The Superheat agent

You probably arrived here from a machine — systemctl status superheatd prints this address, and so does the installer. This page is the orientation; the full operator documentation lives elsewhere and is linked at the bottom.

What it is​

superheatd is the daemon that runs on every GPU host in the Superheat fleet. It is the machine's only connection to the control plane, and it is outbound-only: the agent opens a connection out over HTTPS and WSS, and nothing needs to listen for inbound management traffic.

It is responsible for reporting what the machine has, reporting how it is doing, and running the workloads the control plane sends it.

superheat is the command-line tool that talks to that daemon locally.

The commands you are most likely to want​

All of these are safe to run at any time and none of them disturb a running workload.

# Is the machine healthy, connected, and how many GPUs are free?
superheat status

# Follow the agent log
superheat logs -f

# Run the hardware self-test
superheat self-test --quick

# What version is installed?
superheat version

Every one of these accepts --json if you are scripting against it.

If something is wrong and you need a human to look at it, produce a support bundle and quote the identifier it prints:

superheat logs --bundle

Before you take the machine down​

Do not just stop the service. Open a maintenance window first, so the scheduler stops placing work on the machine and drains what is already there:

superheat maintenance start --duration 2h --category software --reason "replacing a fan"
# ... do the work, reboot if needed ...
superheat maintenance end

Pulling a machine out from under a running workload is the one action that reliably costs somebody else money.

Where the machine keeps its state​

Full documentation​

Installing, enrolling, port forwarding, verification, maintenance, GPU-passthrough VMs and the complete superheat command reference are covered in the Superheat operations documentation, which requires a Superheat Cloud Admin sign-in.

If you operate a machine in the fleet and do not have access, ask your Superheat contact to grant it.

Renting GPUs instead​

If you landed here looking to use GPUs rather than host them, you want the renter documentation instead.