fleet

The problem

Ask your agent to train a model and it starts on your laptop — while a 4090 sits idle across the room and a rented A100 bills you by the hour. It cannot use what it cannot see.

• Your agent is stuck on one machine. It has no idea your other boxes exist.
• Finding a free GPU is manual. SSH into five hosts, run nvidia-smi, compare in your head.
• Handing it a server means pasting credentials into the chat, and hoping.

Install once, then just talk to your agent

On the machine you work from:

curl -LsSf https://raw.githubusercontent.com/lion-zhang/fleet/main/install.sh | sh

_{Windows: powershell -ExecutionPolicy ByPass -c "irm https://raw.githubusercontent.com/lion-zhang/fleet/main/install.ps1 | iex"}

That is the whole setup. This machine becomes your fleet's center, and every supported agent installed on it learns fleet. From here on you say what you want in plain words — no commands to remember. When something is missing, the agent asks (illustrative):

You: add my new GPU server

Agent: Sure — how do you usually connect to it? An SSH command like ssh -p 40001 [email protected] is all I need.

You: ssh [email protected]

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

You: train train.py on whatever has a free 24 GB card

Agent: rtx4090 has 23.1 GB free and an idle GPU; a100-spot is free too but costs $1.89/hr. Starting on rtx4090, logging to train.log.

No hostnames, keys or passwords go into the conversation, and anything irreversible waits for your yes.

Every machine besides the center is a member. A member needs nothing installed — just sshd. For machines where you also want to run fleet, or that you would rather not type a password for, the agent gives you an invite line: pasted there, it installs fleet and joins by itself.

Already in your agent? Install from there

Plus OpenCode, Amp, Windsurf, Cline, Zed, Qwen Code, Goose, Kiro, Hermes — 35+ agents in all: every agent →

On a machine in no fleet yet, these make it a center on first use, like the installer. For a member, paste its invite line first.

Built to be safe

• Nothing to install on your machines. fleet probes with one script over one SSH connection — Linux, macOS or Windows. A NAS or a fresh rental works as it is.
• Keys, not passwords. A password, if needed at all, is typed once by you. Nothing that could be stolen is stored, and the agent never sees a credential.
• Access you control. The center decides which machine may reach which, and a revoke is pushed at once. fleet access gpu-box --allow laptop, --deny to take it back.
• The agent asks, not guesses. It is told to ask which machine you mean, and to leave irreversible commands to you.

Prefer the command line?

Everything the agent does is a plain fleet command, for when you want to drive it yourself:

fleet ls                            # every machine, with what is free right nowfleet ls --tag cuda --tag vram-24g  # by capability: NVIDIA, a card of 24 GB or morefleet top                           # live view, like htop for the whole fleetfleet show gpu-box                  # one machine in detail: GPU processes, services, disksfleet ssh gpu-box -- nvidia-smi     # run something therefleet add "ssh [email protected]"     # add a machine; fleet invite NAME for a join linefleet access nas --allow laptop     # let one machine reach another

Rentals from vast.ai, RunPod or Lambda show their price (fleet edit a100 --cost 1.89), the fleet's burn rate, and an alert when a paid machine sits idle. On Tailscale, ZeroTier or WireGuard? fleet just needs an address it can route to.

How fleet compares

fleet is about the machines you already have. It complements tools that launch new ones.

FAQ

Learn more

• Getting started — the full walkthrough, every command
• Every agent — install commands and config for 35+ agents
• Design — one core per machine, a skill per agent, MCP for the rest; and access — how access is granted, signed and revoked

Status: v0.5. The test suite runs on Linux, macOS and Windows, and every command is run end to end on a real machine of each OS in CI — from a script, as an agent runs it, and at a real terminal, as you do. Multi-machine fleets (key, password, invite, handover) are tested on Linux machines built from scratch.

Issues and pull requests are welcome — uv run pytest -q runs the tests. If fleet saved you a GPU-hour, a ⭐ helps other people find it.

MIT