CoolFace
Apppublic

Thox-ai/ThoxyWeb

sourceHugging Faceotherupdated 26d agoView on Hugging Face
0likes
App README

ThoxyWeb — chat with THOXY in your browser

Your AI. Your Data. Your Rules. ThoxyWeb is the proof-of-concept that lets you chat with THOXY — the THOX assistant — running entirely client-side in your browser on the Gemma-4 WebGPU model. No servers. No telemetry. No account. The model, the tokenizer, and every compute kernel execute inside the browser; nothing you type ever leaves your machine.

  • —Live Space: https://huggingface.co/spaces/Thox-ai/ThoxyWeb
  • —Product line: https://thox.ai
  • —Model mirror: https://huggingface.co/Thox-ai/ThoxWebby-Gemma-4-E2B
THOXY is the assistant. ThoxyWeb is this browser surface that runs it. It shares the THOX design system and chat UX with ThoxOS Web Edition, so the two feel like one product.

The point of this Space

ThoxyWeb proves the local-first thesis end to end: a capable assistant that runs on-device with no backend. Open the page, load the model once (weights cache locally after the first load), and talk to THOXY — the inference happens on your own GPU via WebGPU compute shaders.

ThoxRoute — local-first routing

ThoxyWeb routes through ThoxRoute, the THOX model surface, using the vendored in-browser router (vendor/thoxroute-client.mjs, kept byte-identical to `ttracx/thoxroute-client`). ThoxRoute decides locality before any data exists:

  • —Route: browser-local. Chat runs on-device via Gemma-4 on WebGPU. The router's probeCapability() detects the backend (WebGPU / WASM), device class, and memory budget, shown live in the ThoxRoute panel.
  • —Cloud fallback is a data-boundary crossing, not a silent downgrade. ThoxRoute's only configured chat fallback tier is gated / internal-only and is therefore not enabled on this public brand-safe surface. ThoxyWeb stays on-device; it never sends your prompts off your machine.

Model

This build runs Google's Gemma-4 E2B (QAT Mobile) — a rebrand-and-configuration wrap of `google/gemma-4-E2B-it-qat-mobile-transformers`, mirrored for THOX use at `Thox-ai/ThoxWebby-Gemma-4-E2B`. A larger Gemma-4-E4B browser runtime is being evaluated as the next step; ThoxyWeb ships the measured, working E2B path today rather than an unverified one.

How it works

  1. 1.Open ThoxyWeb in a WebGPU-capable browser (Chrome 113+, Edge 113+, or Safari 18+).
  2. 2.Press Load model — the Gemma-4 E2B (QAT Mobile) weights download once and cache locally.
  3. 3.Chat with THOXY. Every token is generated on your device with WebGPU compute shaders.
  4. 4.No servers. No telemetry. No account.

Attribution

  • —Base model: google/gemma-4-E2B-it-qat-mobile-transformers — Google (Gemma license).
  • —WebGPU kernel authoring: original kernels by Fable 5, upstream Space `webml-community/gemma-4-webgpu-kernels`.
  • —ThoxRoute in-browser router: `ttracx/thoxroute-client` (vendored).
  • —Rebrand, THOX house voice, and product integration: Thox.ai LLC.

Standards

  • —Owner: Thox.ai LLC
  • —Independence: THOX is an independent AI company. This project is not affiliated with, endorsed by, or sponsored by Google, Meta, Alibaba, Anthropic, OpenAI, Mistral, or Qwen. It applies THOX branding, THOX house voice, and THOX product wiring to an open-source model released by Google.
  • —License: Base model weights are governed by the upstream Gemma license. THOX-authored code in this Space (index.html, README, THOX configuration) is provided AS-IS with no warranty, express or implied.

Your AI. Your Data. Your Rules.