Run a fleet of agents, side by side.
Cockpit is a desktop command center for Pi. Open your projects and run as many AI agents and real terminals as you want — each in its own folder, all in one window. It runs on your machine, with your files.
Requires the Pi CLI — guided setup on first launch.
A multiplexer for agents and terminals.
Not a chat next to an editor — a workspace where shells and AI live together, on your own machine.
Many agents, one window
Split the canvas into as many panes as you want — agents, terminals, and files, side by side or stacked. One agent streaming never blocks the others, and your whole layout comes back exactly as you left it.
A composer that keeps up
Paste or drag images, switch the model, dial reasoning from off to high, type @ to reference files and / to fire the agent's commands. Every choice is remembered per tab.
Real terminals
Open a genuine shell in any pane — your default shell on macOS and Linux, PowerShell or cmd on Windows. Full color, resize, copy/paste, and tab titles that track what's running.
An agent-aware file tree
Browse the project on demand with per-type icons. Right-click any folder to spawn an agent or terminal right there, or drag a file straight into the conversation.
Read what the agent touches
Open Markdown, syntax-highlighted code, images, and SVG inline — hundreds of languages. The viewer reads; the agent does the editing.
Sessions you can resume
Every agent is a live Pi session in the folder you chose. Watch text, thinking, and tool calls stream in — and reopen a past conversation right where it stopped.
Native notifications
When an agent finishes a turn and the window is in the background, your OS lets you know. The sidebar also counts agents with new replies, per workspace.
Light or dark, your call
System, light, or dark theme; careful interface and code fonts; syntax palettes; and zoom for the whole UI when you need it bigger.
Multi-project workspaces
Open any folder as a workspace and switch in a click without losing state — agents keep running in the background. Each shows its git branch and whether there are uncommitted changes.
A new branch, fully set up, in one click.
Fork your project onto a fresh branch without breaking your flow. Cockpit uses git worktrees under the hood — and recreates your exact pane-and-tab layout pointed at the new folder.
One click to branch off
Spin up a worktree on a new branch and Cockpit rebuilds the same agents, terminals, and tabs — already aimed at the new directory, so you keep working instead of setting up.
Clean removal, no surprises
Done with it? Removing a worktree warns you first if the branch isn't merged yet, then tidies up both the folder and the branch for you.
Agents that outlive the window.
The same always-on layer behind Remote Pi — now with a desk. Keep agents working around the clock and put them on a schedule, managed right in the app.
24/7 daemons
Promote work to background daemons that run nonstop — create, start, stop, restart, rename, and remove them in-app, with a live “supervisor online” indicator.
Recurring schedules
Run agents cron-style: pick the daemon, the time, and the prompt, with skip-if-busy and time zones. Toggle them, run one now, and read the execution log.
Your agents can talk to each other — across machines.
Cockpit's onboarding sets up the remote-pi extension, which hands every agent the mesh tools. So an agent in one pane can reach another — in the next pane, or paired in from a different machine — and they collaborate over the Remote Pi mesh. That's exactly what the screenshot up top is doing.
The mesh is a capability of the Remote Pi ecosystem — Pi plus the remote-piextension — shown off inside Cockpit, not a separate feature of the app. Cockpit doesn't draw a network map, and you don't drive its panes from your phone: the agents themselves do the talking.
macOS today, Windows and Linux too.
Cockpit ships signed and notarized on macOS, with Windows (.exe) and Linux (.deb/.rpm) installers from the same release pipeline. It drives a local Pi install rather than bundling one — the first-launch onboarding checks your setup and guides anything that's missing.