> CUSTOM_AI.train()

Your use case. Your model.

The 24 ready-made modules cover the common cases. When yours is different, we build and train a model for exactly what you need to detect, and run it on the same on-premise edge platform.

TELL_US_WHAT_TO_DETECTSEE_READY_MADE_MODULES
CAM_11 · WAREHOUSE_A
LIVE
CAM_11 · WAREHOUSE_A camera view
EDGE-NODE-01 · ON-PREM
INFERENCE …

Illustrative example, not a client deployment.

> WHEN.check()

When custom makes sense.

It is not in the library.
A specific product defect, an unusual object, a behavior that only exists on your site.
The library model is not enough.
Your uniforms, vehicles, lighting, or camera angles need better accuracy than a general model gives.
It must fit your process.
Your own classes, rules, and outputs, wired to your systems.
> EXAMPLES.list()

What can be detected.

Illustrative examples of the kind of thing a custom model can do. These are not client projects.

Product or packaging defects on a line

Specific vehicles, forklifts, or container and fleet IDs

Custom PPE or uniform compliance

Behaviors specific to your site, like a blocked emergency exit

Counting items on a conveyor, pallet, or truck bed

Equipment state: running, stopped, or jammed

> PROCESS.run()

From brief to edge.

01
SCOPE
Define together what to detect, where, and what success means.
02
COLLECT
Gather sample footage from your cameras, covering different times of day, lighting, and angles.
03
LABEL & TRAIN
We label the data and train the model, with the training setup agreed per project.
04
VALIDATE ON SITE
Test on your live cameras and review results together.
05
DEPLOY & IMPROVE
The model runs on the edge box, and we retrain as your needs change.
LIVE
RUNNING
  • T+000 [SCOPE] PROJECT_A · define together what to detect, where, and what success means
> SCOPE.define()

What you bring, what you get.

What we need from you
  • A clear description of what to detect, with examples
  • Sample footage from the real cameras, in whatever format your VMS or NVR exports
  • Camera locations and access to them
  • A definition of a "correct" detection, and how you want alerts delivered
What you get
  • A model running on your on-premise edge box, alongside any ready-made modules
  • Alerts and dashboard in the same interface as the other modules
  • Results reviewed together on your own footage before go-live
  • Retraining as your environment changes, scoped with you
> TERMS.read()

Data, ownership & timing.

Data & privacy
  • Your footage is used to build your model for your project.
  • Where training happens and how long footage is kept are agreed per project.
  • After deployment it runs on-premise, with no cloud dependency, like the rest of the platform.
Ownership

Ownership and licensing of the trained model are agreed when the project is scoped.

Timing

It depends on what you need to detect, how varied your footage is, and the accuracy the use case needs. We give a realistic estimate after scoping, not before.

> FAQ.query()

Common questions.

Have a use case in mind.

Describe it and we will tell you what is possible.

REQUEST_DEMO_OR_QUOTE