apple-on-device-ai

To configure on-device AI on Apple Silicon, select Foundation Models, Core ML, MLX Swift, or llama.cpp based on your target hardware, required inference speed, and whether you need native tool-calling or cross-backend coordination. This skill provides a framework selection router to guide the decision.

Yes, you can build offline AI workflows with structured generation schemas on iOS by utilizing on-device inference frameworks like Foundation Models and Core ML. This enables private, low-latency AI execution without any network dependency.

To optimize models for the Apple Neural Engine, apply quantization, palettization, and pruning techniques. This skill provides specific deployment strategies and performance best practices to maximize inference efficiency across Apple hardware.

Yes, it supports building tool-calling AI features and cross-backend coordination for macOS apps. You can integrate on-device tool workflows and structured generation patterns across Foundation Models, MLX Swift, and llama.cpp backends.

When choosing between llama.cpp and MLX Swift for on-device inference, consider that MLX Swift is optimized specifically for Apple Silicon, while llama.cpp offers broader cross-platform compatibility. This skill outlines availability checks and performance trade-offs for each framework.