One worktree per branch. One tmux session per worktree.

kwt is a Git worktree manager with two surfaces. The dashboard is for you: see and manage the worktrees across all your projects from one keyboard-driven view. The CLI is for your coding agents: plain commands and stable JSON for creating worktrees and working inside their tmux sessions.

Install kwt Quickstart

The dashboard, for you¶

go install go.kenn.io/kwt/cmd/kwt@latest
kwt

Run kwt inside a repository and it registers the project, discovers its worktrees, and opens a full-screen dashboard. From one view you can:

• Attach to a worktree's tmux session (enter), creating it when needed.
• Create a branch and worktree (n), or search local and remote branches for one (b).
• See per-branch Git state — dirty, ahead, behind — across every project.
• Delete worktrees (d), kill live sessions (K), and switch the active project (P).

A plain directory can be a workspace too: run kwt there and enter opens a tmux session in place. The quickstart walks through the full key map.

The CLI, for your agents¶

Every dashboard operation is also a plain command, so coding agents and scripts can manage worktrees and their bound tmux sessions without a UI:

kwt add -b feature/new-ui              # isolated checkout + tmux workspace
kwt exec feature/new-ui -- make test   # run a command in the worktree
cd "$(kwt get feature/new-ui)"         # resolve a worktree path
kwt list --json                        # machine-readable worktree state
kwt remove -b feature/new-ui           # delete the worktree and its branch

kwt pr import checks a GitHub pull request out into a fresh worktree, and kwt status shows which branches are dirty, ahead, or behind — useful when several agents are working in parallel. Configurable layouts define the tmux panes each workspace starts with, from a single shell to a preset arrangement of agents.

Guardrails¶

Checkouts of existing or remote branches are created inert: kwt skips repository hooks, setup commands, and workspace launch until you review the files and explicitly open the workspace. Pull-request imports use a protected session boundary and preserve exact push routing. Multi-machine sync stays off until you configure it.