Vibeleaderboard
← Course overviewModule 10 · Operate agents across machines · Beta
Lesson 10.2

Module contents

Operate agents across machines

Lesson 2 of 5

What does a terminal multiplexer actually keep alive?

When the treasurer connects to the rented computer from a phone, a terminal on the phone shows a shell on the host, as module 2 described. If that connection drops, what happens to the programs the shell started depends on how they were started: a program running in the foreground of a shell whose terminal disappears is often ended with it. This lesson introduces the program many practitioners use to avoid that, a terminal multiplexer, reads its own documentation for what it promises, and reads one practitioner's phone workflow against that documentation.

Numbered links identify sources. Underlined terms offer a definition without leaving the lesson.

A multiplexer is a program on the host

A is a program on the host that provides terminals to other programs itself, so that the person's own terminal can come and go. A commonly used one is called tmux, and its project wiki, updated in August 2026, describes its design. A single tmux server process runs on the host and manages the programs started inside it. Clients, meaning tmux client processes running in the person's actual terminal, attach to that server to see and type into those programs, and detach to leave them. The programs are organized into sessions, each holding windows, each holding panes, and the wiki describes detaching as leaving the session and any programs inside it running in the background. 1

Read that with module 2's words. The tmux server is a process on the host. Inside a pane it usually runs a shell, and the report harness, started from that shell, is a child process of it whose terminal is provided by tmux rather than by the treasurer's phone. When the phone's client detaches, or the connection drops, the harness process does not notice, because its terminal is still there. When the treasurer attaches again, from the same phone or a different device, the same session is shown, with what the harness has printed since, up to the amount of output tmux keeps for each pane. This course calls the pair : the client leaves, the programs stay, and a client can return.

The same Remote Control documentation recommends a multiplexer when keeping a terminal session alive after disconnecting from a remote machine. 2

Two things called a session

The word session now means two different things, and this module keeps them apart. Module 4's is the harness's record of one piece of work: messages, tool calls, observations, saved by the harness and resumable by it. A multiplexer session is a container for terminals on a host, managed by the multiplexer's server. The report harness's session lives in its process and in the harness's own storage; the tmux session holds the terminal that process is attached to. Detaching a tmux session saves nothing about the harness's work, since that is the harness's job, though the tmux server does keep its own state, including each pane's recent output, for as long as it runs. Closing a harness session does not end the tmux session it ran in.

What the multiplexer does not promise

The wiki says what tmux does when things end, and it is worth reading for what it leaves out. If the harness exits, the shell it was started from is still in the pane, showing a prompt, unless the harness was the pane's only program, in which case the pane closes. The server, by default, exits when it has no sessions left, and the wiki and the program's manual describe settings that change that. 1 The wiki says nothing about the host restarting, and it does not need to. The tmux server is an ordinary process, and module 2 established what happens to a process when its host restarts: it ends, along with everything in its memory, including the panes' kept output. On a host where nothing restarts it, and this case assumes nothing does, there is no server to attach to afterwards.

So a tmux session survives the person's connection dropping. It does not keep a program running that has stopped on its own, and it does not survive the host going down. The belief that a multiplexer guarantees recovery after a host failure confuses the connection failure it handles with the host failure it cannot. A multiplexer protects against the loss of the viewing device and its connection, and nothing else in the drawing. Process exit and host restart are handled, if at all, by what the process saved, which is module 8's checkpoint, and by what starts it again.

A practitioner's phone workflow, read against the documentation

In August 2026 Pieter Levels, a developer who runs several products alone, posted on X that he keeps one tmux session per project on a rented server and attaches to whichever project he is working on from his phone through a terminal application, adding that if his phone disconnects he reconnects and is back where he was. 3 The site's Intel entry reproduces the post; the original returned an access stub when this module was checked, so the wording is as the entry records it.

Read the post as a dated personal report, and read it against the wiki. "Back where I was" after a phone disconnection is the detaching the wiki describes: the client left and returned, and the programs ran on. The post is good evidence that this works for that person, on that server, with those tools. It says nothing about what happens when a program exits or the server restarts, because the post is not about that, and a reader who takes it as evidence that agent work on a server is durable has read a claim the post did not make.

Carry this forward

  • A terminal multiplexer is a program on the host whose server provides terminals to other programs. Clients attach and detach; detaching leaves the programs running. This protects against the viewing device and its connection going away.
  • A harness session is the harness's saved record of work; a multiplexer session is a container for terminals on a host. The second keeps the first's process alive across a disconnection and keeps only its own state, such as recent pane output.
  • If the harness exits, its shell usually remains in the pane. The server exits by default when it has no sessions, and it ends with the host. It does not keep a stopped program running and does not guarantee recovery after a host failure.
  • A practitioner's report of reconnecting from a phone to where they were is evidence of detach and attach working for them. It is not evidence about process exit or host restart, which it does not mention.

Lesson check

The report harness was started inside a tmux session on the rented computer, where nothing restarts programs automatically. Overnight the hosting company restarts the computer. In the morning the treasurer attaches to tmux from the phone. What does this lesson predict?

Choose an answer for each question.

This is a practice check. Answers reset when you leave this page. Results are not saved and do not place you in another module or award a qualification.

Detaching handles the tunnel. It does not handle a process that exits, a host that restarts, or a notification that never comes. The next lesson takes those cases one at a time.

Sources · Reviewed 2026-09-29
  1. 1. tmux project wiki (updated 14 August 2026): Getting Started
  2. 2. Anthropic, Claude Code documentation: Continue local sessions from any device with Remote Control
  3. 3. VibeLeaderboard Intel: Pieter Levels on X (24 August 2026), One tmux session per project: running coding agents from a phone

Sources support nearby explanations. Product details can change. Scenarios and model results are authored illustrations unless a lesson explicitly identifies a recorded execution. Module 7 identifies its offline code checks.