foundation-models-on-device

To run an LLM on-device in Swift for offline privacy, you integrate Apple's FoundationModels into your iOS app, ensuring model availability checks and single-session handling to keep data entirely on the user's device.

Yes, you can get structured outputs from an on-device LLM in iOS by using the Generable feature, which provides type-safe results for your structured data generation workflows directly on the device.

You stream LLM responses in real-time for your iOS app UI by utilizing snapshot streaming with PartiallyGenerated types, which continuously update your interface as the on-device model generates text.

On-device LLM tool calling works without cloud access in Swift through FoundationModels, enabling custom tool calling workflows entirely offline while maintaining user privacy and requiring no network connectivity.

Limitations of using FoundationModels for on-device text generation include the need for model availability checks and strict single-session handling, meaning you must manage these constraints for stable offline LLM workflows.

Yes, you need to check model availability before starting an on-device LLM session because FoundationModels requires verifying that the device supports the model before you can initialize single-session text generation or tool calling.