Google Gemma

I have enough context. Let me write the description. Gemma Chat is an open-source Electron app that runs Google's Gemma 4 model natively on Apple Silicon Macs, turning a laptop into a self-contained coding agent that works on a plane, in a cabin, or anywhere without Wi-Fi. You describe what you want to build, and it writes the code, HTML, CSS, JavaScript, multi-file projects, with a live preview that updates as the model types. No internet connection needed after the initial model download. The project was built by Ammaar Reshi as a proof-of-concept for fully offline, local-first coding using a small open model. The app ships with a Build Mode that writes multi-file projects into a sandboxed workspace alongside a Chat Mode with tool use for web search, URL fetch, calculator, and bash. It supports hot-swapping between four Gemma 4 variants on the fly , ranging from a 1.5 GB E2B model for fast Q&A up to a 31B variant for maximum quality on machines with 32 GB of RAM. Inference runs through MLX-LM, Apple's framework for running LLMs on Apple Silicon , with local Whisper handling speech input in the browser via WASM. The launch matters because it pairs Google DeepMind's newest open-weights release with a concrete, installable consumer app that keeps prompts, code, and conversations on the user's own machine. For founders and operators evaluating local AI for privacy-sensitive workflows or air-gapped environments, it's a working reference for what a small open model plus Apple's unified memory can actually do without a cloud API.