On-premises AI
On-premises AI, installed on your servers
The engine that reads your rack photos runs on your premises. It proposes a result, marks what it is not sure about, and a person reviews it before any use.

Contents
An engine that runs on your premises
The analysis engine is installed on your servers, along with the rest of the platform. It calls no online service and works without an internet connection: photos are analyzed on site and never leave your network. It takes advantage of a graphics card if the server has one, and also runs on a processor.
What it reads on a photo
- The family of each piece of equipment: network, fiber and transmission, power, server, patching…
- The model, when it is readable on the front panel.
- Its position in the rack, in U.
- A confidence level, shown next to the result.
Visible labels are read on the server, with no external service, and everything feeds a rack diagram graduated in U.
Uncertain results are flagged
When the confidence is too low, the device is marked “to check”. It is never presented as certain. The threshold below which a device is marked is set in the settings by an administrator, and the summary of each analysis shows how many devices remain to check.
The engine of your choice
Several analysis engines can be installed. An administrator chooses the default one, and each analysis can be run with another. The result always shows the engine that produced it, and the System page in the settings shows the installed version, the acceleration used and the services called.
A person always reviews
The engine proposes, a person reviews: devices marked “to check” are checked against the photo, and the diagram is corrected before export. The platform writes to none of your systems; what it produces are results and files that you review before using them. See also our page on EU AI Act.