Intelligence where the data is. GPU compute in the rack.COMING SOON
Inference that runs on the customer's own machines and devices at the edge, with our cloud holding the shared memory and doing the training, so every device of a business learns from every other one while raw data never leaves the building. Behind it: GPU nodes on six-month and annual terms, and JOY-OPS, an assistant for running infrastructure. Camera intelligence was the first product built this way; sensors, meters and machines follow the same pattern.
What happens to one frame.
- 01
A camera feeds the PC
Webcam, IP camera, phone, a YouTube link or the screen itself. Everything that follows runs on the customer's Windows PC, on its GPU if it has one, on the CPU otherwise.
- 02
Six layers of understanding
Pose finds people and their joints; face reads identity and emotion; objects finds what is in view and in hands; behaviour combines them (drinking, on the phone, fighting, fallen); scene names the place; memory remembers who did what and when.
- 03
Only learning features travel
A compact numeric description of an object, a pose feature, small thumbnails and one screenshot a minute go to the Joy cloud. The video stream never does.
- 04
The cloud remembers and trains
A shared memory per business confirms, checks and consolidates what every camera saw, and retrains four models on the CPU in seconds.
- 05
Every camera gets smarter
Updated models and names are pushed to every camera of the business within seconds. Click a thing, name it once, and every camera knows it.
At the edge, in the rack, in the console.
Intelligent Vision
Cameras that understand people, things and moments, on the customer's PC. Every camera in a business learns from every other one through our cloud. Free plan to start.
LIVEGPU compute
NVIDIA L4, L40S and A100-class dedicated nodes on six-month and annual terms as custom builds. Open a Sales ticket with model, memory and term.
BY QUOTEJOY-OPS assistant
An assistant for operations: ask about your fleet, your bills, your incidents, in plain language, with the console's permissions.
COMING SOONOne app, a mode for each place.
Visits, mood, drinks
Returning guests by name, drinks per person, waiting too long, hands raised for service.
Footfall and queues
Dwell time per zone, queue length, what customers pick up, blocked persons.
Presence and attention
Working, meeting, attentive, phone in class, after-hours presence, tailgating.
Safety
Helmet or no helmet, falls, danger zones, strangers at the gate, number plates, fire and smoke.
Download the Windows build, read how the six layers work, and see every use case at vision-ai.joycloud.in.