Skip to main content

​
Prepare a GPU server

Per box:
  • Ubuntu 22.04 or 24.04+
  • An installed NVIDIA driver (nvidia-smi works)
  • Outbound HTTPS to your console URL on port 443 — no inbound access is ever needed
  • A sane clock

​
Install

A token is reusable, so this exact line rolls out a whole rack. Add --site <name> per site, for example --site dal-01. The installer:
  1. Preflights the OS, driver, connectivity, and clock, with loud, actionable failures
  2. Downloads the checksum-verified agent
  3. Installs a hardened systemd unit
  4. Enrolls the machine
  5. Starts streaming
Each node and its GPUs appear in Fleet — in your capacity pool — within two minutes.

​
On a shared box, pass the token in the environment

Command-line arguments are world-readable through ps. On a machine where that matters, hand the token over in the environment instead:
Note the leading space before export — that keeps the line out of shell history. Drop the --token flag entirely in this form. The installer forwards the variable to the agent the same way, so the secret never reaches any process’s argv.

​
Verify on the box

​
Removing the agent

The installer is idempotent — re-running it is safe.

​
Behavior during network outages

Agents buffer 24 hours of telemetry on disk and backfill with original timestamps after outages, up to a 55-minute replay cap. Older gaps are shown honestly as gaps rather than interpolated. Fix the network and telemetry backfills itself; no intervention needed.