Foundation Models Framework Example

Exploring Foundation Models

This project includes playground examples organized by chapters to help you learn everything about Apple's Foundation Models framework.

It is part of the Exploring Foundation Models book.

Requirements

• iOS 26.0+ or macOS 26.0+ (Xcode 26.0+)
• Xcode 26 official is required
• Apple Intelligence enabled
• Compatible Apple device with Apple Silicon

Try it on TestFlight

You can now try Foundation Lab on TestFlight! Join the beta: https://testflight.apple.com/join/JWR9FpP3

Automated TestFlight uploads

This repo includes a repo-local ASC workflow in .asc/workflow.json and a GitHub Actions workflow in .github/workflows/foundation-lab-testflight.yml that uploads the iOS app to the external TestFlight Beta group whenever app changes land on main.

GitHub Actions expects these repository secrets:

• ASC_KEY_ID
• ASC_ISSUER_ID
• ASC_PRIVATE_KEY_B64

It also expects these repository variables:

• FOUNDATION_LAB_APP_ID
• FOUNDATION_LAB_EXTERNAL_GROUP_ID
• FOUNDATION_LAB_BUNDLE_ID
• FOUNDATION_LAB_TEAM_ID

Getting Started

• Clone the repository
• Open FoundationLab.xcodeproj in Xcode
• Ensure you have a device with Apple Intelligence enabled
• Build and run the project
• Explore the different capabilities through the examples!

CLI

This repository also includes fm, a standalone command-line interface backed by shared capability code in FoundationLabCore.

• Local development: cd FoundationLabCLI && swift run fm --help
• One-off response: cd FoundationLabCLI && swift run fm session respond --dry-run --json --prompt "Suggest an uplifting science fiction novel"
• Xcode development: build the fm scheme from FoundationLab.xcodeproj
• Homebrew release prep: use ./Scripts/generate-homebrew-formula.sh <tag> and follow FoundationLabCLI/README.md

What's Inside

The app has four main sections:

Chat

Multi-turn conversations with context management, streaming responses, and a feedback system. Includes automatic context window management with session summarization when needed.

Tools

Nine system integration tools that extend the model's capabilities:

• Weather - Current weather for any location (OpenMeteo API)
• Web Search - Keyless Search1API (free, limited)
• Contacts - Search and access system contacts
• Calendar - Create and manage calendar events
• Reminders - AI-assisted reminder creation with priority levels
• Location - Current location and geocoding
• Health - HealthKit integration for health data queries
• Music - Apple Music search (requires subscription)
• Web Metadata - Extract metadata and generate social media summaries

Voice Interface

Talk to the model using speech:

• Voice-to-text with real-time transcription
• Text-to-speech responses
• Create reminders by voice
• Audio-reactive visualization
• Handles all permissions automatically

Health Dashboard

AI-powered health tracking with HealthKit:

• Personal health coach with contextual insights
• Trend analysis and correlations
• Predictive analytics
• Weekly summaries and personalized health plans
• Multiple health metrics tracking

Integrations Hub

Three sections for exploring advanced features:

• Tools - All nine system integration examples
• Schemas - Dynamic schema examples from basic to expert level
• Languages - Multilingual features and language detection

Examples

Ten different example types showing framework capabilities:

• One-shot prompts
• Journaling
• Creative writing
• Structured data generation
• Streaming responses
• Model availability checking
• Generation guides
• Generation options (temperature, tokens, fitness)
• Health dashboard
• RAG chat with document indexing and search

Features

Core Capabilities

• Chat: Multi-turn conversations with context management
• Streaming: Real-time response streaming
• Structured Generation: Type-safe data with @Generable
• Generation Guides: Constrained outputs with @Guide
• Tool Calling: System integrations for extended functionality
• RAG: Document indexing and semantic search with LumoKit/VecturaKit
• Voice: Speech-to-text and text-to-speech
• Health: HealthKit integration with AI insights
• Multilingual: Works in 10 languages (English, German, Spanish, French, Italian, Japanese, Korean, Portuguese, Chinese)

Dynamic Schemas

The app includes 11 dynamic schema examples ranging from basic to expert:

• Basic schemas
• Arrays and collections
• Enums and union types
• Nested objects
• Schema references
• Form builders
• Invoice processing
• Error handling patterns

Playground Examples

Four chapters with hands-on examples:

• Chapter 2: Getting Started with Sessions (16 examples)
• Chapter 3: Generation Options and Sampling Control (5 examples)
• Chapter 8: Basic Tool Use (9 examples)
• Chapter 13: Languages and Internationalization (7 examples)

Run these directly in Xcode using the #Playground directive.

