Local models
Run open models on your own computer, privately and for free, and use them anywhere in Mach Studio.
Mach Studio runs open models directly on your computer. Nothing leaves your machine, there is nothing to pay for, and the agent keeps working without an internet connection.
Why run models on-device
- Private. Your prompts, files, and screenshots stay on this computer.
- Free. Local models cost nothing to download or run, on every plan.
- Offline. Chat, code, and run tools on a plane or behind a firewall.
- No rate limits. The only limit is how fast your hardware generates tokens.
What your hardware needs
Local models run fastest on Apple Silicon, where the model lives in the same memory the processor uses. Your computer's memory decides which models fit: a model needs roughly its download size in free memory to run well. Mach Studio measures this for you. In Browse, every model carries a Fits or May be tight badge, and the Fits this computer filter is on by default so you only see models you can actually run.
Smaller models answer faster and leave room for other apps. Larger models reason better but need more memory. Start with the model marked Recommended for this computer and move up from there.
Where to go next
- Find and download a model from the Syzygy catalog or Hugging Face.
- Manage your library: start, stop, tune, and free up disk.
- Use a local model in chat, including with Claude Code, Codex CLI, and OpenCode.
- Cloud models and your own API keys when you want a frontier model instead.