Which server should run your agents
The honest answer is: a small Linux server, and not your laptop. The reason is not power. It is the hours.
Agents that work while you sleep need a machine that is awake while you sleep. That one sentence decides almost everything about this choice, and it rules out the machine most people would reach for first.
Not a laptop
A laptop is shut at the hours your fleet is meant to be working. It sleeps, it changes network, it gets carried through an airport, and it is the machine you are using for something else. An agent on it is an agent that works when you are already at your desk, which is the one time you did not need it.
This is also why the comparison with driving a coding agent by hand comes out the way it does: until you want work done at an hour you are not awake for, the laptop wins.
A small Linux server is enough
A €5 VPS is enough to begin with. The work an agent does is mostly waiting on a model and on the services it reaches, not computing, so the machine is cheaper than people expect. Our own figure for what this costs you is about €5 to €20 a month, and the lower end of that is a real starting point rather than a teaser.
- Hetzner or DigitalOcean: the Mac app makes the server for you from one API token.
- Any other VPS: give the app its address and it checks the machine before it does anything to it.
- More than one machine: one subscription covers a fleet with fair use of five machines and ten agents.
What the machine has to carry
The hub, the agents, the vault that holds the logins you grant them, and the git history that is the record of what they did. All of it is on your side of the line. That is the good news and it is also the work: you own the server, the data, the credentials and the record, and so the backups, the disk and the bill are yours too.
Nobody can switch your fleet off, because nothing of ours is in its path — no licence check, no call home, no expiry. Our updates keep the software current; they do not keep your machine alive.
What else the setup needs
Tailscale, which is how the apps reach your machines without anything being exposed to the internet. Its own sign-in makes your Tailscale account as you go, and you paste one key, once. Note that Tailscale’s free plan is for personal use: business use needs their Standard plan at $8 a user a month, which is their price and not ours.
Then your own AI plan, signed in on each machine from the app, in your own browser. We never hold a model provider credential, which is the condition that makes this shape legitimate at all — and it is also why that session expiring is your job to renew.
Questions
Can I run it on a Mac mini at home?
It has to be a machine that is on, reachable through Tailscale, and not the machine you carry around. A Mac mini that stays on at home can be that. A laptop cannot.
Do I need to know Linux?
The Mac app does the install, checks the machine afterwards, and writes the fix for any check that failed. You need enough to own a server: to know it is yours, and that its backups are too.
What does the server cost?
About €5 to €20 a month, and it is your bill, not ours. A €5 VPS is enough to begin with.
The other guides
What makes an AI agent self-hosted
“Self-hosted” is said of setups that keep very different things on your side. Six questions show which parts of the work stay yours, and when a hosted platform is the better choice.
What always-on agents do to your AI plan
They can exhaust it. The limit is reached on your account, because it is your account they are signed in to. We cannot raise it and we do not pay it.
Deciding what an agent may reach
An agent you grant access to acts with that access. This is the part to think about before you grant anything, not after.
What it costs, what the agents actually do, and what you carry before you pay. Something wrong on this page? support@founderizer.com.