Hey everyone, about two months ago I shared an early black-and-white version of this component here (back when it was called agent-robot-avatar). I've kept iterating on it since then, gave it an official name (BoopFace) and built a website. Everything finally went live yesterday, so I wanted to share how it looks now.
If you haven't seen it before: it's an MIT-licensed native Web Component ( <boop-face> ), written in vanilla JS and SVG with zero runtime dependencies. When an LLM or Agent runs tool calls, you can directly call avatar.startWaiting() , avatar.play('wary') , or avatar.play('refuse') to turn states into expressions.
What's new since my last post:
Expressions are now up to 18, plus press-and-hold squeeze and a draggable spring antenna (it gets angry after you pull it).
188 3D-style skins completed so far. Every skin is still drawn in pure SVG with pointer-following light, using no bitmap images at all.…