apple/coreai-models

Core AI Models is a project from Apple that helps developers run artificial intelligence models directly on Apple devices, such as Macs and iPhones, rather than sending data to a server. It is built around Apple's Core AI framework, and it brings together several kinds of tools in one place. There are four main parts. The first is a set of export recipes that convert popular open source models, including ones shared on the Hugging Face website, into Apple's own model format. The second is a collection of Python building blocks for people who want to write their own models in PyTorch, a common framework for machine learning. The third is a Swift package, meaning code for Apple's own programming language, that helps developers actually run these models inside an app. The fourth is a set of skills, which are plugins that help AI coding assistants work with Core AI more effectively. To use the project, you need recent Apple software: macOS or iOS version 27 or later, and Xcode 27 or later, which is Apple's tool for building apps. Models are saved as standalone files with an .aimodel extension. Some models need extra pieces to work, for example a tokenizer for language models, so the export step can produce a folder with everything bundled together. Command line tools are also included so you can run a model straight from a Mac. The repository explains how to list the supported models and how to install the coding-agent skills for tools like Claude Code, Codex, and Gemini. Each skill covers a specific workflow, such as exporting a model, authoring a model for on-device use, or exploring ways to compress a model so it takes up less space. One thing to note is that Apple is not accepting code contributions at this stage. Any pull request will be closed, which the team describes as a deliberate scope choice for this release. They do welcome bug reports and model requests through GitHub Issues. The project is released under the BSD 3-Clause License.