A hosting company runs the physical server and divides it into separate virtual machines using virtualisation, a long-established technique used by almost every hosting provider.

A block of flats is a useful picture: one building, with a separate front door, kitchen and set of keys for each flat.

Why do AI agents run on a VPS?

Mainly because a VPS stays on. An AI agent running on your laptop stops when you close the lid. One on a VPS keeps running, fires its scheduled tasks on time, and replies when you message it at six in the morning from a train.

Location is the other reason. A VPS sits in a specific data centre, such as Frankfurt or Singapore, and where your conversation history and memory are stored can matter under GDPR or your local data-protection law. Our guide on what happens to your data goes into this.

An example

You want a planning message every weekday at 7:30 London time. On your own computer, that only happens if the machine is awake and online at 7:30. On a VPS, the message arrives whether you're at your desk, on a train or asleep.

The trade-off is upkeep. Someone has to install security updates, take backups and keep access locked down, and that's either you or a provider you pay. Our comparison of a VPS versus your own computer sets out the costs of each.

What's the difference between a VPS and shared hosting?

Shared hosting puts many customers' websites in one environment with less separation between them. A VPS gives each customer a separate virtual machine with its own resources. For a simple website, shared hosting is often enough. For an agent that holds conversation history, memory and credentials such as API keys, the extra isolation is worth having.

Common questions

Do I need my own VPS to use an AI agent?

No. A hosted agent service runs the server for you. You'd only rent one yourself if you were self-hosting an open-source agent and wanted it available around the clock.

Is a VPS secure?

It can be, once it's configured properly and kept patched, but a default install isn't. Most incidents come from outdated software, weak remote-login settings or a leaked key rather than from the virtualisation itself.

How powerful does the server need to be?

For a personal agent, not very. The AI model runs on the model provider's machines, so the server mainly has to stay awake, store data and make network requests. A small plan is usually enough, but check the stated requirements of the agent software you plan to run.