Pick a model, load it, and run it right here — WebGPU + WebAssembly, nothing leaves your machine. First load downloads the model weights (cached after).