AI ON YOUR MAC // AGENTS · OPEN-WEIGHT MODELS
Set up your Mac for AI
Coding agents that work in your projects, and open-weight models that run privately on your own Mac. Pick a kit, paste one command, and InitMac shows which models your Mac can handle.
1 · Pick a kit
Want something else? Choose from all AI apps.
2 · Paste into Terminal
Fingerprint (SHA-256), checked automatically before it runs:
It shows the full plan and asks before changing anything. Add --dry-run to just look.
Couldn't create the command. Please try again.
Which models can my Mac run?
Open-weight models load into your Mac's memory, so memory decides what fits. The installer checks yours and marks your row.
qwen3:4b · gemma3:4bqwen3:8b · qwen3:14bqwen3:32b · gemma3:27bllama3.3:70b==> AI tools
Open-weight models on this Mac (48 GB memory):
8 GB+ small 1-4B models (e.g. qwen3:4b, gemma3:4b)
16 GB+ 7-14B models (e.g. qwen3:8b, qwen3:14b)
-> 32 GB+ up to ~32B models (e.g. qwen3:32b, gemma3:27b)
64 GB+ 70B-class models (e.g. llama3.3:70b) and everything smaller
A good first model for this Mac: qwen3:14b (about 9 GB to download).
Start Ollama in the background (it will also start at login) and download qwen3:14b now, about 9 GB? [y/N]
Try your coding agents: open a project folder in Terminal and run
claude # Claude Code
codex # Codex
What you get
- Coding agents that read your project, edit files and run commands, with you reviewing
- Local models that work offline: nothing you type leaves your Mac
- A model guide sized to your Mac, and one starter model if you want it
- The exact commands to try each tool first
What it promises
- Never asks for or stores API keys or accounts: you sign in to each tool yourself
- Downloads a model only if you answer
y(default is No) - Official Homebrew packages only, skipping anything you already have
- Open source and tested in a sandbox: read the code