
Building environment perception systems to help humans and machines better understand the world around them.
Seeing the world through electronics.
Trinet
Third eye for your AI.
A family of hardware-synchronized egocentric cameras for teams training VLA models. Every frame, IMU sample, and audio sample on one clock - with a hardware mute switch for spaces where audio isn't welcome. Stream over standard UVC or record straight to the device.
- Trinet Mono - Ultra-wide 1080p30, 400 Hz motion data and stereo audio on one clock
- Trinet Stereo - Two cameras 70 mm apart, calibrated per unit - self-contained, no RPi or Jetson
- Trinet Stereo GS - The stereo camera with global-shutter sensors, for fast motion without skew
- Trinet Wrist Kit - Head and both hands on one shared timeline, under 1 ms apart
Trinet Studio · Now live
From SD card to dataset.
Upload what your Trinet cameras record. Trinet Studio returns 3D hand poses, head motion and action labels for every frame - ready to review in minutes, ready to export as a LeRobot dataset. Or just check your recordings and split them into tasks.

A Wrist Kit recording in Studio: head and wrist cameras in sync, both hands in 3D, a room scan with the head's path, and the action labels on a timeline.
- 01
Record
Wear a Trinet head camera or a Wrist Kit and do the task.
- 02
Upload
Open the SD cards in Studio. Upload all of it, or only the parts you keep.
- 03
Process
Full annotation, hands only, or QC only. You see the exact price first.
- 04
Review
Every camera in sync with the 3D hands. Accept or correct each label.
- 05
Export
A LeRobot dataset, plain JSON and Parquet, or your original video and MCAP.
- ≈ 5 min
- To ready-to-review, for a 2-minute Wrist Kit recording
- Both hands
- In 3D on every frame, through brief occlusions
- Faces blurred
- In every review video and annotation export
- Per hour
- Of video, with trial credit for new accounts
The road to Physical AI
Capture today. Models tomorrow.
Physical AI is bottlenecked on data. We're building the stack that unblocks it - starting with the hardware that collects it.
Shipping now
Egocentric capture hardware
The Trinet family - head camera, stereo, and the Wrist Cam Kit - wearable, hardware-synchronized cameras that record what your operators see and do, on one shared timeline.
Explore TrinetNow live
Trinet Studio
Upload Trinet recordings and get 3D hands, head motion and action labels for every frame - reviewed in your browser, exported as a LeRobot dataset. Fine-tuning comes next.
Open Trinet StudioIn development
More modalities
A UMI-style data-collection gripper and tactile gloves - manipulation and touch data captured with the same synchronized-timeline discipline as Trinet.
The goal
Models that understand the physical world
Everything we build points the same direction: vision-language-action models trained on rich, precisely-timed egocentric data.
Trinet is for machines. Panocle is for humans.
Our mission runs both ways. Trinet helps machines understand the world around them. Panocle - our AI-first smart glasses, in development - helps humans understand the world around them: perception, understanding, and assistance, from hands-free computing to support for the visually impaired.
Meet PanocleWord on the street

ET Startup Awards 2026 - Nominee, Best On Campus
The Economic Times
Read

30 Startups To Watch
Inc42
Read

Point Break: Teaching AI to See
AIM
Watch

Built in India, Not China
Runtime
Watch

AI: The Access Key to Inclusion
The Economic Times
Read

IIM Calcutta feature
AINA - AI & Analytics Magazine
Read






