</>MCP Agents Market
Skill

coreai-models

by apple2.1kSwiftUpdated 2026-09-04

Model export recipes, Python primitives, and Swift runtime utilities for on-device AI

Claude CodeCodexGemini

coreai-models is a collection of agent skills and utilities for deploying AI models on Apple devices through Core AI. It provides pluggable skills that teach coding agents how to export PyTorch models to Core AI format, author optimized models for Apple silicon, and explore compression configurations. The package includes model export recipes, Python primitives, Swift runtime utilities, and agent skills compatible with Claude Code, Codex CLI, and Gemini CLI.

Key Features

Three specialized agent skills: working-with-coreai for end-to-end deployment, model-authoring for on-device optimization, and model-compression-exploration for systematic quantization testing
Model export recipes to convert Hugging Face and open source models to Core AI .aimodel format
Python primitives for authoring custom Core AI models in PyTorch
Swift runtime utilities package for integrating Core AI models into macOS and iOS apps (requires iOS/macOS 27.0+)
Command-line tools for running exported models directly on Mac with Xcode 27.0+
Model registry with discovery via 'coreai.model.registry --list-models' command
Support for complex model pipelines including language models with tokenizers and multi-model diffusion sequences
Compatible with Claude Code, Codex CLI, and Gemini CLI agent platforms

Use Cases

  • 01Teaching coding agents to export and deploy PyTorch models on Apple silicon devices
  • 02Enabling AI assistants to guide developers through Core AI model authoring best practices
  • 03Automating exploration of weight compression and quantization strategies for on-device models
  • 04Converting Hugging Face models to Apple's Core AI format for iOS and macOS integration
  • 05Building custom on-device AI models optimized for Apple platforms with agent assistance
  • 06Integrating language models and diffusion pipelines into native Apple applications

Related Skills

View more

coreai-models — FAQ

What is coreai-models?+

coreai-models is a repository from Apple containing agent skills, model export recipes, and utilities for deploying AI models on Apple devices using Core AI. It includes pluggable skills that teach coding agents like Claude Code, Codex, and Gemini how to work with Core AI effectively.

How do I install the coreai-skills plugin in Claude Code?+

First register the marketplace with '/plugin marketplace add git@github.com:apple/coreai-models.git', then install the plugin with '/plugin install coreai-skills@coreai-models'. Alternatively, you can register from a local checkout using the file path.

Which AI coding agents work with coreai-models skills?+

The skills are compatible with Claude Code, Codex CLI, and Gemini CLI. Each platform has a different installation process detailed in the documentation.

What are the prerequisites for using coreai-models?+

For model export you need uv package manager installed via Homebrew or curl. For running models and app integration, you need macOS/iOS 27.0+ and Xcode 27.0+. No API keys are required.

Is coreai-models free to use?+

Yes, coreai-models is open source and licensed under the BSD 3-Clause License. It is freely available on GitHub.

What skills are included in the coreai-skills plugin?+

The plugin includes three skills: working-with-coreai for end-to-end PyTorch deployment, model-authoring covering optimization rules for Apple silicon, and model-compression-exploration for systematic quantization testing with coreai-opt.

How do I install coreai-models?+

Open the source repository on GitHub and follow its README. coreai-models is a skill — MCP Agents Market links you directly to the official repo.

Is coreai-models free?+

coreai-models is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches