Libraries such as Transformers.js already support running lightweight models like Phi-3 and Qwen locally in the browser for inference, without requiring network requests.