A map of the screen
The three columns, the board, and how tabs move
The app is made of three columns, left to right.
| Column | What sits there |
|---|---|
| Left | The list of projects, and the board (the state of each session) |
| Middle | Tabs, and what is inside them (terminal, chat) |
| Right | The files of the workspace you have open |
The order is meant to let you watch what is running, give an instruction, and read the result without switching screens.
Just after launch

The list on the left holds your projects (registered directories). Nothing is open yet, so the middle and the right stay empty and wait.
Create a worktree
Click + beside a Git project to open the creation dialog with a suggested name. Names come from French, Italian, and Spanish number words. Click the dice button for another suggestion, or type your own name, then click Create.

If no suggestions are available, enter a name. If a suggested name is taken before you create the worktree, choose another name and try again.
Bring .env and other ignored files
A worktree only contains files tracked by git. To also get files that git ignores, such as .env,
put a .worktreeinclude file at the project root and list them in .gitignore syntax. When a worktree
is created, buster7 copies only the listed files that git ignores from the original folder
(the same convention as Claude Code).
.env
config/secrets/Existing files are never overwritten. Copying stops at 256 MiB or 10,000 files. The number of copied files is shown after the worktree is created.
Run a command after creating a worktree
If you want a command such as pnpm install to run in every new worktree, open Settings → Projects,
choose the project, and save it under "Command to run after creating a worktree". When you start an agent
in a new worktree, the command runs as a terminal session named "setup" and its tab opens too (the agent's
tab stays in front), so you can watch its progress and errors.
The command is never read from files inside the repository; only what you set here runs.
The board
On desktop, use Projects and Board at the top left to switch views. Each view keeps its own width. Switching preserves your search, expanded items, and scroll position without changing the terminal in the center.

All projects appear in the same order as the project list. Cards show a sprite, name, workspace, and state. Hover to see the project, working directory, launch command, and other available details. Unassigned sessions appear in a separate group.
- Waiting — the parent is not working and no observed background task is running
- Working — the agent is working. Claude Code and Codex follow the state the agent reports, so typing alone does not make a session Working. A session is also Working while a subagent or background command is still running, even if the parent is waiting (only observed tasks count)
- Needs attention — an action needs approval or a question needs an answer. The card shows “Needs approval” or “Awaiting answer”. This column takes priority even while background tasks run. Currently supported for Claude Code. The label clears when you answer or press a key in that terminal
A card with a change you have not seen yet (work finished, waiting for approval, or exited) shows a dot at its right edge. It clears when you open it or press a key in that terminal, and your Mac and phone stay in sync. The number of unseen cards appears at the top of the sidebar and on the macOS Dock icon. Click the number or use Go → Next Needing Attention (⌘J) to open them, approvals and questions first, then oldest first. On a phone, use the numbered button in the top bar. Finished work is detected for Claude Code and Codex. An ended process appears in the Waiting column with an ended label.
Use the + beside a project to create a session. Check or select Workspace in the menu, then choose an agent. The view switches to that workspace and opens the new session in its tabs.
Each project keeps three columns: Waiting, Working, and Needs attention. Cards move between columns as their state changes, and empty columns stay visible.
Press a card to switch to the session’s workspace and open its tab in the middle. If the tab is already open, it is selected. Each workspace keeps its own tabs. An ended process does not mean the task was completed. The parent's selected LLM, objective, and current task are not collected automatically.
When a supported agent starts a subagent, it appears beneath its parent on the board and in the project tree. Hover over a subagent to see its reported state and model, when available. Click it to open the parent's terminal. Subagents follow their parent with a short delay when it moves between columns.
Completed, interrupted, and failed states come from the agent's reports. Subagents whose information is unavailable may not appear.
Tabs

Tabs line up along the top of the middle column. One tab is one session. A session can hold more than one pane (terminal, chat), so you can watch the same partner from both.
Closing a tab also stops that session. A tab that vanishes while the work runs on behind it would leave what you see out of step with what is running. So that you do not stop a session by mistake, pressing the X asks for confirmation first.
See what an agent was doing

In an agent's terminal, the bar at the top shows what it is doing in one line: its title and the task in progress. Click it to open the goal, the first and latest requests, and the task list. Hover over a tab to read the same without switching.
The summary comes from the agent's conversation record. When the agent keeps no
task list, only the title and requests appear. In Claude Code, a goal set with
/goal appears too.
Right-click to act
On a tab you can split or close; on a session you can open it, copy its name, or stop it. On a project you can pin it or add a workspace; on a workspace you can copy its path; on a file you can open it, expand a folder, or copy a path or name. Stopping a session and closing a tab both ask for confirmation.
Maximize the window
On a Mac, double-click the title bar or the empty space beside it to maximize. Double-click again to go back to the previous size. Double-clicking an interactive control does not maximize.
Fold the panes away
The board on the left and the files on the right can be folded away with the toolbar buttons, which gives the middle column more room. To give the terminal more space see Terminal; to look at files see Files.
On a phone or a tablet
On touch devices (iPhone / iPad, Android) the skeleton of the screen changes. Instead of squeezing in three columns, the board moves into a drawer on the left, and a row of helper keys (Esc, Ctrl and so on) sits below the terminal. When the keyboard is up, the helper keys sit right above it.
When nothing is open, the screen shows the board. Tap a card to open its session. The flagship opens from a button in the top bar.
On narrow screens, the board shows Needs attention, Working, and Idle in that order, skipping empty states to keep the layout compact. Pull down to refresh. Open Settings from the bottom of the drawer.

No server runs on the phone. It connects to a server running on your Mac or another machine, over the same Wi-Fi or a tunnel such as a VPN. The first time you open the app, it asks for the server address and token, and tests the connection before saving. If it cannot connect, it tells you whether to check the address, the token or the network.
The skeleton follows the device, not the window
Narrowing a macOS window does not turn it into the mobile skeleton. When the window is narrow, only the panes that no longer fit fold away.
macOS menu bar
Use File to add a project or find files, View to switch the board and panels, Terminal to start a shell or agent and split panes, and Go to navigate history and tabs.
Open buster7 → Settings… (⌘,) for settings or Help → Quick Guide for keyboard shortcuts. Focus a text field or terminal to use the editing commands. Unavailable commands are dimmed. Workspace commands are disabled while settings or a dialog is open. Closing a running terminal tab shows the same confirmation as its close button.

Choose a harness

Select a workspace and press + to launch Claude Code, Codex, Gemini CLI, OpenCode, GitHub Copilot CLI, Cursor Agent, Aider, Cline CLI, Goose, Amp, Qwen Code, or Antigravity (agy). On macOS, these are also in the Terminal menu.
Install the CLI on the connected computer and complete its login or API key setup first. The additional harnesses run in the terminal; dedicated chat view and account switching are not supported. Enabling YOLO passes the CLI's option for skipping approval prompts.
Goose and Amp use their own approval settings. Antigravity receives the selected working directory.