Usage Examples

Basic Chat

let session = LanguageModelSession()
let response = try await session.respond(
    to: "Suggest a catchy name for a new coffee shop."
)
print(response.content)

Structured Data Generation

let session = LanguageModelSession()
let bookInfo = try await session.respond(
    to: "Suggest a sci-fi book.",
    generating: BookRecommendation.self
)
print("Title: \(bookInfo.content.title)")
print("Author: \(bookInfo.content.author)")

Tool Calling

// Single tool
let weatherSession = LanguageModelSession(tools: [WeatherTool()])
let response = try await weatherSession.respond(
    to: "Is it hotter in New Delhi or Cupertino?"
)

// Multiple tools
let multiSession = LanguageModelSession(tools: [
    WeatherTool(),
    Search1WebSearchTool(),
    ContactsTool()
])
let multiResponse = try await multiSession.respond(
    to: "Check the weather, search the web, and find my friend John's contact"
)

Streaming Responses

let session = LanguageModelSession()
let stream = session.streamResponse(to: "Write a short poem about technology.")

for try await partialText in stream {
    print("Partial: \(partialText)")
}

Voice Interface

// Speech recognition
let recognizer = SpeechRecognizer()
try recognizer.startRecognition()

// Text-to-speech
try await SpeechSynthesizer.shared.synthesizeAndSpeak(text: "Hello, how can I help you?")

Health Data

let session = LanguageModelSession(tools: [HealthDataTool()])
let response = try await session.respond(
    to: "Show me my step count trends this week"
)

Data Models

The app includes various @Generable data models for different use cases:

General Purpose

@Generable
struct BookRecommendation {
    @Guide(description: "The title of the book")
    let title: String

    @Guide(description: "The author's name")
    let author: String

    @Guide(description: "Genre of the book")
    let genre: Genre
}

@Generable
struct ProductReview {
    @Guide(description: "Product name")
    let productName: String

    @Guide(description: "Rating from 1 to 5")
    let rating: Int

    @Guide(description: "Key pros and cons")
    let pros: [String]
    let cons: [String]
}

@Generable
struct StoryOutline {
    let title: String
    let protagonist: String
    let conflict: String
    let setting: String
    let genre: StoryGenre
    let themes: [String]
}

@Generable
struct JournalEntrySummary {
    let prompt: String
    let upliftingMessage: String
    let sentenceStarters: [String]
    let summaryBullets: [String]
    let themes: [String]
}

Health Models

@Generable
struct HealthAI {
    let greeting: String
    let mood: HealthAIMood
    let motivationalMessage: String
    let focusMetrics: [String]
    let suggestions: [String]
}

@Generable
struct HealthAnalysis {
    let healthScore: Int
    let trends: HealthTrends
    let insights: [HealthInsightDetail]
    let correlations: [MetricCorrelation]
    let predictions: [HealthPrediction]
    let recommendations: [String]
}

@Generable
struct PersonalizedHealthPlan {
    let title: String
    let overview: String
    let currentStatus: String
    let weeklyActivities: [String]
    let nutritionGuidelines: NutritionPlan
    let sleepStrategy: String
    let milestones: [String]
}

Chat Context

@Generable
struct ConversationSummary {
    let summary: String
    let keyTopics: [String]
    let userPreferences: [String]
}

Tools Details

Weather Tool

• Uses OpenMeteo API for real-time weather
• Temperature, humidity, wind speed, conditions
• Automatic geocoding
• No API key required

Web Search Tool

• Uses Search1API keyless endpoint
• Returns search results with snippets
• No API key required (free tier limits)

Contacts Tool

• Search system contacts
• Natural language queries
• Requires contacts permission

Calendar Tool

• Create and manage events
• Timezone and locale aware
• Supports relative dates ("today", "tomorrow")
• Requires calendar permission

Reminders Tool

• Create reminders with AI
• Priority levels: None, Low, Medium, High
• Due dates and notes
• Requires reminders permission

Location Tool

• Current location information
• Geocoding support
• Requires location permission

Health Tool

• HealthKit integration
• Query health metrics
• AI-powered insights
• Requires HealthKit permission

Music Tool

• Apple Music search
• Songs, artists, albums
• Requires Apple Music subscription
• Requires music permission

Web Metadata Tool

• Extract webpage metadata
• Generate social media summaries
• Platform-specific formatting
• No API key required

Multilingual Support

The app works in 10 languages:

• English
• German
• Spanish
• French
• Italian
• Japanese
• Korean
• Portuguese (Brazil)
• Chinese (Simplified)
• Chinese (Traditional)

Language detection and code-switching examples are included in the Integrations section.

Permissions

The app may request the following permissions depending on which features you use:

• Microphone (for voice input)
• Speech Recognition
• Contacts
• Calendar
• Reminders
• Location
• HealthKit
• Apple Music

All permissions are requested at the appropriate time and can be managed in Settings.

Contributing

Contributions are welcome! Please feel free to submit a pull request.

License

This project is licensed under the MIT License - see the LICENSE file for details.