What is the token limit for on-device models in iOS 26?

The framework currently supports a 4,096 token context window, which encompasses the system instructions, the user prompt, and the generated output combined.

What is the @Generable macro used for?

The @Generable macro is used to define Swift types that the on-device model can populate directly. This ensures structured, type-safe data extraction from natural language prompts without manual parsing.

Does this skill require an internet connection?

No, the FoundationModels framework runs entirely on-device, making it ideal for privacy-sensitive tasks and applications that must function in offline environments.

How does snapshot streaming differ from delta streaming?

Snapshot streaming provides a complete partial state of the object with every update. This makes it easier to manage structured UI updates in SwiftUI because each snapshot is a valid (though incomplete) version of the final type.

On-Device Apple Foundation Models

Name: On-Device Apple Foundation Models
Author: nbossn

bynbossn

0•

Mobile Development

Integrates Apple's on-device language models into iOS applications using the FoundationModels framework for privacy-first AI features.

This skill provides expert patterns and implementation guidance for utilizing Apple's native FoundationModels framework in iOS 26 and later. It enables developers to implement high-performance, private, and offline-capable AI features such as structured data extraction via the @Generable macro, custom tool calling for domain-specific tasks, and real-time snapshot streaming for responsive user interfaces. By following these patterns, developers can leverage Apple Intelligence while respecting hardware constraints and maintaining the highest standards of user data privacy.

Características Principales

01Structured output generation with @Generable and @Guide macros

02Privacy-preserving implementation patterns that work fully offline

030 GitHub stars

04Custom tool calling to bridge AI logic with application-specific code

05On-device text generation and summarization without cloud dependencies

06Real-time snapshot streaming for progressive UI updates

Casos de Uso

01Building privacy-sensitive AI assistants that process data exclusively on-device

02Implementing real-time AI features that function without an active internet connection

03Creating structured forms and data extractors from natural language input in mobile apps

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nbossn/claude-code-base foundation-models-on-device

For use in Claude.ai and ChatGPT

Características Principales

01Structured output generation with @Generable and @Guide macros

02Privacy-preserving implementation patterns that work fully offline

030 GitHub stars

04Custom tool calling to bridge AI logic with application-specific code

05On-device text generation and summarization without cloud dependencies

06Real-time snapshot streaming for progressive UI updates

Casos de Uso

01Building privacy-sensitive AI assistants that process data exclusively on-device

02Implementing real-time AI features that function without an active internet connection

03Creating structured forms and data extractors from natural language input in mobile apps

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add nbossn/claude-code-base foundation-models-on-device

For use in Claude.ai and ChatGPT