Test rig · the same avatar ships on /dashboard/assistant

Giving the assistant a face

A test rig for TalkingHead (MIT licence), running on this domain so you can judge it on your own phone. The 3D avatar renders entirely in your browser — no GPU server involved.

Loading the avatar…

This rig shows the face; speech lives in the app. Rendering needs no sign-in, so anyone can judge the avatar here on their own phone. Speaking uses your workspace's self-hosted voice — no third-party key, ever — and runs on the assistant avatar page, which is signed in and loads its 3D libraries from our own bundle rather than a public CDN.

What this test does and doesn't prove

It proves: how the avatar looks and performs on your actual device, how big the download is, and whether the 3D rendering is smooth enough on a phone.

It does not prove: anything about phone calls — a face can only ever appear in a browser, never on the Retell voice line. Speech here is the SAME self-hosted Piper voice the dashboard assistant uses (no third-party key, no per-character billing); word timings are derived from the exact decoded audio length, which is what drives the mouth.