Route Foundation Models and on-device AI tasks with LanguageModelSession and @Generable.
npx skills add https://github.com/CharlesWiltgen/Axiom --skill ios-ai
Please help me install this Agent Skill. Skill: ios-ai Source: https://github.com/CharlesWiltgen/Axiom/tree/main/.claude-plugin/plugins/axiom/skills/ios-ai Command: npx skills add https://github.com/CharlesWiltgen/Axiom --skill ios-ai
This Skill consolidates guidance for implementing Apple Intelligence and Foundation Models on iOS, including on-device AI, LanguageModelSession, @Generable, and tool protocol patterns. It helps with debugging AI generation issues and iOS 26 on-device AI integration.
Use the ios-ai skill to implement an Apple Intelligence feature locally, for example: "Set up a LanguageModelSession to run a Foundation Model locally and enable @Generable structured output."
High-intent search queries and answers about installing and using this skill.
Foundation Models on iOS run locally via LanguageModelSession, which executes on-device AI without server calls. Set up a LanguageModelSession to initialize a model, then use @Generable for structured output and the Tool protocol to route specific tasks, enabling streaming generation and dynamic schema evolution.
@Generable is an annotation for declaring structured output schemas in Foundation Models on iOS. Apply @Generable to your data model, pass it to LanguageModelSession, and the model returns type-safe JSON matching your schema, eliminating manual parsing and validation.
Debug AI generation by routing Foundation Models work through decision-tree logic that categorizes failures into diagnostic buckets—schema mismatches, streaming interrupts, or device constraints. Use dedicated endpoints and API references to trace where generation fails, then adjust LanguageModelSession parameters or schema definitions.
Yes. The Tool protocol integrates with Foundation Models via LanguageModelSession, allowing you to define tool functions that the model calls during generation. Tools receive structured input from @Generable schemas and return results that the model uses to refine outputs.
Streaming generation outputs tokens incrementally as the model produces them, reducing perceived latency and enabling cancellation. Non-streaming waits for the complete response. LanguageModelSession supports both; choose streaming for interactive UI and non-streaming when you need the full result before proceeding.
iOS 26 on-device AI capabilities require iOS 26 or later. Earlier versions may support LanguageModelSession with server-backed models, but local Foundation Models execution and dynamic schema evolution are iOS 26 features.
mintuzIntegrate on-device AI models in iOS apps using Foundation Models and MLX Swift.
justcoonStream agent stdout, stderr, and logs in real time via WebSocket.
foxtrottwistDocument the on-device Foundation Models API for iOS, macOS, and iPadOS.
omarshahineGuide on-device AI model integration across Apple platforms with Core ML and MLX Swift.
