mlc-ai/web-llm-chat ? reverse-engineered prompt

Build me a private AI chat app that runs fully in the browser, so conversations stay on the user’s device and nothing needs to go to my servers.

It should feel like a clean modern chat app, with a sidebar for chats, markdown rendering, dark mode, responsive layout, and a smooth message stream. Let users pick from built in browser models, and also let them connect to a local custom model through an MLC LLM REST API if they want. I’d also like image upload support for vision style chats, plus an experience that works offline after the initial model download. Make the setup simple, include clear settings for model selection and endpoint configuration, and keep the whole thing polished enough to deploy as a real public app. If you need to check current docs for the model API or browser GPU support, look them up online first.