cfahlgren1/webllm-playground — reverse-engineered prompt

Build me a clean browser based playground for running local LLM chats using WebLLM and MLC. I want it to feel like a simple ChatGPT style app, where someone can pick a model, wait for it to load in the browser, type messages, and see the replies stream back without needing a server.

Please include a nice model selector, clear loading status, chat history, and a small stats area that shows useful chat or generation info like speed, tokens, or timing if the WebLLM API provides it. Make the UI polished and responsive with a modern purple or indigo style.

Use TypeScript and the existing Vite style setup. Keep it static so it can be deployed easily, like on Hugging Face Spaces or any static host. Look up the current WebLLM docs online if you need to, and make sure the app handles errors clearly when a model fails to load or the browser doesn’t support what it needs.