fleet-deck

Fleet Deck is a dashboard for people who run many Claude Code coding sessions on one machine at the same time. If you have ever opened a session, then another in a separate worktree, then a third just to check something, and lost track of which terminal tab is doing what, this tool is built for exactly that moment. It gathers every session on your machine onto one local web page and shows each one with a nickname, a status such as working, verifying, needs you, or idle, and a small mailbox for messages.

The board watches for sessions that are about to edit the same file and warns them before they collide, even across separate worktrees of the same repository. When a session hits a permission prompt or asks you a question, that question shows up as a card on the board instead of sitting quietly in a terminal you forgot about, so you can answer it from anywhere. You can also message a single session, a whole repository, or every session at once.

Fleet Deck can start new Claude Code sessions for you inside its own terminal windows, let you click into a live terminal view of any session and type directly into it, or open a grid showing every running session at once, with only one accepting keystrokes at a time so you never send a command to the wrong agent. If the machine restarts, the underlying work and history survive on disk, and a session can be resumed with a single click. There is also a mode for driving a session remotely from a phone, and a feature that captures plans produced by a session in planning mode into a small library you can revisit and run later.

The tool deliberately avoids making its own decisions with an AI model. Picking which idle session should take a new task is handled with a plain database lookup rather than a model call, which keeps your token usage in your control.

It installs as a Claude Code plugin with two commands, requires Node.js version 22.5 or newer, and needs tmux 3.4 or newer if you want it to spawn workers or open terminals in the browser. It runs on Linux, WSL2, and macOS. The project is written in JavaScript and released under the MIT license. The full README is longer than what was shown.