A VPS is the safer home for an agent you rely on: it stays on through power cuts and holidays, has a stable address, and someone else looks after the hardware. Your own computer makes more sense if you already have a spare machine, want the data physically in your building, or are experimenting and don't want another monthly bill.
What you're really choosing is who's responsible when it stops. On a VPS, the provider owns the hardware and you own everything above it. At home you own all of it, including the router that needs rebooting at nine on a Sunday evening.
Disclosure: BeLateral makes Able°, which gives you your own hosted AI agent, so weigh our view accordingly.
Side by side
| VPS | Own computer | |
|---|---|---|
| Cost | A few euros a month (Hetzner from €5.99 excluding VAT), plus model usage | Hardware you may already own, plus electricity, plus your time |
| Uptime | Redundant power and network, with a stated service level | As good as your broadband, wiring and fuse board |
| Control | Full control of the operating system, none of the hardware | Full control of everything, including the physical machine |
| Security | Public address, scanned constantly, yours to harden | Behind your router, but on the same network as everything else you own |
| Maintenance | Patching, updates, monitoring, backups | All of that, plus hardware, power and cooling |
| Best for | An agent that must be awake when you're not | Experiments, data that must stay on site, hardware you already have |
Prices checked on Hetzner's cloud pricing page on 24 October 2026, where the cheapest shared plan is €5.99 a month excluding VAT. The model bill is often larger than the server bill, so compare the total, and confirm the hosting figure before you commit.
What do you get with a VPS?
A virtual private server is a slice of a machine in a data centre: an operating system, a share of processor and memory, a network connection and usually a public address. It behaves like a computer you can only reach over the internet.
Its main strength is staying awake. Data centres have backup power, redundant network links and staff who notice when something fails at three in the morning. For an agent that sends a message at seven, that matters more than any specification.
It's also quick to change. If the agent needs more memory, you resize the instance and reboot; if the region is wrong, you rebuild somewhere else. And it's addressable: a webhook can reach it, your phone can connect from anywhere, and you don't depend on a home router that hands out a new internal address every time it restarts.
The drawbacks. It's someone else's computer. Your memory and conversation history sit on a hypervisor you don't control, in a country you picked from a dropdown, under a provider's terms. For most work that's fine, and for some it rules a VPS out.
It's exposed. A public address attracts automated scanning from the moment it exists, and login attempts start within hours and never stop. Configure it properly from day one: keys instead of passwords, a firewall, updates applied, and nothing running that doesn't need to.
And everything above the hypervisor is yours. The provider guarantees the hardware, not your application. If the disk fills or the process dies, nobody calls you.
What does running it at home involve?
Usually a small desktop, a Mac mini, a spare laptop or a mini PC on a shelf, plugged into your broadband and left running.
You get physical control and a one-off cost. The machine is in the room: you can unplug it, wipe it or lock it in a cupboard. If your work involves material you'd rather not store with a hosting company, keeping it on hardware you own is a legitimate reason to choose this route, and often the deciding one.
It's cheap if the hardware already exists. A machine that would otherwise sit in a drawer costs nothing extra, and a small desktop draws little power, so the electricity bill is rarely the number people imagine.
Local models are the other case. If you want to run a model on your own hardware rather than pay per token, you need a machine with a decent graphics card. Renting a server with a comparable GPU is expensive month after month, so over time the economics can flip.
The drawbacks. Reliability, mostly. A home machine inherits every weakness of a home: power cuts, broadband outages, a router that needs restarting, a cleaner who unplugs the wrong thing, a child who needs the socket for a charger.
Laptops sleep. A closed lid usually means a suspended machine, and a sleeping agent misses its schedule and answers nothing. You can disable sleep, and you should, but it's the kind of setting an update can reset without telling you.
Some home connections, particularly mobile broadband and some newer fibre services, sit behind carrier-grade NAT, where you share a public address with other customers and can't forward a port at all. If your agent only makes outbound connections, which is how most chat-based agents work, that doesn't matter. If it needs to receive a webhook, you'll need a tunnel service.
Finally, the machine shares a network with everything else in the building. A compromised agent can become a foothold for reaching your other devices, so give it the same caution you'd give any unfamiliar device on your network.
Which wins on reliability, security and cost?
Reliability: the VPS. Redundant power and several network paths beat one broadband line and one fuse. A home setup can run perfectly for months and then fail the week you're away, which is the failure that hurts.
