GitHub - kessler/gemma-gem: Gemma Gem runs Google's Gemma 4 model entirely on-device via WebGPU — no API keys, no cloud, no data leaving your machine.

Gemma Gem is an open-source Chrome extension that runs Google's Gemma 4 language model entirely on-device using WebGPU — no API keys or cloud required. It injects a chat overlay into any webpage and can read page content, click elements, fill forms, execute JavaScript, and take screenshots via an agent loop. The architecture splits work across an offscreen document (model inference + agent loop), a service worker (message routing, screenshot/JS execution), and a content script (UI + DOM tools). Two model sizes are supported: E2B (~500MB) and E4B (~1.5GB), both using q4f16 quantization with 128K context. Built with WXT, Hugging Face Transformers.js, and ONNX-quantized Gemma 4 models.