Cin-MinAI 0.0.1 β€” Linux Mint with an AI built in

An operating system with an AI assistant built in. It runs on your own computer, on your own graphics card, and works without the internet. It helps you use your computer, build things, learn and write, and it asks before it changes anything.

Free. No ads, no telemetry, no account. Download it and use it wherever, however you need it.

Website: cin-minai.github.io Β· Source code: github.com/Cin-MinAI

Download

cinminai-0.0.1-amd64.iso β€” 5.9 GB (5,899,288,576 bytes)

SHA-256: ae74be30c25a3e14d3a9ca4620e570de2b4318f50f5912d659de01c2553bdeb0

This is the first public alpha. It works and it's tested on real hardware, but it's early. Try it from a USB stick first: trying it changes nothing on your computer.

Try it

  1. Download the ISO above and check its SHA-256 (Rufus can do it for you).
  2. Write it to a USB stick of 8 GB or more with balenaEtcher or Rufus, from Windows, no commands: step-by-step guide.
  3. Start your computer from the stick (the boot menu key is often F12, F11, F8 or Esc). The assistant works right away, even offline: a small model comes on the stick.
  4. Install it on a drive when you're ready.
  5. Use it every day and find out what it's good for.

No picture? See the hardware notes.

What it does today

  • Your computer's assistant: it reads the system's own records and explains problems in plain words.
  • Learn Linux, coming from Windows: Mint's real menu names, numbered steps, the right settings opened for you β€” in English, Spanish, Portuguese, French, German and Japanese.
  • Coding, teaching and planning (AICUI): it builds projects with you, explains your mistakes before fixing them, breaks the work into goals and ticks each only when the project's own checks pass. It asks before every change.
  • Writing: story projects that ask you questions until the idea comes together, and a private journal.
  • In the lab, coming next: it watches videos and reads photos and scanned letters for you.

What's on the ISO

Linux Mint 22.3 Cinnamon (Ubuntu 24.04), unchanged except for our packages, plus the built-in guide model (CinMin/guide-4b, Qwen3.5-4B fine-tuned, Apache-2.0) beside the live system so it works offline from the stick. llama.cpp runs every model. Updates come from our signed apt repository, cin-minai.github.io/cinminai-apt (key 7EA7 4B93 128A A958 65E2 BA39 77CA 19E6 2644 70F6, kept offline and used by hand for each release).

Checked before release: the live system differs from Mint's by exactly our packages and their named dependencies; the boot setup is Mint's own; md5sum.txt matches; the model is on it once; it boots in a VM, and installs unattended.

What you need

  • A 64-bit PC that runs Linux Mint 22, and an 8 GB USB stick.
  • For full speed, a graphics card with 6 GB or more: NVIDIA (GTX 10 series and newer) or AMD (less tested β€” reports welcome). Without one it still works, slowly, on the processor.
  • Disk space: what Mint needs (20 GB) plus about 3 GB for the built-in model.

Where it's going

Today, Cin-MinAI is an AI assistant built into Linux Mint. The direction β€” a direction, not a promise β€” is AI moving down the stack: an assistant on the OS today, the AI as the session next, and one day an AI-first OS on a small, deterministic core. Every decision, result and failure is written down in the open: PLAN Β· development journal.

Credits and licences

Led by Ian McClenathan; built with AI assistants (Gemini, Claude, Codex), none preferred. It stands on Linux Mint, Ubuntu, Debian, the Linux kernel and GNU, llama.cpp, the Qwen models, LibreOffice and Mozilla.

Our code is GPL-3.0-or-later; documents and data CC BY-SA 4.0; the guide model Apache-2.0. Linux Mint and the other software on the ISO keep their own licences. Cin-MinAI is not affiliated with or endorsed by Linux Mint.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for CinMin/Cin-MinAI-OS

Finetuned
Qwen/Qwen3.5-4B
Quantized
CinMin/guide-4b
Finetuned
(1)
this model