thesysdev/openui ? reverse-engineered prompt

I want to build a "generative UI" chat application using React and TypeScript. The core idea is that when a user talks to the AI, the AI can respond with rich, interactive UI components, not just text. For example, it should be able to generate and display things like charts, data tables, and forms directly in the chat history. The UI should stream in piece by piece as it's being generated to feel really responsive.

Let's use the OpenUI framework to build this. It seems to have a special language optimized for this exact use case that's more token-efficient than JSON. Please scaffold a complete starter project that includes a working chat interface and the necessary OpenUI packages for rendering. Make sure to set up an environment file where I can plug in my OpenAI API key to get it running.