apple-on-device-ai

This skill is a reference and decision guide for integrating on-device AI on Apple platforms. It addresses the problem of choosing among several overlapping frameworks — Apple Foundation Models, Core ML, MLX Swift, and llama.cpp — and deploying models efficiently on Apple Silicon. It provides a framework-selection router that maps use cases (text generation, structured output, image classification, custom model conversion, cross-platform LLM inference) to the appropriate technology, plus best practices and common-mistake checklists.

The content covers Apple Foundation Models (LanguageModelSession, @Generable and @Guide for guided generation, SystemLanguageModel availability checking, tool calling, locale handling, and always-on guardrails), Core ML (coremltools conversion from PyTorch, TensorFlow, scikit-learn, and XGBoost; quantization, palettization, and pruning; Neural Engine targeting; stateful and multifunction models; deployment targets and compute units), MLX Swift for high-throughput inference of open-source LLMs, and llama.cpp for GGUF-based cross-platform deployment. Supporting reference files detail Core ML conversion and optimization, Foundation Models, and MLX Swift. Code examples show availability checks, conversion pipelines, and input/output type definitions.

It targets iOS, macOS, and Apple Silicon developers building privacy-preserving, on-device AI features, as well as ML engineers converting and optimizing custom models. Use cases include adding tool-calling AI features on Apple Intelligence devices, generating structured output with @Generable types, converting a PyTorch or TensorFlow model to Core ML, and running open-source LLMs with maximum throughput on a Mac. It is a documentation-oriented skill; its examples are standard SDK and CLI usage (pip install, coremltools conversion) with no destructive or network-exfiltration behavior.