Cockpit

Free software · Linux, macOS, Windows

A whole project,
in a single window.

Run your AI agents where you work: terminals that survive closing the window, sessions resumed in one click, and next to them your files, Git, notes, tasks, clusters and tickets. A Cockpit project is a name and a folder, nothing more to set up.

Version 1.0.0 · 10/06/2026 · release notes

cockpit — dashboard
Cockpit showing the dashboard: tasks from two projects, each with its progress and due date.

01

It keeps running

Close the window: the terminals carry on, screen and history included.

02

Nothing to configure

A name and a folder. No project file, no six-step wizard.

03

Free, and you can check

Public code under the MIT licence. Anyone can read it and build it.

Terminals

They do not die with the window

A separate service holds the shells. Close Cockpit, open it again: everything is there, including what you had scrolled back to. Even after the machine was shut down, each tab comes back in its folder with what it was showing. Side-by-side panes, history search, a dropped file types its path, and agent sessions resume in one click.

A terminal open in Cockpit, showing the output of git status and git log.

Git

The diff is where you work

Status, coloured diff, staging file by file, commit, branches, history. And worktrees: one working folder per branch, so several things can run in parallel without stepping on each other.

Cockpit's Git tab: the list of changed files and the coloured diff of the selected one.

Files

Read and fix without switching tools

A tree that respects gitignore, syntax colouring for about thirty languages, in-place editing, project-wide search and go-to-definition. The system bin, not a blunt delete.

Cockpit's Files tab: the project tree and a file shown with syntax colouring.

And also

Everything around a project

Kubernetes

Pick a cluster and a namespace: pods grouped by service, the real reason something fails instead of “Running”, live logs, a shell, and failed pods deleted in one click. CPU and memory charted pod by pod. Cockpit reads your kubeconfig and keeps no secret.

Jira

Cloud or Server: your assigned tickets, on the dashboard and in each project. Change a status, comment, log time, create a ticket. “Start” creates the ticket's branch from an up-to-date main.

Containers

Compose starts in the right order, logs and a shell per container, a machine-wide view with cleanup. Entirely optional: with no compose file, the tab stays quiet.

Tasks, notes and meetings

Tasks with due dates and progress, a bell that warns you, notes in a tree with a rich editor and a reading mode, and a meeting recording transcribed then summarised into a note.

Palette and commands

Ctrl+K goes anywhere: projects, terminals, tabs, files. Declare your usual commands per project and run each in its own terminal with one click. Quick links with their live status.

Monitoring

Processor, detailed memory, disks, processes. An alert when a disk fills up or the machine stays saturated, never on a passing spike.

The AI provider you choose

Claude, Codex, Gemini and more: one setting decides, and the interface only shows what your provider can do. Its instructions, skills and plugins are edited from Cockpit, and its usage is displayed.

Make it yours

Themes, accent colour, wallpaper, zoom that stays sharp down to the terminals. French or English. Database backup in one click.

With an account, if you want one

Your projects from one machine to another

An account is all it takes to find your projects, notes and tasks on another machine. Paths stay local to each computer — they do not exist anywhere else.

Cockpit works without an account, and keeps working offline. Syncing is an extra, never a condition: if the server is unreachable, everything works the same.

Download

Pick your system

The binaries are not signed. On first launch, each system shows a warning you have to dismiss by hand. On Windows 11, when Smart App Control is on, it blocks Cockpit with no way around it: turn it off in Windows Security, App & browser control. That is a choice: a publisher certificate is expensive, and the code is public, so anyone can read it and build it themselves.

All releases