building

Requires Xcode with the appropriate platform SDKs.

# Build for macOS
xcodebuild -project "Core AI Language Model Test.xcodeproj" \
  -scheme "Core AI Language Model Test" \
  -destination "platform=macOS" build

# Build for iOS Simulator
xcodebuild -project "Core AI Language Model Test.xcodeproj" \
  -scheme "Core AI Language Model Test" \
  -destination "generic/platform=iOS Simulator" build

# Run tests
xcodebuild -project "Core AI Language Model Test.xcodeproj" \
  -scheme "Core AI Language Model Test" \
  -destination "platform=macOS" test

Or open the project in Xcode and press ⌘R.

Notes

• Bundle identifier: private.Core-AI-Language-Model-Test
• Swift language version: 5.0 (Xcode toolchain)
• The Items tab retains the SwiftData starter template — a timestamped item list with add/delete — alongside the Foundation Models demos.