Give your coding agents one always-on Linux machine, reach it privately
from anywhere, and keep several sessions working when your laptop closes.
You do not need a GPU or a dedicated AI server.
CPU8 shared x86 vCPU
MEMORY16 GB RAM
STORAGE100–160 GB SSD
SYSTEMCurrent Ubuntu LTS
Have an agent help
Relay the setup through an agent
Give this brief to an agent on the computer in front of you. You
keep control of provider login, billing, credentials, and approval.
Help me set up a cloud virtual machine for a team of AI agents and Ronin Cowork.
Start by opening and following this agent-facing guide:
https://github.com/ronincowork/ronin-cowork/blob/master/docs/hetzner-vm.md
Explain each choice before acting. I will handle provider login, billing, passwords, authentication, and final purchase approval. Never ask me to paste a private key, password, token, recovery code, or payment detail into chat. Finish by handing the remote agent to docs/install.md and return the private Ronin address.
01
Use an ordinary cloud VM—not a GPU box
Claude Code, Codex, Gemini CLI, and similar agents send model work
to their providers. The VM runs their terminal applications,
repositories, tools, and sessions; it usually does not run the
foundation model itself. That makes RAM and reliable storage more
useful than a local GPU.
Shared CPU is normally the economical choice. Dedicated CPU is
useful only when your own builds, tests, databases, or local models
need sustained compute—not merely because the workload involves AI.
02
Choose a provider
Any reputable provider that offers an x86 Ubuntu VM, SSH access,
persistent SSD storage, and a region you can reach will work. Compare
the complete live price, including IPv4 and backups, before buying.
Ronin's proved value choice is Shared Cost-Optimized x86 CX43:
8 vCPU, 16 GB RAM, and 160 GB disk. Capacity is limited and can
disappear temporarily. CPX22 is a smaller Regular Performance
x86 bridge that can later rescale to CX43.
Choose a Basic shared-CPU x86 Droplet near the same memory and
storage target. DigitalOcean's control panel is approachable;
its dedicated General Purpose plans are usually unnecessary for
terminal agents.
Vultr Cloud Compute is another shared-CPU VM option. Select an
x86 Ubuntu image, then size memory and storage to the team you
expect to keep active.
03
Size for concurrent agents
A useful planning allowance is roughly 700 MB for each live agent,
plus memory for Ubuntu, Ronin, tmux, Node, editors, builds, and any
services. The 8-vCPU / 16-GB recommendation leaves room for a real
team; a 2-vCPU / 4-GB VM is a starter box for setup and a few light
sessions, not the same capacity.
Small start: 2–4 vCPU, 4–8 GB RAM, at least 80 GB disk.
Large team or heavy builds: 16 vCPU, 32 GB RAM, 200+ GB disk.
Set a maximum number of concurrent sessions. When RAM is exhausted,
Linux may end an existing process; it does not politely queue the next agent.
04
Set up access without exposing the agents
Select your SSH public key when the VM is created.
Use the initial root login only to create an ordinary sudo account.
Run every agent, repository, tmux session, and Ronin under that account.
Put the VM and your devices on the same private Tailscale network.
Never open the Ronin application port directly to the public internet.
Adding an SSH key to a provider account after creation may not install
it on an existing VM. Keep the provider console as a recovery route until
direct SSH and private-network access are both proved.
05
Install the agents, then give them a coworking space
Install each agent CLI under the ordinary account and authenticate it
directly with its own provider. Ronin Cowork then gives those terminal
sessions a Roster, roles, tasks, handoffs, persistent tmux sessions, and
one private browser workspace.
The agent-facing VM guide hands the remote agent directly to Ronin's
installation guide. When setup finishes, it returns the private Cowork
address for you to open from your computer or phone.
Start with the machine you have—or let an agent set one up.
Load Ronin offers the command, the existing-machine guide, and the VM setup guide.
