CodeRunner provides a secure, sandboxed environment for executing AI-generated code locally on macOS, specifically designed for Apple Silicon Macs. It enables users to process local files with remote LLMs like Claude or ChatGPT without uploading sensitive data, by running the LLM-generated code within isolated Apple containers.

How It Works

CodeRunner leverages Apple's native container technology, offering VM-level isolation for code execution. This approach ensures that code runs in a secure, minimal environment with reduced resource utilization and attack surface. An MCP (Model Context Protocol) server facilitates communication between AI models and the sandboxed execution environment, allowing seamless integration with tools like Claude Desktop and OpenAI agents.

Quick Start & Requirements

• Install: Clone the repository, make install.sh executable, and run sudo ./install.sh.
• Prerequisites: macOS with Apple Silicon (M1/M2/M3/M4), Python 3.10+.
• Integration: Install requirements with pip install -r examples/requirements.txt.
• Docs: https://github.com/instavm/coderunner

Highlighted Details

• Secure execution via VM-level isolated Apple containers.
• Enables local file processing with remote LLMs, enhancing data privacy.
• Integrates with Claude Desktop, OpenAI agents, Gemini-CLI, and Kiro.
• MCP server for flexible AI model communication.

Maintenance & Community

The project is actively maintained by instavm. Contributions are welcomed.

Licensing & Compatibility

Licensed under the Apache 2.0 License. This license is permissive and generally compatible with commercial and closed-source applications.

Limitations & Caveats

Currently limited to macOS with Apple Silicon hardware. The integration examples require specific configuration steps for each supported AI tool.