tambo-ai/tambo ? reverse-engineered prompt

I want to build a React application where the AI doesn't just respond with text, but actually generates parts of the user interface. It should have a chat input where I can make requests, and the system dynamically renders the right React component. For example, if I ask it to "show me sales by region," it should render a chart. If I tell it to "add a task to my board," it should update a task list component that's already on the screen.

The key is that I want to be able to "teach" the AI about my own custom components, defining what they do and what kind of data they need. The framework should handle all the hard stuff like streaming the props to the components and managing the conversation state. I'll want to plug in my own API key from a service like OpenAI or Anthropic. For the backend, it would be great to have the choice between a simple hosted cloud option or self-hosting it with Docker.