Join a Hetzner GPU server to your denv cluster.

Hetzner dedicated GPU servers are a first-class path, not a footnote. Status: available. There is no Hetzner API key in the catalog flow. You run the one-line join on each server. It connects out to your cluster. You open no inbound ports and you set up no VPN.

The GPU is whatever is in the server you rented from Hetzner. denv does not pick it from a menu and does not bill you for it. Hetzner does.

What happens after the join

The command installs what the node needs, including the GPU driver, and runs a GPU smoke test before the node is ready. Your servers are a fixed pool. denv can repair them: reset, reboot, driver reinstall. denv cannot order you a replacement from Hetzner. When repair fails, the node stays out and you get the evidence.

That pool can serve the same endpoint as Lambda, Nebius or EC2. A Hetzner site is a good steady base. A cloud site is a good place to burst, if you have set the maximum and you accept that provider’s bill. Failover between them is the point of one cluster.

Air gap and colo

Hetzner is not air-gapped. If you need air-gapped, read self-hosted inference. The join pattern is the same one colo and on-prem servers use. Docs: providers and bare metal.

The product is early. The join command in the docs is the one the console shows once you have an account.