a2ui-project/a2ui ? reverse-engineered prompt

Build me an early preview of A2UI, an open source way for AI agents to send rich interfaces as JSON instead of raw code. I want a working project where an agent can describe screens like cards, forms, buttons, text fields, and simple dashboards, and the client renders them using trusted native components so it stays safe.

Please make it support incremental updates so the UI can stream in or change as the conversation continues. Include a simple end to end demo, something like a restaurant finder or booking flow, plus examples of a remote sub agent returning a form or an approval style panel inside the main app.

It should feel portable across clients, with web support and Flutter support if that is already part of this repo, and it should be easy to extend with custom components later. Please wire up the docs, samples, and local run experience so I can clone it, add an API key, and see it working right away. Look up current docs online if you need to.