"This book has allowed me to be what I saw myself as being with Foundation Models. I feel I can legitimately use them expressively now. I needed this.
~Adrian Eves

This book teaches iOS, macOS, and visionOS developers how to integrate Apple's Foundation Models directly into their apps. Build features like streaming interfaces, structured data extraction, tool calling, and more.

All running on-device with Apple's privacy guarantees.

What you get:

169 pages PDF
EPUB for Kindle
LLM text for AI (~52,000 tokens)
Based on the most popular open-source example repository with 36+ playground examples

Foundation Models with Apple Intelligence

1. Introduction to Foundation Models

Apple's framework for accessing on-device Apple Intelligence models, including guided generation, tool calling, and stateful sessions for iOS 26.0+

The Model
Guided Generation
Streaming with Snapshots
Tool Calling
Stateful Sessions and Multi-Turn Conversations
Foundation Models vs MLX Swift
Developer Experience
Limitations and Considerations

2. Getting Started with Sessions

Setting up your first AI session and understanding model availability, error handling, and different generation configuration options

Project Setup
Checking Availability
Exploring with Playgrounds
Your First Session
Physiqa Example: Workout Assistant Session
Instructions vs Prompts
Prompt Engineering Best Practices
Understanding the Model's Capabilities
Basic Error Handling
Session Safety
Performance Optimization

3. Streaming and Snapshots

Building responsive UIs with partial results and understanding how Foundation Models streams complete object snapshots instead of token deltas

Why Snapshots Instead of Token Deltas?
Streaming Plain Text
Streaming Structured Results
Cancellation and Backpressure
Case Study: Pokémon Snapshot Streaming
When to Stream vs Generate Once

4. Generation Options and Sampling Control

Shaping model behavior with temperature, penalties, and other parameters to control output quality and creativity

Understanding Generation Options
Sampling Modes
Foundation Models vs MLX Swift Parameters
Parameter Tuning Tips

5. Structured Generation with Schemas

Creating type-safe, structured output directly from AI responses using the @Generable framework

Traditional AI Responses
Understanding the @Generable Macro
Understanding the @Guide Macro
Nested Structures
Optional Fields and Collections
Advanced @Guide Features
Journaling and Structured Analysis
Nutrition Data Parsing
Streaming with Structured Generation

6. Basic Tool Use

Enabling your AI to perform actions and access real-world data with practical examples of working with web search APIs

How Tools Work
Understanding the Tool Protocol
API Integration: Search Tool
Using Tools in Your App
Tool Building Best Practices
Health Data Integration with Zenther

7. Advanced Chat Patterns

Building production-ready conversation interfaces with context management, memory handling, and graceful error recovery

Working with Conversation Memory
Streaming Responses
Sliding Window Context Management
Learning from Users

8. Safety and Best Practices

Implementing responsible AI features with proper guardrails, user protection, and Apple's safety principles for trustworthy experiences

Apple's Safety Philosophy
Understanding Model Limitations
Built-in Safety Layers
Working with Guardrails in Depth
Safety in Instructions
Input Patterns with Risk Management
Using Guided Generation for Safety
Domain-Specific Safety Considerations
Building Trust

9. Integrating External JSON APIs

Reusing your Foundation Models schemas with external providers like OpenAI, Anthropic, and Google through JSON Schema compatibility

Architecture and Streaming
Defining the @Generable Structure
Including the JSON Schema
Making a Request to OpenAI with JSON Schema
Using AIProxy or Other Packages
Nutrition JSON to Foundation Models

11. Supported Languages and Internationalization

Working with Foundation Models across 14 supported locales, handling multilingual conversations, and session management strategies for global apps

Building Multilingual Experiences
Querying Available Languages
Creating Language Selection
Language Support Matrix
Generating Responses in Multiple Languages
Handling Multilingual Scenarios
Physiqa Example: Production Language Detection and Localization

12. Training Custom Adapters

Specializing Foundation Models for your app's domain and writing style using Apple's adapter training toolkit with cost-effective cloud GPU workflows

Understanding Foundation Models Adapters
When to Consider Custom Adapters
Training on Resource-Constrained Hardware
Scaling Up with the M5 MacBook Pro
Setting Up the Training Environment
Training Your First Adapter
Testing Your Adapter: Before and After
Using Adapter Studio CLI Wrapper for Easier Training
Exporting the Adapter
Entitlements and Device Support
Adapter Studio: Side-by-Side Evaluation on macOS
Loading Adapters in Your App
Compile the Draft Model (optional)
Locale and Language Support
Integration Checklist

Upcoming chapters:

13. Foundation Models Beyond the Main App

Using Foundation Models in app extensions, widgets, shortcuts, and system integrations while maintaining session-based patterns

14. Advanced Tool Patterns

Production integrations with external APIs and services, including retry strategies, fallback handling, and security considerations

15. Dynamic Generation Schemas

Runtime schema construction and complex data modeling for advanced structured generation patterns

16. Performance Optimization and Profiling

Using Foundation Models instruments in Xcode for production deployment, including memory management, token optimization, and performance monitoring

17. Voice-First AI Experiences

Building conversational interfaces and voice-driven interactions with Foundation Models

Use the code "STUDENT" for a discount if you are a student.

Exploring Foundation Models Framework