Skip to content

Getting started

This walks you through fleet on the machine you work from: installing it, adding one other machine, and a first job on that machine. It takes about ten minutes. You talk to your coding agent the whole way. The command behind each step is shown too, for when you would rather type it yourself.

You need:

  • a coding agent that can run commands — Claude Code, Codex, Gemini CLI, Copilot CLI, OpenCode or any other;
  • one more machine you can already reach with ssh from this one: a server, a desktop, a cloud VM or a GPU rental. Nothing has to be installed on it. If the two are behind different NATs, put them on a mesh network such as Tailscale or ZeroTier first (why).

fleet connects to that machine exactly as ssh does from here — your ssh config, your keys, Tailscale SSH — so if ssh gets in, fleet does too. It manages no access of its own, and never asks for a password.

1. Install fleet

Ask your agent:

Install fleet from https://github.com/lion-zhang/fleet

The agent runs the installer. It installs the fleet command, records this machine, and teaches every supported agent on this machine to use fleet — not only the one you asked. It ends by listing what to try next. Start a new agent session afterwards, so the agent loads fleet's skill.

Installing it yourself 
curl -LsSf https://raw.githubusercontent.com/lion-zhang/fleet/main/install.sh | sh
On Windows, in PowerShell: 
powershell -ExecutionPolicy ByPass -c "irm https://raw.githubusercontent.com/lion-zhang/fleet/main/install.ps1 | iex"
Options, other package managers and upgrading: [Installing and upgrading](guides/install.md).

2. Look at this machine

You: show me this machine

The agent runs fleet show and describes what fleet measured: CPU and load, memory, GPUs and their free VRAM, disks, the busiest processes, the services listening on ports, and the facts agents use to pick a machine (linux, cuda, vram-24g, …).

3. Add a machine

Give the agent the SSH command you already use for the other machine:

You: add ssh ubuntu@10.0.0.7

Agent: Added as gpu-box: 2× RTX 4090, both idle, 46 GB free. It's ready to use.

The command is fleet add "ssh ubuntu@10.0.0.7". fleet connects with whatever already works for you — your ssh-agent, your ~/.ssh/config, a key file you named with -i — measures the machine, and records it in this machine's list. It writes nothing on the other machine.

Or let the agent look for you:

You: what machines could I add?

It runs fleet ls --discover, which lists the hosts in your ~/.ssh/config and the machines on your tailnet, and offers them to you; fleet add NAME adds one.

If the machine does not let fleet in, nothing is recorded, and fleet says why and what to do — usually to put your key on it once. fleet doctor NAME prints the exact line; at a terminal, fleet add offers to run ssh-copy-id, which asks for the machine's password itself. For rentals, Windows and Mac machines and Tailscale SSH, see Reaching your machines.

4. Use it

Ask for what you need, not for a machine by name:

You: what's free right now?

You: run nvidia-smi on the box with the most free VRAM

The agent picks the machine from what each one has and what is free on it at that moment, then runs the command there with fleet ssh and brings the result back. Passwords and keys never go into the conversation; the agent may see a machine's address, as fleet show prints it.

To watch every machine live yourself, run fleet top in a terminal (q quits).

5. Optional: fleet on another machine too

The machine you added is watched from here; its own agents do not use fleet. If you also work on that machine, install fleet there the same way, or from here:

You: install fleet on gpu-box

The agent runs fleet install gpu-box. Each machine running fleet keeps its own list, of the machines its own ssh reaches — there is no center, and nothing to keep in sync. To use the same list on both, share machines.yaml (fleet paths shows where) with your dotfiles.

Next steps