Flockdeck docs

Using Flockdeck

Getting started

Flockdeck runs your coding agents on hardware you control, your desk or a machine that never sleeps. With the optional Flockdeck Remote add-on you can use it from a paired phone, tablet or browser too. Every agent pane is a real program in a pseudo-terminal, so it behaves exactly as it does in a normal terminal: permission prompts, slash commands, plan mode, colours, mouse. Claude Code is what a pane runs unless you say otherwise; any other coding agent, and any model API spoken to directly, is a pick away.

With several agents running at once, Flockdeck answers one question at a glance: which agent needs you right now. The rest of this page covers what you need once there is more than one.

Three things to know first

The dot in each pane header says what that agent is doing. A filled cyan circle is working, an amber triangle is waiting on you, a hollow grey circle is idle, a red square means it failed and a hollow square that the process exited. Shapes as well as colours tell them apart. A tab holding a waiting agent is marked ▲, and so is the window title, so an agent that blocks while you are looking elsewhere still reaches you.

Panes are moved, not recreated. Drag a pane by its header to another edge, another tab, or a tab of its own; nothing restarts, and the conversation, working directory and scrollback come with it.

Ctrl+Shift+K opens the command palette, and so does Commands at the right of the top bar. It holds every action in the application, searchable, and goes to any tab or project by name. If you remember one shortcut, remember that one.

A first session

  1. Press Ctrl+Shift+T for an agent tab, or Ctrl+Shift+D to put a second agent beside the one you have. Both take the default agent; New agent tab (choose agent)… is the same thing with the picker in front of it, for when you want a different agent or a different model.
  2. Type into a pane as you would in any terminal. Click a pane to give it the keyboard.
  3. When an agent blocks on a permission prompt its dot turns amber and its tab is marked. Answer it, and it goes back to work.
  4. When one has produced something, press Ctrl+Shift+S to see the diff, commit and push it without leaving the window.

When one agent is not enough

Two agents editing the same files will fight. Give each its own checkout instead: Ctrl+Shift+G opens the worktrees, where you can create them, and Start a team starts a manager agent that starts the rest, each in a worktree of its own. Git worktrees and Agents and models cover both.

Which agent, which model

A pane's header names what is in it, such as claude · sonnet or codex · gpt-5.6-sol, beside its branch. Agents and models explains the picker, what changes when an agent cannot report its own status, and how to add one of your own. For the API agents, and for Claude Code where its status line is read, the header also estimates what the conversation has spent and shows how near a usage limit it is; Spend and limits says how, and when Claude Code's are read.

Finding the rest of it

The rail down the left of the window has a tile for each open project (one with an amber triangle badge has an agent waiting on you, and one with a cyan badge has agents at work), and under them the tools: Agents, Team, Broadcast, Changes, Resume and Worktrees, then, below a rule, Help and Settings. GitHub, API keys and Flockdeck Remote are in the command palette and in the settings. Rest the pointer on one, or reach it with the keyboard, and it says what it is and the key that does the same. On a phone or a narrow window the rail folds into the menu at the left of the top bar.

Most tools (all but Broadcast, Help and Settings) open as a sheet beside the rail, with your panes still in view. While one is open the keyboard stays in it, and a click on the panes puts it away; so does Esc, or the shortcut that opened it.

The first run opens this page once, unasked, and never again. F1 brings it back from anywhere.