Runs 100% in your browser via
transformers.js.
No server, no API calls — the model
(SmolLM2-360M-Instruct)
downloads once and generates locally. Uses WebGPU when available, WASM otherwise.
Click send to load the model (~300 MB, cached after first load).