foundation-models-on-device

You can achieve on-device LLM text generation by initializing a FoundationModels session on a supported iOS 26+ device, which executes language models locally to keep user data completely private and offline.

Structured output with @Generable in FoundationModels applies guided generation constraints to language model responses, ensuring safer data handling by producing strictly typed, predictable structured data directly on the device.

Yes, the FoundationModels framework for on-device AI requires iOS 26+ devices. You must also perform model availability checks before creating a session to ensure the local environment supports on-device execution.

To implement tool calling and snapshot streaming for real-time UI updates in iOS, use the FoundationModels framework to execute local language models that stream generation snapshots and invoke external tools directly on-device.

Yes, Swift integration with the FoundationModels framework allows you to programmatically check model availability before creating an on-device LLM session, ensuring your iOS app safely initializes private, offline text generation.

Limitations of on-device foundation models include strict dependency on iOS 26+ environments and the requirement to perform model availability checks, as local hardware constraints may restrict model size and offline execution capabilities.