Download AI Cockpit

One download. Free for everything you run on your own computer. Partway through the install you choose an edition, and you can always run both.

Try the demo first: nothing to install

v2.0.0 — Latest

cockpit_install.tar.gz

The whole application: PHP backends, front end, scripts, documentation and tools in one archive.

Download cockpit_install.tar.gz

~18,636 KB · 53 files · 210 downloads · every file has a SHA-256 checksum in MANIFEST.txt

On Linux, open a terminal in your Downloads folder and run:

tar -xzf cockpit_install.tar.gz && cd cockpit && bash install_cockpit.sh

A setup window opens (over SSH: the same questions in the terminal): edition, models, optional tools. A few pages, one password prompt, then it installs on its own. Running it again later is safe; it remembers your answers.

Two editions, one download

Full (the default)

Asks for a free account once, at first run: no card, ever. Opens the community Marketplace, and later the optional paid plans (see Pricing). Everything that runs on your computer stays free.

Stand-alone (V1)

No account and no calls to our servers, ever. The trade: no Marketplace, no Cruise or AutoCognition, no phone remote on this install. Everything else runs fully local.

The choice only sets what is switched on: the code is the same. Every bundle carries an integrity manifest, so the app can check that nothing was changed. Older cockpit_full_project.tar.gz and cockpit_v1_noaccount.tar.gz links still work.

Install steps by system

Tested on six distributions

Linux

Linux Mint, Ubuntu 24.04, Debian 13, Fedora 44, Manjaro/Arch and openSUSE Tumbleweed. Hand-tested on fresh installs of each, and an unattended install passes in our automated checks.

Use the command above. Setup Guide

Tested: Windows 11 + WSL2

Windows

Right-click the download → Extract All, then run install_windows.ps1 in an administrator PowerShell. It sets up WSL2 and a Start-menu shortcut, and the Linux installer runs inside WSL. Tested end to end on a clean Windows 11.

Windows steps · Tell us how it goes

Install tested; interface not yet

macOS

Run tar -xzf on the archive in Terminal, then the installer. On Apple Silicon our automated check runs the real installer to the end and the app answers. The interface itself has not been hand-tested on a Mac yet.

macOS steps

Install tested on ARM64

Raspberry Pi / ARM Linux

The same archive and install_cockpit.sh as regular Linux. An unattended install passes on a real ARM64 runner.

Setup Guide

Other ways to get it

Docker · not tested yet

Docker bundle

A small bundle (Dockerfile + docker-compose.yml + README) that packs Cockpit and Ollama into one container. Build it locally, then open http://localhost:8080. Written and reviewed, but not yet built end to end: tell us how it goes.

Download cockpit_docker.zip

~6 KB · Docker section of the Setup Guide

No script

Manual-install package

The app files already extracted, plus a step-by-step guide for Linux, macOS and Windows with no script to run. The installer scripts are still inside if you change your mind.

Download cockpit_manual_install.zip

~18,789 KB · read the guide online

Open source

Native window sources

Cockpit opens in its own small window (GTK 3 + WebKitGTK, written in Free Pascal). The archives above include it ready built; this is the source alone, under the Apache License 2.0, to read or rebuild with bash build.sh (needs only fpc).

Download cockpit_shell_src.tar.gz

Everywhere · beta

Phone, tablet, anything else

Android, iOS or any other device can reach your own Cockpit through the Everywhere relay: a free account with the Everywhere tier, and nothing installed on the phone beyond a browser tab.

Open the Everywhere relay

Before you install

You need

One of the systems above (the installer adds what it can for you)
Apache 2.4 + PHP 8.3 or PHP's CGI package for the self-contained start, which needs no Apache
PHP extensions: curl, mbstring, xml, zip
tmux, jq, curl
WebKitGTK 4.1 + GTK 3 for Cockpit's own window (the installer adds them; without them Cockpit opens in your browser)
Node.js 20 LTS, only if you use the Claude, Codex or Gemini command-line agents
RAM and a graphics card only decide how fast local models run: 4 GB of RAM is enough to run Cockpit itself (what runs on what)

Nice to have

pandoc + texlive-xetex (document conversion)
poppler-utils (PDF text)
ffmpeg (media)
whisper.cpp (speech to text)
LibreTranslate (local translation)
archmage (CHM files)
imagemagick (images)

Agents: Claude Code, Codex, Gemini and others each need their own sign-in or key; local models need Ollama or LocalAI. See Features.

Quick install (Ubuntu / Mint)

sudo apt install apache2 php8.3 \ php8.3-curl php8.3-mbstring \ php8.3-xml php8.3-zip \ libapache2-mod-php8.3 \ tmux jq curl wget \ pandoc ffmpeg # Node.js via NodeSource: curl -fsSL https://deb.nodesource.com/\ setup_20.x | sudo bash - sudo apt install nodejs # Claude Code + Gemini CLIs: npm install -g \ @anthropic-ai/claude-code \ @google/gemini-cli

After you install

The installer adds AI Cockpit to your applications menu, the cockpit-start and cockpit-stop commands, and offers a Ctrl+Alt+C shortcut that brings Cockpit up from any application (and starts it when it is not running). Closing its window runs the proper exit procedure. Install both editions and Stand-alone gets its own menu entry and cockpit-v1-start; only one runs at a time.

menu_book