Skip to content

GenClass v0.1.0 — browser extension + 32M model

Latest
Latest

Choose a tag to compare

View all tags
@MeharPro MeharPro released this 04 Oct 13:49
· 2 commits to main since this release
v0.1.0
c2ddfa5

GenClass v0.1.0 is the first public build of the browser extension. It is an independent, open-source reimplementation of a Jev-style decision model and is not affiliated with TypeSafe.

Install

  1. Download genclass-0.1.0.zip below and unzip it.
  2. Open chrome://extensions, turn on Developer mode, click Load unpacked and pick the folder. Requires Chrome 116 or newer.
  3. Click the GenClass icon to open the side panel. On first run it downloads the model and the Moonshine speech model, and caches both. It starts in dry-run mode; turn on Live in Settings.

What's in it

  • Voice browser control. Moonshine speech runs on-device (first partial in about 150 ms). Actions fire mid-sentence. Destructive actions need a spoken "confirm", password fields are blocked, and there is a kill switch.
  • Focus mode, which parks off-task tabs with a grace period and undo. Experimental.
  • RAM manager, which unloads idle tabs. Experimental.
  • Content filter. Experimental.
  • Speech engines: Moonshine is the default; Whisper base.en and Whisper Turbo (WebGPU only) are optional.

Assets

  • genclass-q8.onnx (WASM) and genclass-fp16.onnx (WebGPU): the 32M computer-use decision model, trained only on synthetic data, Apache-2.0.
  • tokenizer.json, calibration.json, meta.json, model.json.
  • ASSETS.json: sha256 of every file. The extension verifies each download against it.

See BENCHMARKS.md for the head-to-head numbers against Jev, including where it loses.

Known gaps: live microphone and WebGPU have only been tested in CI-style headless Chromium so far. The Chrome Web Store listing is pending.

Assets 10
Loading