Setups for different situations are in Deployment examples by situation, and installation steps are in the Installation chapter.

Overall setup

Install Edge SDK on the field PCs (small edge PCs placed in the factory) and MLOps Central Platform on the central server. MLOps Central Platform provides the management screen, so staff PCs need only a web browser.

The whole setup: robot arms, PLCs and cameras in the factory send data to a field PC running Edge SDK; the field PC exchanges data, new AI models and instructions with the central server running MLOps Central Platform. Operations staff, AI developers and approvers open the central server's management screen in a web browser. Field workers get verdicts from the field PC

In the pictures, the chip icon marks an edge PC running Edge SDK, and the M icon marks a server running MLOps Central Platform. The pictures in Deployment examples by situation use the same icons.

ComponentLocationWhat it does
Field devicesFactoryDevices such as robot arms, PLCs and cameras produce readings and product photos
Edge SDKField PC (small edge PC)Gathers data from several devices and runs the AI model on site for a verdict. When the link drops it keeps the data and sends it once reconnected
MLOps Central PlatformCentral serverStores field data and trains AI models. Decides through approval which model goes into production, and sends the models staff choose to the field. Provides the management screen
Web browserStaff PCOpens the management screen. Shows device status and handles training, approval and deployment

Users

Centre users can use only the features their role allows.

UserWorks inWhat they doPlatform role
Field workerFactoryHandles the devices and sees AI verdicts on the field program's screenNone (does not use the platform screen)
Operations staffHead office · centreWatches device status and alerts, and deploys models to the fieldOperator
AI developerHead office · centreBuilds datasets, trains models and compares resultsDeveloper
ApproverHead office · centreApproves or rejects the models that go into productionApprover

Connection principles

Both principles apply to every deployment example.

1. The field connects first

Edge SDK in the field connects to the centre, reports its status and asks for work. New models and instructions come back as the response on that connection. The centre never connects into the factory first, so no inbound path is opened in the factory firewall. The only address that must be reachable is the central server's.

A connection from inside going out is outbound; one from outside coming in is inbound.

WhereOutbound (inside to outside)Inbound (outside to inside)
Factory (field PC)Uses one connection out to the central serverNot opened
Central serverNever connects to the field firstOpens only the server address so field PCs can come in

The field connects first and the centre answers on that connection. The path from the centre into the factory stops at the firewall, and there is no need to open it

While the link is down, data is kept on the field PC; when the link returns, the most important data goes first. AI verdicts are made on site, so the production line keeps getting verdicts while disconnected.

Next: Deployment examples by situation

Written for the platform as of 2026-10-02.

© MLOps Central Platform