mmux vs CCManager

What is CCManager?

CCManager describes itself as a CLI application for managing multiple AI coding assistant sessions across git worktrees and projects [1]. It started as a Claude Code manager and now documents Claude Code, Gemini CLI, Codex CLI, Cursor Agent, Copilot CLI, Cline CLI, OpenCode, Kimi CLI and MiniMax Code, with a state detection strategy per tool that marks each session busy, waiting or idle [1]. It is written in TypeScript, MIT licensed, and installed with npm install -g ccmanager or run with npx ccmanager; the npm package pulls in a prebuilt binary for macOS (arm64, x64), Linux (arm64, x64) or Windows x64 [2]. The latest release at the time of writing is v4.4.6 from 2026-10-10 [8].

The interface is a menu: pick a worktree, work with its agent, press Ctrl+E to return to the menu [1]. From that menu you create, merge and delete worktrees; a multi-project mode (--multi-project with CCMANAGER_MULTI_PROJECT_ROOT) discovers every git repository under a root directory [6]. Settings live in ~/.config/ccmanager/config.json with a per-repository .ccmanager.json on top [1].

CCManager deliberately avoids tmux. Each agent runs in a pseudo-terminal owned by the CCManager process, so the README notes that closing CCManager kills the processes it started. It records the open sessions and offers to start them again on the next launch, which reruns each command in its worktree but does not bring back the conversation or the terminal output [1].

Where CCManager is better

• No tmux. CCManager needs nothing besides itself. mmux requires tmux on your PATH, which is one more dependency and one more thing to keep apart if you already have a tmux setup.
• Windows. CCManager ships a Windows x64 binary [2]. mmux runs on macOS and Linux only.
• Devcontainers. With --devc-up-command and --devc-exec-command, CCManager stays on the host and runs the agents inside a container, so a flag like --dangerously-skip-permissions only reaches the sandbox [5]. mmux has no equivalent built in.
• Conversation context in new worktrees. When it creates a worktree, CCManager can copy the Claude Code session data from ~/.claude/projects of the source checkout, so the new branch starts with the existing conversation [1]. A new mmux worktree starts its agents fresh.
• Status hooks. CCManager runs a command of your choice whenever a session changes between idle, busy and waiting, with the old and new state and the worktree in environment variables [3]. That makes it easy to post to Slack or log time per state. mmux notifications are triggered by the bell or a notification escape from the program, not by every state change.
• Wider list of detected agents. CCManager has state detection for Cursor Agent, Copilot CLI, Cline CLI, Kimi CLI and MiniMax Code, none of which mmux ships a preset for [1].
• Experimental auto approval. CCManager can ask a model (Claude Haiku by default, or your own command) whether a permission prompt is safe and approve it, falling back to manual approval on errors [4]. mmux leaves approval to each agent's own permission modes.

Where mmux is better

• Agents outlive the terminal. mmux runs inside an invisible tmux session per directory. Close the terminal window or drop SSH and every agent keeps working; run mmux again, or mmux attach from anywhere, to rejoin.
• Conversations come back after a quit. After a quit, a crash or a restart to update, reopening the directory resumes the conversations of Claude, Codex, Pi and Grok agents, not just their commands. Terminals reopen in the directory you left them in.
• More than agents. The sidebar also holds Terminals and Processes (dev servers, scripts) defined in mmux.yaml, with optional stop: teardown commands. Across worktrees only one checkout runs a given process at a time, so localhost:3000 is always the branch you last started it in.
• A git panel. A native panel shows changes, branches and commits, a live diff preview, commits with an AI-suggested message, scheduled commits and push. CCManager shows line counts and ahead/behind numbers per worktree.
• Scriptable from outside. mmux ls, mmux read, mmux ask, mmux wait and mmux start let a script or another agent list sessions, read a dev server's output, hand a prompt to a new agent and wait for its answer. CCManager's CLI accepts only startup flags and requires an interactive terminal [7].
• Worktrees that tidy up. A worktree opens as its own project box, with gitignored files such as .env copied at every depth and an optional setup: command. A finished worktree (clean, merged or pushed, idle) is removed after 30 minutes by default.
• Notifications over SSH without setup. A bell or notification escape from any session becomes a native desktop notification on the machine your terminal runs on, even when mmux runs on a remote box.

When to use which

Moving from CCManager to mmux

1. Install mmux and make sure tmux is on your PATH (Homebrew installs it for you).
2. Run mmux init in a project, or mmux init workspace in the parent folder to get several repositories in one sidebar, similar to CCManager's multi-project mode.
3. Move your command presets to agents: in ~/.mmux/config.yaml: each entry has a name, cmd and args.
4. Your .worktreeinclude entries map to worktrees.copy, and a post-creation worktree hook maps to worktrees.setup in mmux.yaml.
5. Press w to create a worktree, M to merge it back, X to remove it.

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

# ./mmux.yaml
worktrees:
  copy: [.env, .env.local]
  setup: pnpm install

To keep CCManager and still get persistence, open a terminal row in mmux and run npx ccmanager there. The row lives in mmux's tmux session, so CCManager and its agents keep running when you detach.