Local AI Hardware Test Lab

A shareable, no-account hardware test for browser-local AI. It checks whether this browser can use GPU acceleration, applies the model's approved fallback policy, runs a real pretrained model when supported, and creates a report you can copy or download yourself.

đź”’ Your files never leave your machine.

Model code and weights may be downloaded from third-party hosting on first use. The image you choose is passed to the model inside this browser and is not uploaded for inference. No test report is submitted automatically.

Compatibility

WebGPU API: Checking…
Test ID: Creating…
Model: Checking…
Device class: Checking…
Preferred backend: Checking…
Fallback policy: Checking…
Checking whether this device can run the model efficiently and privately…

Real pretrained vision test

Choose any ordinary photo. MobileNetV4 will classify it locally if this device passes the execution policy. The first run can take longer while model/runtime assets download.

No image selected.
Not run yet.

Test report

The report contains technical compatibility information only. It does not contain your image or its contents. Nothing is sent automatically.

Report will appear after compatibility checking.

Share the report manually with the Clean Local Tools maintainer if you choose.

Foundation compute probe

Backend: Checking…
Reason: Checking…
Model: Checking…

This tiny deterministic model validates our worker plumbing without downloading the real vision model.

Not run yet.