𝗟𝗼𝗰𝗮𝗹 𝗖𝗼𝗱𝗶𝗻𝗴 𝗔𝗴𝗲𝗻𝘁 𝗼𝗻 𝗠𝗮𝗰𝗢𝗦

You can run a local coding agent on any Apple Silicon Mac. This setup keeps your code private and runs on your own hardware.

Here is how to set it up:

  1. Install Dependencies You need Python installed. Run these commands to install the MLX libraries:

pip install mlx mlx-lm mlx-vlm

Or use uv:

uv tool install mlx-lm && uv tool install mlx-vlm

  1. Pick a Model Use the mlx-community on Hugging Face to find models like Gemma, Llama, or Qwen. Model names follow a pattern: {name}-{size}-{instruction tuned}-{precision}.

If you struggle to choose, use this tool I built for MacOS: https://github.com/ruarfff/help-pick-local-model

  1. Start the Local Server Run the server using your chosen model. I use port 7777 to avoid conflicts.

mlx_lm.server --model mlx-community/gemma-4-26B-A4B-it-OptiQ-4bit --port 7777

For models that handle images and text, use mlx_vlm instead:

mlx_vlm.server --model mlx-community/gemma-4-12B-it-OptiQ-4bit --port 7777

  1. Connect Your Tools Point your coding agent to your local server.

For GitHub Copilot CLI: Create a shell function in your .zshrc or .bashrc to set the provider to OpenAI and the base URL to http://127.0.0.1:7777/v1.

For OpenCode: Update your ~/.config/opencode/opencode.json file to include your local MLX provider and model details.

For Pi: Update your ~/.pi/agent/models.json file with the local provider configuration.

Running models locally gives you control over your development environment.

Source: https://dev.to/ruarfff/local-coding-agent-on-macos-2gga

Optional learning community: https://t.me/GyaanSetuAi