What are the requirements for using FoundationModels on-device?

Implementation requires iOS 26 or later and a device compatible with Apple Intelligence. Developers must also handle model availability checks to ensure the device is eligible and the model is fully downloaded.

Does this skill support streaming for better UX?

It supports 'snapshot streaming' for structured outputs, where the model provides partially generated versions of the final object so the UI can update incrementally.

How does this skill handle structured data?

It utilizes the @Generable macro to define Swift types that the model can populate directly, providing compile-time safety and eliminating the need for manual string parsing.

Can I use custom tools with the on-device model?

Yes, the skill provides patterns for defining a Tool protocol that allows the LLM to call your Swift code to perform specific tasks like database lookups or calculations.

Is there a limit to how much text the on-device model can process?

Yes, there is currently a 4,096 token limit that encompasses the system instructions, the user prompt, and the generated response.

Foundation Models On-Device

Name: Foundation Models On-Device
Author: affaan-m

byaffaan-m

•

130,863

•

Mobile Development

Integrates Apple's FoundationModels framework for privacy-preserving, on-device large language model capabilities in iOS applications.

This skill provides specialized guidance and implementation patterns for Apple's FoundationModels framework, enabling on-device AI for iOS 26+. It allows developers to build privacy-centric applications that leverage local LLMs for text generation, structured data extraction via the @Generable macro, and custom tool calling. The skill emphasizes best practices for handling hardware constraints, such as the 4,096 token context window, and provides robust patterns for device eligibility checking and real-time snapshot streaming to ensure responsive, offline-capable AI experiences.

主な機能

01Real-time UI integration using snapshot streaming

02On-device text generation and multi-turn conversation patterns

03130,863 GitHub stars

04Comprehensive model availability and device eligibility management

05Implementation of custom tool calling for domain-specific AI actions

06Structured output generation using the Swift @Generable macro

ユースケース

01Building privacy-first mobile apps where sensitive data never leaves the device

02Creating offline-capable AI assistants for travel, utility, or productivity apps

03Extracting structured data from natural language inputs for automated form filling

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add affaan-m/everything-claude-code foundation-models-on-device

For use in Claude.ai and ChatGPT

主な機能

01Real-time UI integration using snapshot streaming

02On-device text generation and multi-turn conversation patterns

03130,863 GitHub stars

04Comprehensive model availability and device eligibility management

05Implementation of custom tool calling for domain-specific AI actions

06Structured output generation using the Swift @Generable macro

ユースケース

01Building privacy-first mobile apps where sensitive data never leaves the device

02Creating offline-capable AI assistants for travel, utility, or productivity apps

03Extracting structured data from natural language inputs for automated form filling

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add affaan-m/everything-claude-code foundation-models-on-device

For use in Claude.ai and ChatGPT