A five-minute phone video becomes a face your agent can wear
Uruvam reconstructs a photoreal 3D digital human from ordinary phone footage, then ships it as a 12 MB browser asset that animates on the user's own device — so an AI agent gets a face without a render farm behind it.
Capture once, animate everywhere
Capture on a phone
Five minutes of ordinary phone video is the whole input. No rig, no studio, no depth sensor.
Build the face
The capture is reconstructed into a photoreal 3D head and compressed into a single 12 MB asset a browser can load.
Ship it once
The asset transfers on first load and is then cached. Every later session reuses the face already on the device.
Animate from speech
The server sends only the agent's speech. The device drives the face from that audio, so nothing has to be rendered upstream.
The server sends speech, never video
That single decision is what removes the render GPU, the encode delay and the per-session cost.
Sub-second responses
Nothing waits on a video encode or a render queue. Audio arrives and the face is already moving on the client.
Zero render GPU
There is no per-session GPU to provision or pay for. Cost per concurrent user is the cost of streaming speech.
Video never leaves the device
Because the server transmits speech rather than frames, no rendered likeness of the user is sent over the wire during a session.
Where the work happens changes everything downstream
| Uruvam | Server-rendered avatar | |
|---|---|---|
| What the server sends per turn | Speech audio | Rendered video frames |
| GPU per concurrent session | None | One render GPU |
| Response latency | Sub-second | Encode + network bound |
| Cost as users scale | Bandwidth for audio | Linear GPU cost |
| Capture requirement | 5 minutes of phone video | Studio session or rig |
Put a face on your agent
Send us five minutes of phone video and we will show you the face running in your own browser.