Reference

Under the hood

What runs your AI, where its software comes from, and how to check what it sends.

The engine

Your AI runs on Ollama, pinned to a tested version and checked against its published fingerprint before it installs. It listens only inside the box. OxOnOS refuses to use an engine anywhere else.

The default models

Kind of workModelSize on disk
Everydaygpt-oss 20B (MXFP4)13.8 GB
Deep thinkingQwen3.8 27B (Q4_K_M)17.7 GB
CodingQwen3-Coder 30B (Q4_K_M)18.6 GB
Everyday, no graphics cardPhi-4-mini 3.8B (Q4_K_M)2.5 GB

A new box starts with Everyday. Every model file is checked against its fingerprint before the box uses it. More are on the AI screen.

How updates arrive

  • From your own boxes. An OxOnOS update comes from another of your boxes, over your own network, when you press Install now.
  • Signed. Before running any of it, the box checks the release’s signature against its own list of trusted signers. Unsigned or altered, it is refused.
  • Tested, then kept. It installs beside the version you have and checks itself. If that fails, it goes back.
  • Security fixes. The box fetches security fixes for its core system once a day, by itself. The kernel and graphics driver stay pinned, and the box never restarts on its own for them.

Installing on a PC downloads the graphics driver and some software during setup.

Check “no telemetry” yourself

OxOnOS carries no tracking code and reports nothing to us. Three ways to see it:

  1. Reaching outIn Settings › Privacy, read live from the box: may OxOnOS’s own service reach the internet? It ships Sealed.
  2. The recordIn the same place, Every time something left this computer lists, with dates, where OxOnOS’s service connected to. Where, never what.
  3. Your routerWatch the box from outside. Beyond the housekeeping of the system underneath (security fixes, the clock), OxOnOS connects out only for a model you add, the OxOn Direct relay while Direct is on, and a service you connected while it is in use.