Can the model call local Swift functions?

Yes, by defining custom Tools within the LanguageModelSession, you can allow the model to trigger specific local functions and logic based on user input.

Does this skill require an internet connection?

No, the primary benefit of using this skill is implementing AI features that run entirely on-device, supporting offline usage and ensuring data privacy.

What is the context window for on-device models?

The current limit for Apple's on-device foundation models is 4,096 tokens, which covers the combined total of instructions, prompts, and output.

What is the FoundationModels framework?

It is an Apple framework introduced in iOS 26+ that allows developers to run large language models directly on the device for privacy and offline use.

How do I ensure the model returns structured data?

This skill utilizes the @Generable macro and @Guide constraints to define Swift types that the model fills, providing type-safe structured output instead of raw strings.

Apple On-Device Foundation Models

Name: Apple On-Device Foundation Models
Author: xu-xiang

byxu-xiang

•

323

•

Mobile Development

Integrates Apple's FoundationModels framework for privacy-focused, on-device LLM capabilities in iOS applications.

This skill provides a comprehensive pattern for utilizing Apple's native LLM framework on iOS 26+. it guides developers through implementing secure, offline AI features including text generation, structured data extraction with Swift macros, and custom tool calling. By leveraging the FoundationModels framework, developers can build responsive, privacy-preserving applications that handle natural language processing entirely on the user's device without relying on external cloud services.

주요 기능

01On-device text generation and summarization

02Real-time snapshot streaming for responsive UIs

03323 GitHub stars

04Structured output using @Generable Swift macros

05Automatic model availability and eligibility checks

06Custom tool calling for domain-specific AI tasks

사용 사례

01Building privacy-sensitive AI applications that process data locally

02Extracting structured data from natural language input for forms or databases

03Implementing offline AI assistants and text processing tools

주요 기능

01On-device text generation and summarization

02Real-time snapshot streaming for responsive UIs

03323 GitHub stars

04Structured output using @Generable Swift macros

05Automatic model availability and eligibility checks

06Custom tool calling for domain-specific AI tasks

사용 사례

01Building privacy-sensitive AI applications that process data locally

02Extracting structured data from natural language input for forms or databases

03Implementing offline AI assistants and text processing tools