Security: closer than it looks. A VPS is exposed to the internet by default and needs hardening on day one. A home machine sits behind a router that filters unsolicited inbound traffic, but it usually shares a network with personal devices, gets patched when somebody remembers, and may sit in a room other people walk through. Neither is safe by accident. The VPS is easier to secure well, because the practices are documented and the provider supplies the tools; the home machine is easier to leave insecure, because it feels private.
Cost over three years: depends what you own. The home machine wins if you already have it and usually loses if you buy one for the purpose. A VPS is a small recurring cost that never quite ends; a machine is a larger one-off cost plus electricity plus your time, and hardware eventually fails. Do the sums over the period you expect to use it.
Control and data location: your own computer. You own the hardware and know which country it's in without reading a terms of service. A VPS gives you the software and none of the physical layer, which is enough for most people.
Effort: the VPS, after the first evening. Hardening a public machine makes the start harder. After that, patching and monitoring are the whole job. At home the first evening is easier, but the ongoing work includes things you can't patch: hardware, power, and a router with opinions.
Recovery: the VPS, comfortably. A snapshot restores a machine in minutes. At home, a failed disk means a rebuild, and a failed disk without a backup means six months of agent memory is gone. Decide the backup before the first useful conversation.
Who should choose which?
Choose a VPS if the agent needs to be awake at a set time, you travel, you want a stable address, or you'd rather not own the hardware. It's also right if nobody in the building enjoys system administration and you're willing to pay a little for a managed layer on top. The hosted vs self-hosted guide covers where that line sits.
Choose your own computer if you have a spare machine, the data must stay physically with you, you want to run a model locally, or you're still experimenting and don't want a subscription. A mini desktop on a shelf with a wired connection and sleep disabled is a perfectly reasonable agent host, and plenty of people run exactly that.
Choose a hosted service if you don't want to patch anything, or the agent is for a team rather than just you. Self-hosting either way makes you the support desk. A hosted service costs more per month and less per hour of your attention, and for a small team that trade usually pays.
Use both if you want a cheap always-on host for the schedule and a local copy of the memory. It's more moving parts than most people need, and it's what a careful setup looks like when the data matters.
What should you sort out on the first evening?
The list is much the same on either route.
An operating system you can update. A current long-term-support release, minimal packages, automatic security updates on.
Access without passwords. SSH keys and a firewall that allows only what you need on a VPS. At home, a strong local account and remote access through a tunnel rather than an open port.
A user that isn't root. The agent shouldn't run as an administrator. If it's compromised, that's the difference between an incident and a rebuild.
The right time zone. Schedules fire in the machine's time zone. An agent that sends the Monday summary on Sunday evening usually has its time zone set wrong.
A backup you've restored. Configuring backups isn't enough: restore one deliberately, once, before you need to.
A way to know it's alive. A daily message, a heartbeat or an outside check. If the agent stops and nobody notices, a week of scheduled work simply doesn't happen.
A plan for updates. Monthly is fine for a personal agent. Machines that never get updated are how most self-hosted setups end up compromised.
The steps are identical whether the machine is in Frankfurt or your spare room. What changes is who you call when the backup restore fails.
Common questions
Is a VPS safe for personal data?
It can be, if you choose the region deliberately, encrypt the disk, keep it patched and read the provider's terms. It isn't private by default, because the provider can access the host. If that matters for your data, use a machine you own or a hosted service with a clear data policy.
Can I run an agent on a laptop that sleeps?
Not reliably. Sleep suspends the process, so scheduled tasks miss and conversations time out. You can disable sleep and run it lid-open or on external power, but updates and power settings tend to undo this, so a small desktop is a better fit.
Do I need a static IP address?
Only if something has to reach the agent from outside. Most chat-based agents connect outbound, so a changing home address is fine. For inbound connections, a tunnel service solves it without a business broadband contract.
What happens when my home internet goes down?
The agent stops answering and misses anything scheduled. It usually resumes when the connection returns, and whether missed runs catch up depends on the scheduler. Ask that question of any setup you rely on, hosted or not.
Can I move from one to the other later?
Yes, and it's worth planning for. Your configuration, memory files and skills move with you; the operating system, network and permissions need redoing. Keep a written note of what you installed and why, and the move takes an evening.