Core AI Models

Core AI Models is Apple's open-source companion project to the Core AI framework. It provides a curated catalog of popular models exported to run on Apple Silicon, along with the Python and Swift tooling needed to author, export, and run them on device. The repository's stated purpose is to supply model export recipes, reusable Python primitives, and Swift runtime utilities for building on-device AI with Core AI.

What It Is

Core AI Models is a GitHub repository published by Apple under a BSD 3-Clause license. The README describes four main parts: export recipes that convert popular open-source models from Hugging Face and other sources into Core AI's .aimodel format; reusable Python building blocks for authoring custom models in PyTorch; a Swift package with runtime utilities for integrating those models into macOS and iOS apps; and a set of skills that help coding agents work with Core AI. It is meant to be used alongside the Core AI framework rather than as a standalone runtime.

The Model Catalog

The repository centers on a model catalog with per-model README files and export recipes. Developers can list supported models through a model registry command-line tool included in the project. According to the documentation, exported models are produced as standalone .aimodel files, and models that need extra resources — such as a tokenizer for language models or a multi-stage pipeline for diffusion models — are exported as a resource folder containing the model files plus any required assets.

Authoring and Runtime Tooling

Beyond ready-made exports, the project ships Python primitives for authoring new models for on-device execution, and a Swift package, also named coreai-models, that provides runtime utilities for loading and running exported models inside an app. Command-line tools are included for running exported models directly on a Mac. The README notes the project builds on top of the Core AI framework and pairs with Apple's coreai-torch converter for the PyTorch-to-device workflow.

Skills for Coding Agents

A distinctive part of the project is a plugin of skills aimed at coding agents. The README lists skills for the end-to-end deployment workflow, for authoring PyTorch models that run well on Apple platforms, and for exploring weight-compression configurations. Apple documents installation steps for several agent environments, including Claude Code, Codex CLI, and Gemini CLI, and says the skills activate automatically based on the task or can be invoked explicitly.

Project Status and Scope

Apple states that it is not accepting code contributions at launch, describing this as a deliberate scope decision while it learns how the community uses the project; pull requests will be closed. The maintainers say they do welcome bug reports and model requests through GitHub Issues. Running and integrating the exported models requires macOS or iOS 27.0 or later and Xcode 27.0 or later.