← All posts

Installable PWA with Full Browser Model Integration

Keating is now an installable PWA that works offline with the browser model. WebGPU-powered Gemma 4 E4B streams tokens in real-time.

Keating is now a fully installable Progressive Web App. Install it from your browser and run it offline with the browser model. The WebGPU-powered Gemma 4 E4B model streams tokens in real-time directly in your browser.

Key Changes

  • Browser Model → Agent: The local Gemma model is now fully wired into the Agent infrastructure. Token streaming works via a custom stream function that dispatches between WebGPU and cloud providers.
  • Automatic Fallback: On load, the app checks for WebGPU support. If unavailable, it automatically falls back to gemini-3.1-pro-preview.
  • Model Selector: The model selector now shows WebGPU status in real-time. Browser option is disabled with a clear message when WebGPU isn't available.
  • PWA Manifest: Added service worker with intelligent caching for WASM files and model weights from HuggingFace CDN.

How It Works

// Hybrid stream function dispatches based on model selection const hybridStreamFn = async (model, context, options) => { if (selectedModelId === 'browser' && webGpuAvailable) { return createBrowserStreamFn()(model, context, options); } return streamSimple(model, context, options); };

Install: Visit keating.help in Chrome/Edge and click "Install" in the address bar, or use the browser's menu → "Install app".