What an agent is doing before it answers: live steps, searched sources, tool calls, streaming reasoning and a 'Thought for 12s' summary.
pnpm dlx shadcn@latest add "https://ui.knowns.dev/r/thinking-state.json"