Every AI app I worked on needed the same small pieces: a thinking indicator, a task checklist

that shows progress, a voice orb, a card that shows what a tool call is doing.

I kept rewriting them, so I collected them into one library.

What I focused on:

- Plain React + Tailwind. The code is copied into your project, no runtime package.

- Calm motion, and everything respects prefers-reduced-motion.

- Each component has a props playground so you can tweak it before copying.

The newest one is a tool-call card: the arguments stream in, a status pill runs,

then the result slides in. It takes a Vercel AI SDK tool part directly.

Still early, so I'd really like feedback: which part of an agent UI do you find hardest to get right?

(Link in the comments.)