oss

tmux-ide is an open-source project by Prototyper.View source(opens in a new tab)

NewThe tmux-ide app in 2.9

The open-source workspace for coding agents.

Run and coordinate a team of coding agents, such as Claude Code and Codex, in the tmux sessions you already use. See what every agent is doing, and jump straight to the one that needs you.

Install method

tmux-ide app

macOS and glibc Linux · installs to ~/.local · no sudo

Requires Node.js 20 or newer

Copies Read https://tmux-ide.com/agents.md, set up tmux-ide for this project, and tell me what you did.

macOS native alpha →Docs
Animated tmux-ide app showing agent status, terminal panes, window controls, and Commands
Fig 1.Production app / sessions, agents, panes, and CommandsHow this demo is made →
One workspace, every agent accounted for

Named agents with live status.Go straight to the right pane.

Name each agent, watch its state, and open its exact pane.

01Name

Names you can talk about

Name an agent when you create it — Architect, Reviewer — or rename any pane from its menu. Unnamed agents show their harness, Claude Code or Codex; other new panes get a memorable fallback such as warm-redwood until a title or running program names them.

Coordinate a team of agents →
Fig 2.Agent identity / named agents and panes
02Monitor

Live agent indicators

See every agent's status in tmux: working, blocked, done or idle, in the sidebar and pane headers, so a multi-agent workspace stays readable without opening every terminal.

How agent status is detected →
Fig 3.Agent state / live indicators
03Navigate

Exact agent-to-pane navigation

Click an agent or choose it from the keyboard. tmux-ide resolves the exact session, window, and pane before transferring focus—no scanning a wall of terminals.

Fig 4.Agent routing / exact pane focus

Systems / Agent coordination

Claude Code, Codex and opencodein one agent team.

One agent per tmux pane, across harnesses. Coordinate them with tmux-ide send and wait, or use Claude Code agent teams for all-Claude teams; tmux-ide shows every agent and its status. It does not orchestrate them.

  1. a. Coordination. Turns your goal into focused tasks and hands them to the team.
  2. b. Goal ownership. Owns one goal and asks specialists to resolve specific questions.
  3. c. Parallel research. Each researcher works one bounded question in parallel.

Fig 5.A goal goes to an execution agent, which splits bounded questions across three researchers; findings and the finished outcome flow back up for review.

Reading the figure. Each frame is an agent process in its own pane, with the command it runs. Pairings and tasks are illustrative.

Systems / Runtime architecture

tmux owns the processes.tmux-ide adds the view.

Close the app or lose SSH; every agent keeps running in tmux.

  1. a. Terminal truth. tmux owns every process, pane and session, locally or over SSH.
  2. b. One authority. For the app and automation, the daemon is the only path to live tmux state.
  3. c. Many views. The app and automation act through the daemon; plain tmux clients still attach directly. Connect over SSH → Restore after a crash →

Fig 6.The daemon publishes tmux state to the app and automation clients and applies their commands back to tmux.

Reading the figure. Every action is an ordinary tmux operation you could also run from a tmux client.

Everything remains ordinary tmux

Windows, splits and resizing are real tmux operations.

The visual layer maps directly onto familiar tmux operations. Use it when it helps, then drop back to tmux whenever you want.

01

Create

Open clean windows and give agents, panes, and sessions names your team can remember.

  • new windows
  • named agents and panes
Fig 7.Create / windows and names
02

Arrange

Split and resize the workspace while the real tmux layout remains the source of truth.

  • split panes
  • pane resize
Fig 8.Arrange / splits and resize
03

Operate

Focus exact agent targets and close panes deliberately without disturbing the session.

  • precise focus
  • explicit close
Fig 9.Operate / focus and close
Every action remains inspectable from an ordinary tmux client.Explore all commands →
Questions, answered

tmux-ide FAQ

What is tmux-ide?

tmux-ide is an open-source (MIT) terminal app for running coding agents such as Claude Code and Codex inside ordinary tmux sessions. Home lists every agent across your local and SSH machines and shows which ones are working or need you. Terminals mirrors the live tmux session with clickable windows, pane headers and controls. tmux keeps owning processes, panes and persistence, so closing tmux-ide never stops an agent.

How do I install tmux-ide?

Run curl -fsSL https://tmux-ide.com/install.sh | sh, then tmux-ide app. The installer works on macOS and glibc Linux, needs no sudo, and bundles tmux 3.7c and its own Node.js runtime. With Node.js 20 or newer you can use npm install -g tmux-ide instead.

Is tmux-ide a new terminal multiplexer?

No. tmux remains responsible for processes, PTYs, sessions, windows, panes, and persistence. tmux-ide is a visual, agent-aware control surface for the tmux sessions you already own.

What happens when I close tmux-ide?

Your work keeps running in tmux. Open tmux-ide again, attach with an ordinary tmux client, or reconnect over SSH and continue from the same durable session.

Which coding agents does it recognize?

Claude Code, Codex and Aider, with detection tuned on their real output, plus conservative rules for 14 other agent CLIs. Any agent can report its own state with one tmux option, and ordinary shells keep working alongside them.

Does it work with Claude Code agent teams?

Yes. With teammateMode set to tmux, or auto when Claude Code runs inside tmux, each teammate opens in its own pane, and tmux-ide groups them under their team with their teammate names and live status. Claude Code still owns the team, tasks and messaging; in-process teammates have no pane of their own. See Claude Code agent teams.

Can I jump directly to a specific agent?

Yes. Selecting an agent resolves its exact tmux session, window, and pane, then focuses that pane. The names you give agents and panes make targets easy to identify; routing always uses the exact pane, never the name.

Is there a tmux sidebar for agents?

Yes. In the app, F10 shows or hides a sidebar that lists your machines, sessions and agents, with each agent's live status. In plain tmux, tmux-ide adopt <session> adds tmux chrome, and prefix b opens a sidebar pane listing your projects, sessions and windows with live status glyphs. See tmux chrome.

Does it work over SSH?

Yes. Install the same tmux-ide version on the remote machine and start its daemon there with tmux-ide --headless. Then run tmux-ide app --ssh <host> on your computer. If the connection drops, the remote tmux sessions and agents keep running. See remote machines for the setup.

Can it restore my sessions after the tmux server crashes?

Yes. While tmux-ide's background updater runs, which starts once you adopt a session or create one from the app, it snapshots your sessions about every 30 seconds. After a crash, tmux-ide restore --resume-agents rebuilds windows, layouts and directories and resumes supported agent conversations, such as Claude Code once tmux-ide integration install claude is set up.

Do I need a workspace configuration file?

No. tmux-ide discovers ordinary live tmux sessions by default. A .tmux-ide/workspace.yml file is optional when you want a repeatable declarative layout.

Ready when you are

Build your team of agents.

Install tmux-ide, open the app, and see every agent in the tmux sessions you already use.

Then run tmux-ide appRead the guide →