foundation-models-on-device

You can generate structured AI outputs privately on-device by using the @Generable and @Guide macros to define typed models, then prompting the on-device LLM via LanguageModelSession to receive strongly typed results instead of parsing raw text.

Stream partial SwiftUI responses from an on-device LLM by calling the streamResponse method on a LanguageModelSession and handling PartiallyGenerated states to progressively update your user interface in real time.

Check Apple Intelligence availability before on-device text generation by performing model availability handling to detect device eligibility, verify Apple Intelligence enablement, and confirm model download and readiness states.

Yes, you can use custom tool calling with FoundationModels by defining app-specific tools with typed arguments, allowing the on-device language model to invoke domain-specific actions during a LanguageModelSession.

The best way to extract typed data without cloud networking is using @Generable constrained structured generation, which ensures the on-device LLM returns typed outputs and partial snapshots directly, eliminating fragile text parsing.

Maintain multi-turn conversations with an on-device language model by utilizing a LanguageModelSession, which preserves context across single-turn and multi-turn responses for continuous text generation and tool interactions.