Advertise with Us | Sign Up to the Newsletter

MobilePro #220: WWDC Week Special: Getting Started with Apple's Foundation Models

Jun 10, 2026 

Hi there,

We're entering an era where every app is expected to be intelligent.

Not long ago, adding AI to an app meant wiring up external APIs, managing tokens, and sending user data to the cloud. Today, Apple is pushing a very different vision. With Foundation Models, developers can tap directly into an on-device language model using a native Swift API, bringing AI features into apps without sacrificing privacy or responsiveness.

That shift feels especially significant after WWDC 2026. Between Siri AI, Apple Intelligence, and new developer APIs, Apple is making it clear that AI isn't just another feature category—it's becoming part of the platform itself. This week's article explores Foundation Models through a practical journaling app example and offers a glimpse into what building AI-powered iOS apps may soon look like.

TL;DR

Rust is becoming a strong choice for building fast, reliable AI applications.

We are conducting a live workshop with Francesco Ciulla, you’ll learn how to use Claude and Codex to scaffold, debug, refactor, test, and ship production-ready Rust apps faster.

You’ll build a practical Rust application live while exploring AI-native workflows for backend setup, AI integrations, retrieval/chat systems, and production-ready engineering patterns.

AI Agents Simplified cuts through AI noise with clear, actionable breakdowns of agents, automation, and what's actually worth your attention.

🌍 Trusted by 58,000+ subscribers worldwide
📬 New issues every week — no fluff, no hype
🧠 Built for curious minds, not just engineers

Want to feature your product to a highly engaged AI audience? Email to explore partnership options — Mention this newsletter and get a discount on your first sponsorship package.

Ever wished your iOS app could write for you? That’s exactly what Apple’s Foundation Models framework makes possible. Introduced at WWDC25 and now a centerpiece of the Apple Intelligence story, Foundation Models gives developers direct, on-device access to a powerful large language model — no cloud calls, no API keys, no latency spikes. The model lives right on the device, meaning it works even in airplane mode and keeps user data strictly private.

To see how it works in practice, we’ll walk through adding a Foundation Models-powered “Assist” screen to a journal app called JRNL. The screen lets a user type a short prompt — say, “a rainy afternoon hike in the hills” — and have the model draft a full journal entry. It’s a small feature, but it illustrates the full pattern you’ll use in any app that calls on this framework.

At its core, the Foundation Models framework exposes Apple’s on-device language model through a clean Swift API. You create a LanguageModelSession, optionally with a system-level instruction that shapes the model’s behavior, and then call session.respond(to:) with a user prompt. That’s it. The framework handles everything underneath: model loading, memory management, and inference.

Because the model runs locally, response times are fast and consistent. More importantly, user data — the prompts they type, the content they generate — never leaves the device. For categories like journaling, health, or finance, that privacy guarantee is not just a nice-to-have; it may be the reason a user chooses your app over a competitor’s.

One important caveat: not every device supports Apple Intelligence. Before calling into the framework, you need to check availability via SystemLanguageModel.default.availability. The availability enum covers three distinct failure states — device not eligible, Apple Intelligence disabled by the user, and model not yet downloaded — each of which calls for a different response in your UI.

The first step is importing the framework and creating a session. In the JRNL example, the session is declared as a stored property on the view controller so it persists for the lifetime of the screen:

import FoundationModels

private let session = LanguageModelSession(

    instructions: "Write a journal entry based on the user's prompt.")

The instructions: parameter is your system prompt — the framing that tells the model what role it is playing. Keep it concise and specific. A tightly-scoped instruction produces more predictable, on-brand output than a vague one.

Text generation is an asynchronous operation, so your generation method must be marked async. A clean pattern is to isolate generation in its own private method that returns a plain String, making error handling easy and keeping your action handler readable:

private func generateResult(prompt: String) async -> String {

  do {

    let result = try await session.respond(to: prompt)

    return result.content

  } catch {

    return "Could not generate content, please try again."

  }

}

In your button action, wrap the call in a Swift Task { } block and disable the button while the model is running so users can’t fire off duplicate requests. When the result arrives, assign it directly to your text view. The whole interaction — from button tap to displayed text — is typically well under a second on a capable device.

Journaling is just one of many natural fits for this framework. Text summarization is an obvious next step — feed in a long entry and ask the model to produce a one-sentence recap for a timeline view. Extraction is another compelling use case: structured data like locations, people, or mood keywords can be pulled from free-form text without a regex in sight. You can also use the model to provide contextual suggestions — a workout app that drafts a reflection on your run, a recipe app that generates variations based on what’s in your fridge.

The key to getting good results is prompt engineering. The instructions: parameter is powerful — use it to set tone, length, format, and persona. Pair that with a well-crafted user prompt and you will be surprised how capable the on-device model is for focused, domain-specific tasks.

Apple’s documentation links straight from the framework header, and the WWDC25 session (developer.apple.com, session 286) walks through the architecture in depth. Start there, then experiment. The best way to develop an intuition for what the model can do is to give it a problem from your own app and see what comes back.

This week’s WWDC 2026 makes this story even more exciting. Apple unveiled next-generation Apple Foundation Models built in collaboration with Google’s Gemini — with a cloud-tier model described as matching Gemini’s frontier quality — alongside a completely rebuilt Siri AI capable of contextual, cross-app reasoning. Third-party developers can now build on models like Claude and Gemini directly within Xcode via new coding agents. It’s hard not to feel a sense of wonder at the pace of it all: what started as a simple on-device text-generation API at WWDC25 has, in a single year, grown into an open, multi-model, multi-cloud platform powering the next generation of Apple software. The Foundation Models you build with today are just the beginning.

If you're looking to start your journey into iOS app development, iOS 26 Programming for Beginners by Ahmad Sahar offers a practical, hands-on introduction to building modern iPhone and iPad apps with Swift 6 and Xcode 26. Through a project-based approach, you'll learn UIKit, data persistence, media integration, maps, Apple Intelligence, and the new Liquid Glass design system while creating a fully functional journal app from scratch.

🧑‍💻 Learn to integrate Apple Intelligence and the sleek new Liquid Glass UI for modern app experiences

🛠️ Have fun building your first iOS app and start your iOS programming career

📱 Establish a solid foundation with UIKit, testing, and deployment best practices

We’ll be moving MobilePro to Substack soon. From that point forward, all issues will come from [email protected].

To ensure uninterrupted delivery, please whitelist this address in your mail client. No other action is required.

You’ll continue receiving the newsletter on the same weekly cadence, and on Substack you’ll also gain more granular control over your preferences if you wish to adjust them later.

When a new framework drops at WWDC, how quickly do you actually adopt it?
Reply and let me know.

Interested in sponsoring this newsletter and reaching a highly engaged audience of tech professionals? Simply reply to this email and our team will get in touch with next steps.

Cheers,

Runcil Rebello,

Editor-in-Chief, MobilePro

Copyright (C) 2025 Packt Publishing. All rights reserved. 

 Our mailing address is: 

Packt Publishing, Grosvenor House, 

11 St Paul's Square, Birmingham, 

West Midlands, B3 1RB, United Kingdom 

Want to change how you receive these emails? 

You can update your preferences or unsubscribe