Your AI and its models
Everything about the box’s AI is on one screen, called AI. No terminal, ever.
Three kinds of work
Questions, writing, summaries.
Harder problems, worth a slower answer.
Writing and fixing code.
A model for each kind, picked when you ask. A box with no graphics card answers on its processor, with a smaller model.
Add a model
- Open AIScroll to Add AI to this computer.
- Read the fitEvery model is marked for this box: Fits, Tight (it needs the processor to help, so it answers more slowly) or Won’t fit. A model that won’t fit has no Add button.
- Press AddThen Download. The model comes from Ollama’s public library. Nothing of yours is sent, only the request for the model.
- Let it finishThe box downloads it, checks its files, sets it up and asks it a test question. It is marked ready only once it has answered.
Only the owner can add a model. A box ships sealed from the internet, and then the list shows each model but nothing downloads: new models arrive with an update or on a seed stick, and the screen says so. Add appears once the box is open to the internet.
What you can add
Alongside the built-in models there is a catalogue of open models, including Qwen3, Gemma, gpt-oss, DeepSeek-R1, Qwen-Coder and Mistral Small, plus vision models that can read pictures. The fit marks tell you which ones suit your box.
Remove a model
Open the model and press Remove this AI, then Remove for everyone. You can set it up again later from the same screen. The box will not let you remove its only model, or one it is using for Everyday, Deep thinking or Coding.