foundation-models-on-device

Run on-device AI text generation in Swift by initializing a LanguageModelSession with the FoundationModels framework, which processes text locally on iOS 26+ to ensure data privacy and offline availability.

Yes, you can get structured output from an on-device LLM in Swift by applying the @Generable macro, which enforces type-safe data extraction directly from the model during generation.

Yes, the Apple FoundationModels framework supports custom tool calling, allowing you to extend model capabilities for domain-specific tasks like local database lookups without sending data to the cloud.

Update iOS UI in real-time during structured LLM generation by implementing snapshot streaming, which streams structured responses from the FoundationModels framework directly to your interface as they are created.

Build a privacy-preserving AI app that works offline on iPhone by integrating the FoundationModels framework, ensuring all AI text generation and data extraction occur entirely on-device without leaving the hardware.

Limitations of on-device AI for iOS apps include a strict dependency on iOS 26+ and Apple Intelligence hardware, meaning older devices cannot run the FoundationModels framework for local text generation.