At its WWDC 2025 event, Apple introduced the Foundation Models framework, a new developer tool that allows apps to access Apple Intelligence AI models offline, directly on-device. This marks a major step in bringing privacy-focused, AI-powered experiences to iOS users — without needing an internet connection or relying on cloud APIs.
What Is the Foundation Models Framework?
According to Craig Federighi, Apple’s VP of Software Engineering, the Foundation Models framework lets developers integrate Apple's own large language models (LLMs) into their apps with ease. These models are part of Apple Intelligence, the company’s suite of AI technologies that now power many smart features across iOS and other platforms.
One key benefit of this approach? No cloud costs and no compromise on user privacy. Everything runs on the user's device.
Real-World Use Cases
Apple shared several early examples of how this framework is being used:
- Kahoot: Can generate personalized quizzes from student notes — all offline.
- AllTrails: Suggests hiking routes tailored to a user's preferences using on-device AI.
- Day One (by Automattic): Integrates AI to enhance journaling experiences with guided content suggestions.
Developers can use the Swift programming language to access these models with just three lines of code. The framework also supports features like guided generation and tool calling, giving devs more control over how AI behaves inside their apps.
Privacy and Performance Combined
Apple emphasized that the Foundation Models framework is designed with privacy at its core. Because all AI processing happens locally, user data never leaves the device. This ensures a high level of security, low latency, and offline accessibility — ideal for apps in education, health, travel, and productivity.
Availability
The Foundation Models framework is available for testing now through the Apple Developer Program. A public beta will be released early next month, allowing even more developers to try out Apple's on-device AI power.
