Without a central publishing point, every production station could over time end up running a different model version — with no record of who put it there or when. Server prevents that: it runs as a local deployment (Docker or NAS) inside the plant network and distributes packages only to the stations you point it to.
Access to publishing new versions is controlled — not everyone with network access can change the model running on production. Every publication is recorded in a history, so a question like "which version was running on line 3 last week" has an immediate answer.
Server does not require a connection to a vendor cloud. All communication happens inside the plant network, between the Trainer workstation, Server and the Production stations.