The way we build iOS apps is changing, and it’s changing fast. With the release of iOS 26, Apple has doubled down on on-device AI by introducing Foundational Models, CoreML, and CreateML as first-class citizens in the developer toolkit. For brands, that shift means new opportunities to deliver smarter, more responsive app experiences without relying on a heavy backend.
For years, AI in mobile apps was mostly handled on the server side. Apple’s move signals a future where personalization, recommendations, and even generative experiences can be powered natively, securely, and instantly on users’ devices. This is more than a performance boost — it’s a paradigm shift.
What’s in Apple’s Native AI Toolkit?
Apple’s AI capabilities now fall into three major categories:
• Foundational Models: Lightweight, OS-level APIs that can understand context and generate insights.
• CoreML: A framework for running machine learning models on-device with optimised performance.
• CreateML: A tool for training custom ML models using user data or mock data directly on a Mac.
Together, these tools empower developers to move beyond static app flows into dynamic, adaptive, and personalized experiences.
Exploring Apple’s On-Device AI Capabilities
At Heady, we’ve been actively experimenting with these frameworks to understand how they can bring value to our partners and their users.
A few early explorations have included:
• A Foundational Models proof of concept that recommended design elements based on a customer’s style preferences.
• A CoreML experiment predicting how much sleep a person needs, based on lifestyle inputs.
These were small steps, but they helped us validate Apple’s vision: that on-device intelligence is the future of mobile apps.
Real-World Use Case: On-Device Recommendations
Our team recently created a proof of concept using CoreML’s recommender engine.
The problem we focused on is one that nearly every ecommerce app faces: “When a user likes or engages with one product, how can we instantly recommend similar products they might love?”
Using CoreML, we trained a lightweight model to understand product relationships. The app now generates on-device recommendations whenever a user interacts with an item.
Here's why this matters:
• Speed: No network calls required; recommendations appear instantly.
• Privacy: All data stays on-device, aligned with Apple’s privacy-first approach.
• Business impact: Smarter recommendations mean higher engagement and better conversions, critical for retail and lifestyle apps.
Here's a look at the recommendation function in action:
Working on this POC surfaced a few important takeaways:
Pros
• CoreML makes personalization accessible. You don’t need a massive server pipeline to deliver value.
• CreateML + CoreML is a powerful combo: Training and deploying models is smoother than expected.
Cons
• The tech is still very young and evolving, which means it may change significantly as it matures.
• There's a learning and adapting curve. Teams will need time to fully leverage these tools.
What's Next for On-Device AI?
This is just the beginning. With Apple’s Foundational Models Framework, we’ll soon be able to integrate even richer generative AI features — everything from conversational interfaces to real-time content creation — without leaving the iOS ecosystem.
For brands, this opens the door to:
• Personalized product feeds
• Smarter in-app recommendations
• Seamless, privacy-first customer experiences
And for us as developers, it’s a chance to reimagine what mobile apps can be.