mmux vs Claude Squad

What is Claude Squad?

Claude Squad is an open-source terminal app by smtg-ai that "manages multiple Claude Code, Codex, Gemini (and other local agents including Aider) in separate workspaces, allowing you to work on multiple tasks simultaneously" [1]. It is written in Go, installs as cs through Homebrew or an install script, and needs tmux and the GitHub CLI gh [1]. The latest release is v1.0.20 from August 2026, and the repository has about 8,600 GitHub stars [3].

Its model is simple: tmux gives every agent an isolated terminal session, git worktrees give every session its own branch, and a TUI lists the sessions [1]. You press n (or N to add a prompt) to start a session, ↵ to attach, ctrl-q to detach, and tab to switch between the preview and the diff of what the agent changed. s commits and pushes the branch to GitHub, c commits and pauses the session, and r resumes it [1]. Recent releases added a terminal tab per session [5], starting a session from an existing branch and named agent profiles [4].

Pausing removes the worktree directory but keeps the branch [8]. With the experimental -y flag, Claude Squad accepts Claude Code and Aider prompts on its own, and a background daemon keeps doing that after you quit [6][9].

Where Claude Squad is better

• Isolation by default. Every new session gets its own worktree and branch without a decision on your part, so two agents never edit the same checkout. In mmux an agent starts in the main checkout unless you cut a worktree first.
• Quitting is harmless. q in Claude Squad saves the session list and leaves every agent running in tmux [7]. In mmux, q ends all sessions (it asks first and offers to detach instead), and you rely on the restore when you reopen.
• Pause and resume. c commits the work, removes the worktree and keeps the branch; r brings the session back [1][8]. mmux can remove a worktree and keep an unmerged branch, but it has no paused state to resume from.
• Prompt answering that outlives the UI. The -y daemon keeps accepting Claude Code and Aider prompts while cs is closed [6]. mmux relies on the agent's own permission flags instead and does not answer prompts on its own.
• More prebuilt platforms. Releases include builds for Intel Macs and Windows [3]; its install script points Windows users to WSL for tmux [10]. mmux ships prebuilt binaries for Apple Silicon Macs and Linux only; on an Intel Mac, Homebrew builds it from source.
• Less to learn. One list, three tabs (preview, diff, terminal) and about a dozen keys. mmux has a sidebar, a git panel, workspaces and a control CLI, which is more to take in if all you want is parallel agents.

Where mmux is better

• Dev processes next to the agents. mmux has a Processes section for dev servers and scripts you start, stop and read logs from, with an optional stop: teardown command. Across worktrees only one checkout runs a given process at a time, so localhost:3000 is always the branch you last started it in.
• Conversations come back. After a quit, crash or restart, reopening the directory resumes Claude, Codex, Pi and Grok conversations, and terminals reopen in the directory you left them in. Claude Squad keeps the worktree after a reboot, but the agent starts a new conversation [8].
• Knowing which agent needs you. Sidebar rows show working (spinner), ready (green) and asking (yellow ?). A bell or notification escape from an agent becomes a native desktop notification, and that works over SSH because it travels as a terminal escape.
• A full git panel. Stage files, switch branches, browse up to 200 commits, revert or uncommit, commit with a message suggested by Claude or Codex, schedule a commit, and merge a worktree back into its base with M. Finished worktrees (merged or pushed, clean, idle) are cleared away after 30 minutes.
• Worktrees that run. New checkouts get the gitignored files they need (.env and friends, at every depth) and an optional one-time setup: command, configured under worktrees:.
• Agents can drive it. Any agent can list sessions, read a dev server's output, send prompts, wait for another agent, or cut a worktree with its own agent through the mmux CLI, and everything it starts shows up in your sidebar.
• Several projects in one window. A workspace: manifest puts multiple projects in one sidebar, and mmux runs in directories that are not git repositories.

When to use which

Moving from Claude Squad to mmux

1. Push or merge any Claude Squad branch you want to keep. Its sessions, worktrees and config stay under ~/.claude-squad/ and its own tmux sessions; mmux does not touch them.
2. Install mmux and set up the repository. mmux init offers the agent presets it finds on your PATH.
3. Pick a launch mode per agent with mmux agents (or a in the sidebar). If you ran cs -y, the closest match is the danger mode, which skips all approvals through the agent's own flag.
4. For Claude Squad's "one task, one branch" flow, press w and start an agent in the new worktree box, or do both in one command as below.
5. When you no longer need Claude Squad, cs reset removes its stored instances, tmux sessions and worktrees [6].

curl -fsSL https://mmux.org/install.sh | sh
cd ~/code/my-repo
mmux init
mmux
mmux worktree new fix-auth --agent claude --prompt "fix the login bug"