Some jobs take an hour: the full test suite, a video export, copying a photo archive to a new disk, re-indexing a database. You start it, and then you want to leave. The question in every thread about this is the same: how do I see how it's going without walking back to that screen?
There are three good answers, and they suit different jobs.
tmux plus SSH: the terminal, from anywhere
Start the job inside tmux (or screen) on the machine. Reach the machine over Tailscale or your VPN, SSH in from your phone with an app like Termius or Blink, and run tmux attach. You see exactly what you'd see at the desk.
Good for: anything you started by hand in a terminal, and anyone comfortable with a terminal on a phone keyboard. Watch out for: the job dies if you started it outside tmux and the terminal closes. And a 6-inch screen full of compiler output isn't a summary.
Claude Code's Remote Control: the same coding session, on another device
If the long job is a Claude Code session, Claude Code has this built in. Run claude remote-control on the machine (or type /remote-control inside a session) and pick the session up from claude.ai/code or the Claude app. The work keeps running on your computer; the phone is a second window onto the same conversation. It needs a paid Claude plan and a claude.ai login.
Good for: an interactive coding session you want to steer from the couch. Watch out for: it's one session on one machine. It doesn't help with a backup script or a job on a second computer.
Ask your AI chat to start it, then ask again later
The third pattern starts the job from the chat itself. You ask ChatGPT or Claude to run the job on a named computer; the job keeps running on that computer after the chat answers; later you come back to the same conversation and ask how it's going. Your chats sync between desktop and phone, so "later" can be from the train.
I build Local Commander, which works this way, so I'm biased. A real exchange looks like this:
On Home PC, copy D:\Photos to the new disk E:\Photos with robocopy, keeping timestamps. Log to E:\copy.log. Tell me when it's started; don't delete anything on D.
The command starts and the chat answers within seconds with "started". An hour later, from the phone, in the same chat:
How's the copy going? Show me the last lines of the log and whether it finished with errors.
The AI asks the computer for the job's status and output and gives you a one-paragraph answer instead of a wall of log. If something is wrong: "Stop the copy." It sends the cancel to that computer.
Good for: jobs that aren't a coding session — copies, backups, test runs, exports — and machines other than the one in front of you. Watch out for: it reports output and status, not the screen. And as with the other two, the computer has to stay awake.
Which one when
| Job | Best fit |
|---|---|
| Something you started by hand in a terminal | tmux + SSH |
| A Claude Code session you want to keep steering | Remote Control |
| A copy, backup, export or test run you'd rather describe than type | Ask the chat to start it |
| Several jobs on several computers | Ask the chat; name each computer |
The part everyone forgets: sleep
Whatever you pick, a laptop that sleeps when you leave stops the job. On Windows, set "When plugged in, put my device to sleep after" to Never for the duration, or run powercfg /change standby-timeout-ac 0 and set it back later. On macOS, caffeinate -i in front of the command keeps the Mac awake until it finishes. On Linux, systemd-inhibit does the same.
Setting up the chat way
- On the computer, run
npx local-commander-agent@latest(Node.js 20+) and approve it in the browser. - Choose its access: the whole user account, or only the folders the job needs.
- Add
https://localcmd.app/mcpas a connector in ChatGPT or Claude and sign in.
It isn't a sandbox: an approved computer lets the AI do what your account can do, so for a long unattended job, give it only the folders it needs. You can cancel the job or cut the computer off at any time.
Steps and examples: localcmd.app/control-multiple-computers-with-ai