Usuba Checker Model
Usuba is a model, not an app or a plugin. The Usuba Checker Model is a small model that runs locally on your own laptop, for free, trained only on Omarchy plugin data.
Usuba has two complementary parts: the scripts that test a plugin, and the model that judges it. The scripts are plain Python, not AI, and they are available now — see the Checker Scripts. The model is in training.
Status
Section titled “Status”| Piece | What it is | Status |
|---|---|---|
| Usuba Checker Scripts | The tests that check a plugin | Available now |
| Usuba Standard v0.1 | The rulebook the scripts enforce | Available now |
| Usuba Checker Model | The model that reviews a plugin | In training |
What it runs on
Section titled “What it runs on”- Based on Qwen 3.5 9B.
- Runs at 4-bit quantization, about 5.5 GB of weights.
- On a typical 32 GB laptop, with no graphics card.
What it is trained on
Section titled “What it is trained on”Omarchy plugins verified with the Checker Scripts and tested on real Omarchy hardware, from a sweep of more than 1,400 published plugins so far. The scripts’ results across those plugins are part of the training data too.
What it will do
Section titled “What it will do”The model reviews a plugin and says what to fix. Give it the plugin, the scripts’ report, and what the plugin shows on screen, and it judges what the scripts can’t: whether the plugin really works, whether the data is real or faked, whether the buttons do anything, whether it breaks in use, and what Omarchy’s plugin reviewer would flag. Then it explains what to fix.
It is in training, so this is what it is built to do, not what it has proven.
No published results yet:
The Usuba Checker Model is in training. There are no published model results. When there are, they will appear here with the test method, the model versions and the dates attached.