Skip to content

Fleet view

The fleet view shows every agent you run, on every host, in one list. Use it to find the agents that wait for you, then jump straight into the right tab.

Fleet view with agents grouped by status, waiting agents first

  • Click the summary line under the mocante name at the top of the sidebar (for example 4 agents · 1 waiting · 0 errors).
  • Or open the command palette (⌘K) and pick Fleet overview.

The line under the Fleet heading counts your agents, projects, waiting agents and errors. It also shows how many agents are offline and the total token count, when there are any.

Each row is one agent tab. A session with no agent shows its first tab. A row shows:

  • a status dot;
  • the workspace and session, and the tab title for an agent that was detected in a shell;
  • the project, when you group by status;
  • what the agent is doing: its pending prompt when it needs input, the error line when it failed, otherwise its current task or last line of output;
  • the token count, the runtime and the host.

Click a row to open that tab. Right-click a row for the tab’s actions.

Use the Status and Project switch at the top right of the view. The palette has the same choice as Group fleet by status and Group fleet by project.

  • Status puts the agents that need you first: Needs input, then Errors, Working, Completed and Idle. The first two sections always show, so an empty Needs input tells you that nothing waits on you.
  • Project makes one section per project, in sidebar order, with the agents sorted by status inside it. Agents outside any project go under No project. Click a project name to open the project.

The statuses are explained on the agents page.

The fleet view shows a warning line, with a View hosts button, when:

  • a host is unreachable. Its agents drop out of the list and are counted as offline. See remote hosts.
  • the agent hooks on a host come from an older mocante. Status can lag until you update them. See keep hooks up to date.

Running dev servers also show on the fleet view.

mocante counts the agent tabs that need input and shows the count in three places:

  • The sidebar header shows the totals for all projects. Click it to open the fleet view.
  • Each project header in the sidebar shows a badge with the number of waiting agents in that project.
  • The dock icon of the desktop app carries the total.

On a Mac, the desktop app adds a mocante item to the menu bar, with the number of waiting agents beside it. Click it to see each waiting agent, and pick one to go to its tab from any app. Right-click the dock icon for the same list.

The menu bar item also has Open mocante and Quit. A list longer than twenty agents ends in a row that counts the rest and opens the app.

To hide the item, click Show in Menu Bar in its menu, turn off Show in menu bar in Settings › Appearance, or run Hide from menu bar from the command palette. To show it again, use the setting or Show in menu bar in the command palette. The dock menu and the count on the dock icon stay when the item is hidden.

The list comes from the app’s window. If you close the last main window, the count clears and the menus offer Open mocante to see waiting agents until you open a window again, since agents may still wait.

When an agent starts to wait for you and its tab is not on screen, mocante shows a desktop notification, Agent needs input, with the agent’s task or last line. Click it to open the tab.

mocante also notifies you when:

  • a program rings the terminal bell in a tab you are not looking at (Bell in the tab name);
  • an agent shares a file with you;
  • an agent finishes its turn (Agent finished) or stops on an error (Agent stopped on an error), if you turn on Notify when an agent finishes in Settings, under Notifications, or with the palette command Turn finish notifications on. It is off at first, since agents finish often.

To play a short tone with each notification, turn on Notification sound in Settings, under Notifications. The palette command Turn notification sound on does the same.

On a Mac with macOS 27, turn on Summarize with the on-device model in Settings, under Notifications, to have Agent needs input say what the agent wants in one short line, and Agent finished or Agent stopped on an error say what it did or what went wrong. The line comes from the fm command, which runs on the model on your Mac, so the terminal text stays on your Mac. The notification appears at once with the agent’s own text, and changes if a line arrives within a few seconds, with no second sound. The banner may show again when it changes. Treat the line as a hint: it is written from terminal output, which a program can fill with anything.

In the web build, notifications come from the browser, so allow them when the browser asks.

When Claude Code or Codex asks to run a tool and the tab is not on screen, the desktop app shows a notification such as claude code asks to run Bash, with the command or path underneath. It has an Approve button, with Deny behind it in the button’s menu.

  • Approve or Deny answers the agent. You do not need to open the terminal.
  • Click the notification body to open the tab. The agent then shows its own prompt in the terminal.
  • Close the notification to hand the question back to the agent’s own prompt.

If you go to the tab, or the tab is already on screen, the agent shows its own prompt as usual. The notification is a shortcut, not a gate. A request that nobody answers goes back to the agent after about five minutes.

Inside the app, a request also shows on the tab’s row in the fleet view, with the command and Approve and Deny buttons. If it comes from another tab of the session you are looking at, the same buttons show above the terminal. Its right-click menu and the palette also offer Approve permission and Deny permission. A request to read 1Password secrets is approved only in its session, which lists every reference; the fleet row can only deny it.

In Claude Code’s auto mode, the classifier decides each call, and Claude Code does not take an approval from a notification. mocante does not ask you then. If the classifier does not approve the call and Claude Code shows its own prompt, the tab shows needs input a few seconds later and a notification tells you to answer in the terminal.

You do not need the fleet view to reach the next agent that needs you:

Keys (macOS) Keys (Linux) Action
⌘⌥⇧↓ Ctrl+Alt+Shift+↓ Next session that needs you
⌘⌥⇧↑ Ctrl+Alt+Shift+↑ Previous session that needs you
⌘K Ctrl+K Command palette, waiting agents first