mmux vs tmux

What is tmux?

tmux describes itself as "a terminal multiplexer: it enables a number of terminals to be created, accessed, and controlled from a single screen", which "may be detached from a screen and continue running in the background, then later reattached" [1]. Every terminal lives in a pane, panes live in windows, and windows are grouped into sessions; commands are reached through a prefix key, Ctrl-b by default, and the server reads ~/.tmux.conf at startup [4].

The project started in 2007 [2], is written in C, is ISC licensed [3], and runs on OpenBSD, FreeBSD, NetBSD, Linux, macOS and Solaris [1]. Version 3.8, released on 2026-09-09, extended floating and modal panes, added built-in light and dark themes, and added monitor hooks that fire when a format changes [2] [7]. Control mode lets other programs drive tmux over a text protocol; iTerm2 uses it to show tmux panes as native windows [5].

GNU Screen is the older tool in the same category: it multiplexes one terminal between several processes, keeps a scrollback buffer per window, and keeps programs running while the session is detached [9]. Its latest release is 5.0.2 (July 2026) [10]. Everything on this page about tmux versus mmux applies to Screen in broad strokes, with one difference: mmux is built on tmux, so it needs tmux installed even if you use Screen for everything else.

Where tmux is better

• Layout freedom. tmux splits a window into as many panes as you like, tiled or floating, across as many windows as you like, so you can watch several programs side by side. mmux shows one program at a time in its main pane.
• It fits any workflow. tmux knows nothing about agents, projects or git, which is the point: the same tool works for a database console, a log tail on a server, or a pair session with several clients attached to one session [4].
• Configuration depth. Every key binding, option, hook and status line element is yours to change, and third-party plugins such as tmux-resurrect build on it [8]. mmux's keys are fixed; its YAML only describes agents, processes and workspaces.
• Scripting depth. tmux has a full command language and control mode, which terminals like iTerm2 build on [5]. mmux's CLI is narrower and aimed at listing, reading and prompting its own sessions.
• Platforms and footprint. tmux runs on the BSDs and Solaris as well as Linux and macOS, and binary packages exist for many of them [1]. mmux ships binaries for macOS on Apple Silicon and Linux, and itself depends on tmux.
• Maturity. tmux has been developed since 2007 [2]. mmux is a young project at version 0.13.

Where mmux is better

• It knows which agent needs you. A working agent shows a gray spinner, an agent that finished or waits on you turns green, and one blocked on a question shows a yellow ?. tmux can flag activity or silence in a window, but it has no notion of an agent's state.
• Desktop notifications, also over SSH. When a program rings the bell or emits a notification escape (Claude Code does when it is done), mmux raises a native desktop notification on the machine your terminal runs on.
• Sessions come back after a quit. After q, a crash or a restart-to-update, reopening the directory restores the agents and terminals you had open. Claude, Codex, Pi and Grok agents resume their conversation; terminals reopen in the directory you left them in as fresh shells.
• Dev processes as named rows. Processes from mmux.yaml start, stop and restart from the sidebar, can autostart, and can carry a stop: teardown such as docker compose down that runs on stop or quit.
• Git and worktrees built in. A git panel stages, commits (with an AI-suggested message), switches branches and pushes. w opens a git worktree as another project in the same sidebar with its own agents, and M merges it back.
• One session per directory, no setup. mmux in a project opens or reattaches that project's session, and mmux attach picks from every running session and recent directory. With tmux you name and track sessions yourself.
• Agents can drive it. mmux ask starts an agent, waits and prints its reply; mmux read returns a dev server's recent output. Each action shows up in your sidebar.

When to use which

Using tmux and mmux together

mmux already uses tmux underneath: each project directory gets an invisible tmux session with its status bar and prefix turned off. Those options are set on mmux's sessions only; the one server-level change is enabling extended-key support. You can keep tmux for everything else and use mmux for agent work:

1. Install mmux. tmux must be on your PATH, which it already is if you use it.
2. In any tmux pane, cd into a project and run mmux. It unsets TMUX when it attaches, so it works inside another tmux.
3. Both tools use Ctrl-b by default, and your outer tmux sees it first. Press Ctrl-b twice to pass it through to mmux (tmux's default send-prefix binding [6]), or give your outer tmux a different prefix.
4. Inside mmux, Ctrl-b b sends a literal Ctrl-b to the focused program, for a tmux running in an mmux terminal.
5. Put your agents in ~/.mmux/config.yaml and the project's processes in mmux.yaml. Closing the outer tmux pane only detaches mmux; run mmux again to reattach.

# ~/.mmux/config.yaml
agents:
  - name: Claude
    cmd: claude

# ./mmux.yaml
processes:
  - name: Dev server
    cmd: npm
    args: ["run", "dev"]