apple-foundation-models

Apple Foundation Models Skill

Early version: This skill is still in active development. It may lag behind the latest FoundationModels APIs, tools, or OS releases, and some material may be incomplete or incorrect. Treat it as guidance only. Verify behavior with Apple’s Foundation Models documentation and your target SDK before you ship.

Expert guidance for any AI coding assistant that supports the Agent Skills open format — unlocking the power of Apple's on-device generative AI via the iOS 26+ and macOS 26+ FoundationModels framework.

This repository aims to give your AI agent actionable, deeply technical references for building with SystemLanguageModel, with best-practice patterns for Swift 6 concurrency, hardware performance, structured data extraction, and tool calling.

Who this is for

• Developers integrating Apple Intelligence into their native apps.
• Teams migrating server-side LLM calls to secure, on-device execution.
• Engineers navigating the strict concurrency and context-window limitations of local foundation models.
• Anyone building native AI features using @Generable, real-time streaming, and custom Tool implementations.

How to Use This Skill

Option A: Using skills.sh

Install this skill with a single command via the Agent Skills CLI:

npx skills add https://github.com/alessiorubicini/Apple-Foundation-Models-Agent-Skill --skill apple-foundation-models-skill

Then prompt your AI agent (e.g., Cursor, Claude, or Copilot):

"Use the apple foundation models skill and help me write a custom Tool for fetching weather data to inject into my LanguageModelSession."

Option B: Claude Code Plugin

To install this Skill for your personal use in Claude Code:

1. Add the marketplace:

/plugin marketplace add alessiorubicini/Apple-Foundation-Models-Agent-Skill

2. Install the Skill:

/plugin install foundation-models@apple-foundation-models-skill

Project Configuration (For Teams)

To automatically provide this Skill to everyone working in a repository, configure your project's .claude/settings.json:

{
  "enabledPlugins": {
    "foundation-models@apple-foundation-models-skill": true
  },
  "extraKnownMarketplaces": {
    "apple-foundation-models-skill": {
      "source": {
        "source": "github",
        "repo": "alessiorubicini/Apple-Foundation-Models-Agent-Skill"
      }
    }
  }
}

Option C: Manual Install (Codex / OpenAI-compatible tools)

1. Clone this repository.
2. Copy or symlink the apple-foundation-models-skill/ folder into your tool's designated skills directory. For example, with Codex:

cp -R apple-foundation-models-skill/ "$CODEX_HOME/skills/apple-foundation-models-skill"

3. Use your AI tool as usual and explicitly ask it to refer to the "apple foundation models skill".

How to verify it's working: Your agent should start acknowledging hardware availability checks (SystemLanguageModel.default.availability), using @Generable for structured outputs, and wrapping LLM calls in proper Swift 6 @MainActor or actor isolations as defined in the SKILL.md workflow.

What's Inside

This skill covers the entire surface area of the FoundationModels framework. Because reference files load on demand, your agent gets highly targeted, context-efficient guidance exactly when it needs it.