
A chat that answers the moment it lands on a page, built on the Vercel AI SDK. The message list and the composer come with it, and text, reasoning, tool calls, citations and attachments each get a renderer of their own, so a stream arrives drawn rather than as raw JSON. A scripted conversation stands in until you point it at your own model.
@shadcncraft
lib/demo-chat.ts is the file to delete: point the transport at your own route, then drop @shadcn/helpers. Your route reads the model, effort, tools and connectors out of the request body beside messages, and should check the model id against the list composer-1 ships before spending anything. Each file says at the top what it is for and what to change in it.






















