The moment you want an AI agent to do real work, not just answer questions, you hit the same wall everyone hits: where does it live? The default advice is to rent a small Linux box, install your tools, wire up cron, and let it run. It works. It is also more machine than most people need, and it puts your agent somewhere you do not actually control.
There is a simpler answer sitting on your desk.
The server tax nobody mentions
A rented VPS is not just a monthly line item. It is a second computer you now own the upkeep of:
- A Linux box to patch and keep secure.
- Cron jobs and shell scripts to hand-roll for anything recurring.
- Credentials and access tokens living on a machine outside your home.
- A bill that arrives whether the agent did anything useful that month or not.
None of that is hard on its own. Together it is a day of setup and a slow drip of maintenance, all to keep an agent alive. And the whole point of an agent was to save you that kind of work.
A Mac you own beats a server you rent
If you want an agent that is always there, the requirement is modest: a machine that stays on and runs your tools. A cheap Mac mini left on in a corner does this beautifully. It is already yours, it already runs the apps your life lives in, and it does not send you an invoice.
The best home for an agent is the computer you already trust with your mail, your calendar and your files.
Running locally changes the security story too. Your data does not get copied to someone else’s server to be processed. The work happens on your machine, and the only thing that leaves is whatever your AI model needs to think, exactly as it does today when you use it by hand.
What still needs solving
Local is not automatically safe. An agent with a free hand on your real inbox can still do damage, especially if a booby-trapped email tries to redirect it. So the home is only half the answer. The other half is a boundary:
- Effect awareness. The agent should know which actions only read, which create, and which are destructive.
- Approval on anything irreversible. Sending, deleting and editing should wait for a human, and the agent should never be able to approve its own action.
- A log you can read back. Every step, written down, so you can see exactly what ran while you were away.
That is the part Alfos handles. It gives the AI coding agent you already run a set of safe, gated, logged actions across your Mac apps, on demand or on a schedule. You bring the model. Alfos is the hands you can trust.
A Mac you own is the right place for an agent. A clear boundary is what makes leaving it running there a good idea instead of a scary one.