foundation-models-on-device

On-device LLM execution in iOS uses Apple's FoundationModels to generate text entirely locally, ensuring data privacy by eliminating cloud dependencies. You instantiate a LanguageModelSession to process prompts securely offline.

Structured output is achieved using the Generable framework with FoundationModels, allowing your on-device LLM to return typed, structured data safely. This enables reliable data parsing for iOS assistants without network calls.

Tool calling with FoundationModels lets your on-device LLM trigger domain-specific actions within your iOS app. You must safely handle tool calls during the LanguageModelSession to prevent data leaks and execute functions locally.

Streaming LLM responses in SwiftUI uses snapshot streaming to provide real-time UI updates as the model generates text. This requires proper session management and availability checks to ensure the on-device model is ready.

Availability checks are required before instantiating a LanguageModelSession to verify the iOS device supports FoundationModels. This prevents runtime errors and ensures your privacy-first assistant operates safely offline.

Offline LLMs process sensitive data directly on the device using FoundationModels, guaranteeing zero data leaks compared to cloud APIs. This privacy-first approach is essential for secure iOS content generation and data handling.