VM1
Sold out
$ 13,455/ mo
- 1 vCPU
- 1 GB RAM
- 15 GB SSD
- 1 static public IP
- Unmetered transfer (fair use)
- Windows or Linux
- Physical location: Argentina
AI agent VPS
AI agent VPS: stop depending on your laptop
A VPS with a fixed Argentine IP to run your agent —OpenClaw, Hermes Agent or whichever you use— as one more system service. It keeps working after you close the laptop, and webhooks from Telegram or Discord always find it awake.
The short answer
You do not need to leave a computer on to keep an AI agent running: a Linux VPS with Ubuntu 24.04 and root access is enough. The agent is installed as a systemd service, comes back on its own after a reboot and keeps its configuration and memory on the server disk. At SIS Argentina the VM2 (2 vCPU, 2 GB) is enough for one agent; the model is consumed over API with your own key.
- OpenClaw
- Hermes Agent
- Docker
- systemd
- Node.js
- Python
- Telegram
- Discord
- Slack
- Ubuntu 24.04 LTS
What you get
Not tied to your machine
Your agent stops dying every time you close the laptop or the power goes out at home: it runs on a Buenos Aires server that is on all the time.
It comes back on its own
It runs as a systemd unit. If the server reboots, the agent comes back without you having to SSH in and start it.
Your key, your account
The model is consumed over API with the key you load, against the provider you choose. We never see it or manage it.
Docker for whatever you add
It is a VPS with root: you can run the agent in a container and add a database or a queue, each in its own.
Memory stays on disk
Configuration, history and skills live in the user’s home directory on the server, so you back them up like any other file.
Fixed Argentine IP
Webhooks from Telegram, Discord or your CRM always reach the same address, with no homemade tunnels dropping out.
AI agent VPS plans
These are the usual Cloud VPS plans, and the cart opens with the system already selected. Monthly prices, final price. USD amounts are the official store list; local invoicing is issued in pesos.
Most chosen
VM2
$ 26,910/ mo
- 2 vCPU
- 2 GB RAM
- 30 GB storage
- 1 static public IP
- Unmetered transfer (fair use)
- Windows or Linux
- Physical location: Argentina
VM3
$ 40,365/ mo
- 3 vCPU
- 3 GB RAM
- 45 GB storage
- 1 static public IP
- Unmetered transfer (fair use)
- Windows or Linux
- Physical location: Argentina
VM4
$ 53,820/ mo
- 4 vCPU
- 4 GB RAM
- 60 GB storage
- 1 static public IP
- Unmetered transfer (fair use)
- Windows or Linux
- Physical location: Argentina
Your agent running in three steps
-
Create the agent’s user
The gateway is not meant to run as root, and without lingering systemd stops the user’s services at logout:
adduser --disabled-password --comment "" juan && loginctl enable-linger juan -
Install the agent
OpenClaw’s official script brings Node.js and leaves the command ready, without starting the wizard yet:
su - juan -c 'curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard' -
Add the key and leave it as a service
The wizard asks for the API key on screen —you never type it as an argument, so it stays out of your shell history— and leaves the gateway as a systemd service that starts with the server. Check it with openclaw gateway status.
su - juan -c 'openclaw onboard --install-daemon'
Read on: Primeros pasos al recibir un VPS: la checklist de seguridad · Cuánta RAM necesita tu VPS
Frequently asked questions
Which plan do I need for an agent?
The VM2, with 2 vCPU and 2 GB, is enough for one agent and a light database. If you plan to run several agents, a headless browser or a vector database on the same server, the VM3 or VM4 give more room. You can move up a plan without migrating or reinstalling.
Does the model run on the VPS?
No. The agent runs on the VPS and the model is consumed over API with your key, against whichever provider you use. Large models do not run on a 2 GB CPU VPS: if what you want is to run your own open models, that calls for a hybrid cloud server with plenty of memory.
Which agents can I install?
Anything that runs on Linux. OpenClaw (MIT licence, on Node.js) and Nous Research’s Hermes Agent (MIT licence, on Python) install with their own official script and run headless. OpenClaw also publishes a container image on ghcr.io if you would rather start it with Docker.
How do I reach the agent’s dashboard?
OpenClaw’s gateway listens on loopback only by default, on port 18789, and requires authentication. From your machine you reach it over a VPN or an SSH tunnel: ssh -N -L 18789:127.0.0.1:18789 juan@203.0.113.24. There is no need to expose it to the internet.
What happens if the server reboots?
With lingering enabled, systemd starts that user’s services at boot and the agent comes back on its own, with its configuration and history intact on disk.
Is the server managed?
No. VPS plans and cloud servers are delivered unmanaged: we provide the server with a clean operating system, the static IP and root access, and from there on installing, configuring, updating, securing and backing it up are yours. The commands on this page are the ones in the tool’s official documentation, verified on Ubuntu 24.04, and they are enough to get it running; support for the tool itself is its own documentation and community. If you get stuck with something on the server, write to us and we will lend a hand. And if you would rather not deal with it, our software factory builds custom work on this same infrastructure. Shared hosting with a control panel is a different story: that one we do install and maintain.