Apple’s Foundation Models explained: A new way to build with AI

With iOS 26, iPadOS 26, and macOS 26, Apple has quietly pushed a new tool into the hands of developers: the Foundation Models framework. It sits at the center of Apple Intelligence, giving apps access to an on-device large language model that runs privately, offline, and without additional cost.

Instead of flashy AI hype, the framework is being tested in more grounded ways. Health and fitness apps are using it to make training feel less mechanical. SmartGym, for instance, turns a user’s rough description of a workout into a structured routine, then explains why it suggests adjusting reps or weights. It also generates monthly summaries, personal notes, and even dynamic greetings tied to fitness progress.

In journaling, Stoic uses the model to suggest prompts that respond to a user’s state of mind. Log poor sleep, and the app responds with a compassionate nudge rather than a generic notification. The prompts are generated entirely on-device, keeping personal entries private. Similarly, Gratitude and Motivation are finding ways to transform journal entries into affirmations or mood-based reminders.

Education apps have also jumped in. CellWalk, which visualises cells in 3D, now layers conversational explanations of scientific terms on top of its graphics, tailoring answers to a learner’s knowledge level. Grammo builds on-the-fly grammar exercises, while Vocabulary automatically sorts saved words into themes like “verbs” or “anatomy.”

Creativity and productivity apps are experimenting in different directions. Stuff, a lightweight task manager, can parse natural language like “Do laundry tonight” or “Call Sophia Friday,” instantly filing them into the right lists. VLLO, a video editor, analyses scenes to recommend background music and stickers reducing the small frictions that slow down new creators. Apps like Signeasy, Agenda, and OmniFocus are using the framework to generate summaries, surface relevant notes, or propose next steps.

What makes this noteworthy isn’t just the features, but how quickly smaller teams are building them. The framework is tightly integrated with Swift, designed to slot into existing code, and supports guided generation so responses arrive in predictable formats. That lowers the barrier to experimenting with AI features, without requiring heavy infrastructure or sending sensitive data to the cloud.The Foundation Models framework won’t grab headlines the way new hardware does, but it may prove more significant in the long run. By making AI feel like a natural extension of apps rather than a separate destination, Apple is nudging the ecosystem toward intelligence that’s subtle, personal, and private the kind that works best when you barely notice it’s there.