Thread

Model is downloaded and runs locally. It uses WASM/WebGPU if available and runs using Web Workers, so the experience doesn't affect the UI.