Apple Offline AI: Developers Gain Access to On-Device Models

Apple Introduces Foundation Models Framework for On-Device AI

Apple has announced the release of its Foundation Models framework. This new system empowers developers to utilize the company’s AI models directly on a user’s device, without requiring an internet connection.

On-Device AI Capabilities

During the WWDC 2025 presentation on Monday, Craig Federighi, Apple’s VP of software engineering, detailed how the Foundation Models framework will function. Apps will be able to leverage Apple-created AI models to enhance user experiences.

These models are integrated as part of Apple Intelligence, the suite of models that underpin various features within iOS.

Practical Applications

Federighi illustrated the framework’s potential with a practical example. “If you are preparing for a test, an application such as Kahoot can generate a customized quiz based on your study materials, making learning more interactive.”

A key benefit is that processing occurs locally on the device. This eliminates cloud API costs and ensures user privacy.

Developer Accessibility

According to an official Apple blog post, the Foundation Models framework offers native support for Swift, Apple’s primary programming language.

Developers can integrate Apple Intelligence models into their applications with minimal code – as little as three lines, Apple claims.

Framework Features and Early Adoption

The Foundation Models framework incorporates features like guided generation and tool calling.

Automattic is already implementing the framework within its Day One journaling app. Furthermore, the AllTrails mapping application is utilizing it to suggest personalized hiking trails.

Availability

The Foundation Models framework is currently available for testing through the Apple Developer Program. A public beta program is scheduled to launch early next month.

This allows developers to begin experimenting with and integrating Apple’s on-device AI capabilities into their applications.