What Is The Agent Mac, And Why Run Agents On One?

The Agent Mac is an ordinary Mac, in my case a used 2019 iMac, configured to do one job: stay awake, boot itself back up after anything goes wrong, and run my AI agents around the clock without a person at the keyboard. It is not a product and not a server you rent. It is the exact machine under my agent fleet, with the power settings, the boot chain, the security posture, and the mistakes written down so you can clone it in an afternoon. You run agents on one because agents that are supposed to work on a schedule, respond to events, and finish long jobs overnight need a machine that is actually on at 3am, and your laptop is not that machine.

The one-sentence version

An always-on Mac, dedicated to running agents, that nobody uses for anything else. That is the whole idea. Everything else in the guide is detail in service of those three properties: always on, dedicated, unattended.

The reference machine is a 2019 27-inch Intel iMac with 72 GB of RAM and a 3 TB SSD. It runs macOS, it sits in the corner with the display asleep, and at the moment I captured the configuration it had been up for five days with the disk 1 percent used. It runs Claude Code agents, the scheduled jobs I give them, a Redis instance, and the bridge that lets me talk to those agents from my phone. It has no monthly bill beyond electricity.

Why "always on" is the whole point

Most business owners meet agents through a chat window. You type, it answers, you close the tab. That is useful, but it is not where the leverage is. The leverage is in agents that run without you: the morning status sweep that is waiting in your inbox when you wake up, the inbound lead that gets a response at 11pm on a Saturday, the long research job that starts when you go to bed and is finished when you get up.

Every one of those requires a machine that is awake when you are not. A laptop closes. A desktop you also use for email gets rebooted by a software update at the wrong moment. A cloud server works, but now you are paying rent, managing SSH keys, and you have lost access to the Apple apps your business already lives in. The Agent Mac is the answer to all three: a dedicated machine, in your building, that treats sleep as a bug.

What the machine actually does across a day, hour by hour, is laid out in what an always-on Agent Mac does, shift by shift.

What makes a Mac the right box

I get asked why not Windows, why not Linux, why not a cheap cloud VM. The honest answer has four parts.

1. launchd plus a Unix userland. macOS has a real process manager built in that starts jobs at boot, restarts them when they die, and fires them on a schedule. Combined with a normal shell, that is everything an unattended agent needs and nothing to install.
2. iMessage, Contacts, Calendar, Mail. If your business runs on Apple apps, a Mac is the only machine where an agent can touch them. Linux can do the automation but cannot see your texts. That access is granted through macOS privacy controls, and the guide covers exactly which grants you need and why some cannot be given remotely.
3. The same Claude Code binary you use at your desk. No translation layer. What works on your laptop works on the Agent Mac.
4. Used 2019 iMacs are cheap and silent. They sit at idle for years without complaint, and the RAM is upgradeable. A Mac mini M4 is the modern alternative: faster, sips power, same setup minus the RAM upgrade.

The trade-offs against renting a server are real and worth understanding before you buy anything. The Agent Mac vs a cloud server goes through what you give up in each direction.

What is actually in the configuration

The guide is the configuration, not a philosophy. The pieces are:

Power. A handful of pmset settings that turn off system sleep and disk sleep, let the display sleep after fifteen minutes, make the machine restart itself after a power failure, and disable Power Nap so nothing half-wakes. Each line is explained against Apple's default and why the default is wrong for this job.

The boot chain. What happens, in order, when power returns: the FileVault unlock, Redis coming up, a single node entry point launching, secrets pulled from a vault via a command-line password manager rather than sitting in plain text, the bridge that connects the agents to messaging, and the timer jobs that run on a schedule.

Security. Disk encrypted. Nothing listening on inbound ports. Full Disk Access granted to one stable binary, not to a dozen shell scripts. The agents reach out; nothing reaches in.

The rakes. Nine mistakes I already stepped on so you do not have to, from privacy grants that silently fail on bash wrappers to a package manager update that moved the node binary out from under the whole chain.

The clone checklist. Twenty-five steps from a fresh macOS install to a passing health check. The practical walkthrough of how to approach that is in how to set up an always-on Agent Mac for your business.

Who should own one

If you run a $5M to $50M business and you have started using agents for anything more than drafting, you should own one. The tell is the first time you want an agent to do something at a time you will not be at your desk. That moment arrives fast, and the laptop answer stops working the same week.

If you are earlier than that, still curious, not yet running anything on a schedule, the guide costs nothing and the machine can wait. Read it, understand what unattended operation requires, and buy the iMac the day you need it.

And to be clear about what this is and is not: the Agent Mac is the machine. It is not the operating system for the agents, which is Optimus OS, and it is not the agent you talk to in Telegram, which is Mako. Those sit on top. The Mac is the floor they stand on, and the floor is free to copy.

FAQ

Do I need a new Mac for this?

No. The reference machine is a used 2019 27-inch Intel iMac with the RAM upgraded. A Mac mini M4 or a current MacBook Pro also works and runs cooler and faster. What matters is that the machine is dedicated to the job, plugged into power and wired network, and never used as someone's daily driver.

Why not just run agents on my laptop?

Because your laptop closes, sleeps, travels, and reboots for updates whenever it likes. An agent that is supposed to run at 6am every day or respond to a lead at 11pm needs a machine that is awake at 6am and 11pm. The Agent Mac is that machine. Your laptop stays yours.

Is the Agent Mac the same thing as Optimus OS or Mako?

No. The Agent Mac is the hardware and the operating configuration underneath. Mako is the agent you talk to in Telegram, Orca orchestrates a school of agents, and both ship inside Optimus OS. The Mac is what they run on. The setup guide is free and stands on its own.

What does it cost to run?

Electricity, plus whatever the agents themselves cost in model usage. There is no server to rent and no monthly hosting bill. The machine sits silent at idle and draws real power only when the agents are working.