Best VPS for Running an AI Agent
Any VPS with 2 cores, 2 GB of RAM and 2 GB of storage on Ubuntu 24.04 will run a personal agent like OpenClaw. Hetzner, DigitalOcean and Vultr all have suitable plans from roughly $5 to $20 a month; Railway is the easiest if you want a one-click deploy and no server admin.
A personal AI agent is a light workload. The thinking happens in the LLM API, not on your server, so you do not need a powerful box. What you need is reliable uptime and enough headroom to run the agent process and its memory.
The good news is that almost any modern VPS clears the bar. The choice comes down to price, region and how much server admin you are willing to do.
How it works
OpenClaw runs comfortably on its stated minimum of roughly 2 cores, 2 GB RAM and 2 GB storage, with Ubuntu 24.04 recommended. See OpenClaw system requirements for the detail. Against that baseline, here is how the common options stack up:
- Hetzner Cloud. Best value in Europe; small instances start around $5 a month and comfortably exceed the minimum specs.
- DigitalOcean. Reliable, well-documented, easy to learn on; basic droplets from roughly $6.
- Vultr / Linode. Similar pricing and global regions; good if you want a server close to you.
- Railway. Not a raw VPS but a deploy platform with an OpenClaw one-click template. Costs a little more but removes server admin entirely; the fastest path if you do not want to touch a terminal.
- A spare machine at home. Zero hosting cost, full control, but you own uptime and networking.
Pick a region near you to keep latency low, give yourself a little RAM headroom above the minimum, and you are set.
Worked example
| Option | From | Specs vs minimum | Server admin |
|---|---|---|---|
| Hetzner Cloud | ~$5/mo | Exceeds | Yes |
| DigitalOcean | ~$6/mo | Exceeds | Yes |
| Vultr / Linode | ~$5–6/mo | Exceeds | Yes |
| Railway (one-click) | Higher | Managed | Minimal |
| Home server | $0 | Depends on hardware | Yes, plus uptime |
For most people the sensible default is a small Hetzner or DigitalOcean instance: cheap, reliable, and more than enough for one agent. If you would rather not manage a server at all, Railway’s one-click template is the shortcut. Either way, the VPS is the small part of the bill; the LLM API is the swing factor, as cost to self-host an AI agent explains. Once you have a host, how to self-host an AI agent walks through the deploy.
Try this in Liv
If picking and patching a VPS is not how you want to spend your time, Liv runs the agent on managed infrastructure so there is no server to choose at all.
- Start a 14-day free trial at app.liv4all.com, no credit card.
- Message Liv on Telegram, the default channel.
- Connect Gmail and Google Calendar via Google OAuth, revocable any time.
- Optionally link WhatsApp (invite-only, needs a dedicated eSIM).
Liv is early access with batched onboarding, so you may join a queue.
Common questions
Do I need a GPU?
No. The LLM runs in the provider’s cloud via API, so a basic CPU VPS is fine.
How much RAM is enough?
2 GB meets the minimum. 4 GB gives comfortable headroom for memory files and updates without much extra cost.
Which OS should I use?
Ubuntu 24.04 is recommended, but OpenClaw runs on any OS, including via Docker.
Is Railway a VPS?
Not exactly. It is a managed deploy platform with a one-click OpenClaw template, which removes most server admin in exchange for a slightly higher cost.
Can I run it at home instead?
Yes, on a spare laptop or mini PC. You save the hosting fee but take on uptime and networking yourself. See do I need a server for a personal AI agent.
Will a cheap VPS slow the agent down?
Rarely. The latency you notice comes from the LLM, not the server, so a $5 instance feels the same as a pricier one.