Getting started with Core AI
Core AI is a comprehensive set of technologies purpose-built for Apple silicon. Load and run AI models entirely on device — keeping your data private, your apps responsive, and your costs zero.
Overview
The Core AI framework provides a modern, memory-safe Swift API to load and run AI models entirely on device with zero server dependencies and zero token costs. Models are automatically specialized for the hardware they run on, with ahead-of-time compilation support for instant load times. Fine-grained control over inference memory, zero-copy data paths, and stateful execution give you the performance you need to run everything from compact vision models to large-scale generative AI across iPhone, iPad, Mac, and Apple Vision Pro.
Get started with models from the research community, converted and optimized for Core AI. Browse available models or download Swift packages to integrate directly into your projects. Also included are Generative AI Skills to help you get started and navigate specific workflows easily.
Core AI PyTorch extensions convert your PyTorch models into Core AI assets optimized for Apple Silicon. Export one or more inference functions into a single model artifact, take advantage of built-in hardware-optimized operations for common patterns like attention and normalization, or bring your own custom Metal 4 kernels for maximum performance. For advanced use cases, re-author your model with target-aware patterns and layouts to unlock the full power efficiency of the underlying device families.
Core Al Optimization reduces model size and improves inference performance through techniques such as quantization and palettization - with minimal accuracy loss. Highly customizable configurations let you control exactly which layers to compress, choose different techniques per layer, and tune granularity to find the optimal balance between model quality, size, and speed. Core AI Optimization gives you the flexibility to match your compression strategy to your deployment target.
New Xcode integration for Core AI lets you inspect Core AI graphs, profile model performance, and validate artifacts before deployment. The Core AI Debugger — a dedicated macOS application — provides deep visibility into model behavior and performance across the entire pipeline, tracing data directly back to your original Python source code.