Claude Code on a Mac mini.

A Mac mini is small, quiet, and draws little power when idle. It can sit on a shelf with your repositories, tools, and logins already in place. A laptop sleeps when you close it and goes wherever you go. A mini stays put, so an agent can keep working on it while you're away from the desk. Once one mini is set up, adding a second one gives you more parallel work without filling your own machine with more terminals.

You set this up once, at the mini, before you rely on it from your phone:

1. Don't let it sleep. In System Settings › Energy, turn on Prevent automatic sleeping when the display is off and Wake for network access.
2. Bring it back after an outage. In the same pane, set Start up when power is connected to After power failure. On older macOS this is a toggle named Start up automatically after a power failure.
3. Use Ethernet if you can. A wired connection is the most dependable way to reach a mini that's on its own.
4. Turn on Remote Login and Screen Sharing (System Settings › General › Sharing) so you can still get in when something needs a hand.

Decide on FileVault versus automatic login. With FileVault on, macOS turns off automatic login. After a restart, the Mac waits at the login window until someone types the password, and your agents can't run until then. You can turn FileVault off to allow automatic login, but then anyone who can restart the Mac can get to your files. Since macOS 26 Tahoe there's a middle path. With Remote Login turned on beforehand, you can unlock a FileVault Mac over SSH after a restart, so you don't have to be in the room.

In Terminal on the mini, run Anthropic's installer:

curl -fsSL https://claude.ai/install.sh | bash

Then run claude in a project folder and use /login to sign in with your Claude account. To use Remote Control (below), you need a Claude Pro, Max, Team, or Enterprise plan. API keys don't work with it.

1. Claude Code Remote Control (built in, from Anthropic)

On the mini, run this in a project folder:

claude remote-control

The mini then shows up as a device card in the Code tab of the Claude app on your phone. Tap the card, pick a folder, and a session starts on the mini. You can also continue a session you started at the desk by running /remote-control inside it. Claude can send a push notification when a long task finishes or needs a decision.

It's free with your Claude plan and needs nothing else installed. The limits: it runs Claude Code only, and the claude remote-control process has to keep running on the mini. If you close it, the session goes offline until you start it again. If you started it over SSH, Anthropic suggests running it inside tmux so it outlives your connection.

2. An SSH app plus tmux, over Tailscale

This is the do-it-yourself route, and it works with any command-line agent:

1. Install Tailscale on the mini and the phone. The free Personal plan covers up to 6 users and unlimited devices. Your phone can then reach the mini from anywhere without opening ports on your router.
2. On your iPhone, install an SSH app such as Termius (free tier) or Blink Shell ($19.99 a year after a two-week trial). Both support Mosh, which holds a session together on a shaky mobile connection.
3. SSH in, start tmux, and run claude (or codex) inside it. Detach, put the phone away, and reattach later.

You get a real terminal and full control. The cost is that you're reading a terminal on a phone screen, with no notifications when the agent needs you unless you build them yourself.

3. Halfday by Late Sun Studio

Halfday is our app for iPhone, iPad, and Mac. It runs Claude Code, Codex, and OpenCode on your own computers. To add the mini, install Halfday on it and sign in. Halfday installs the agents it needs, and you don't type anything in Terminal. On a desktop Mac it asks to Stay awake during setup. Each computer's settings also have an Always On section with Stay Awake, Restart After Power Failure, and Keep Screen Unlocked.

Agents run in the background, so no terminal window has to stay open. From the phone you start a task in a sentence, typed or spoken. You answer the agent's questions, approve steps, and read what it changed. When a pull request is ready to land, you get one notification. If the mini needs a hand on its screen, you can control it from the iPhone too. If you have more than one mini, a MacBook, a Linux server, or a cloud computer, they all appear in one list, and you send the next task to whichever is free.

On a desktop Mac, Halfday’s Always On section also notes that after a restart, someone has to sign in at the Mac once.

Halfday is invite-only for now. The Mac app comes with an invite, and the iPhone app is in App Store review. Email support@latesun.studio for early access.

• One mini, Claude Code only, already on a Claude plan: start with Remote Control. It's free, built in, and from Anthropic.
• You like the terminal and want any agent: SSH plus tmux over Tailscale.
• Several computers, more than one agent (Claude Code and Codex), or a team sharing the minis: that's what Halfday is for.

These can live side by side on the same mini.

• The mini has to be on. Every option here runs on your own machine. If it's off, asleep, or offline, the work waits.
• Restarts need a plan. With FileVault on, someone has to unlock the Mac after a restart, either at the Mac or over SSH on macOS 26 and later.
• Your plan's usage limits still apply. A second mini gives you more places to work, not more Claude usage.
• Parallel sessions in one folder can collide. Give each task its own git worktree. Remote Control's --spawn worktree and Halfday both do this.

• Every way to use Claude Code from iPhone and iPad
• Halfday and Claude Code Remote Control compared
• What is Halfday?