foundation-models-on-device

On-device text generation in iOS is achieved by initializing an Apple FoundationModels session within your Swift app, which processes prompts entirely locally to ensure user privacy and offline capabilities.

Structured typed outputs from an on-device language model are generated using the @Generable property wrapper, which maps the model's output directly to predefined Swift types for reliable data extraction.

Custom tool calling is supported by Apple FoundationModels on-device, allowing your iOS app to invoke specific domain actions and functions locally during a single generative session.

Streaming LLM generation results progressively in an iOS UI is handled through snapshot streaming, which provides real-time updates to the interface as the on-device model generates text.

Apple FoundationModels supports offline apps needing private data extraction by running the language model entirely on the device, ensuring no cloud data is required and user privacy is preserved.

Checking if an iOS device supports on-device FoundationModels is done through built-in availability checks, which verify the device hardware before initializing a generative session.