Core AI Models is Apple’s official open-source repository of model export recipes, Python primitives and Swift runtime utilities for on-device AI. It gives developers the building blocks to take popular open models, convert them with proven recipes, and run them efficiently on Apple silicon. Published by Apple on GitHub, it has 2,201 stars and is under active development.
Shipping AI features on Apple devices means getting models into the right format and runtime shape for the Neural Engine and GPU. Core AI Models collects Apple’s own recipes for doing exactly that: export paths for well-known open models, Python-side primitives used during conversion, and Swift utilities for loading and running the results inside apps. For developers targeting macOS and iOS, it is the closest thing to a first-party cookbook for on-device deployment.
The repository is organised around practical deployment questions: which format to export to, how to handle tokenisation and pre-processing, and how to structure the Swift side so inference stays responsive inside a real app. As Apple’s on-device AI stack evolves, the recipes are updated to match, making the repo a living reference rather than a one-off sample dump.
Core AI Models is completely free and open source under the BSD-3-Clause license. There are no fees; you build and ship with your normal Apple developer tooling.
Core AI Models is for iOS and macOS developers who want to run AI models locally in their apps instead of depending on cloud APIs. If you are converting open models for Apple hardware, these recipes can save you significant trial and error.