Codex on Linux: Install the Official ChatGPT Desktop App

OpenAI now has an official ChatGPT desktop app for Linux, and it includes Codex, so you can run your coding agent in a desktop window instead of only in the terminal. In this guide I’ll show you how I installed it on Debian, fixed a dependency error during install, opened a project, and used skills, usage limits, Fast mode and voice dictation. I’ll also cover which distros are officially supported, because that matters before you start.

Prefer to watch? See the full walkthrough on YouTube. I build the whole thing on screen.

What the ChatGPT Linux app is

For a long time the only way to get a desktop Codex experience on Linux was an unofficial app from GitHub. I’d already made tutorials on that. Then Codex in my terminal showed a notice saying “Try the desktop app on Linux” with a link, and I went straight to it.

According to OpenAI’s official Linux app guide, the ChatGPT desktop app for Linux is in preview. You install a package for your distro, sign in with your ChatGPT account, and work with projects, local files and Codex.

Check your distro and architecture first

As of October 2026, the official guide lists these supported desktop distributions:

• Ubuntu 24.04 LTS and 26.04 LTS
• Debian 13
• Fedora 43 and 44
• Arch Linux (current, fully updated rolling release)

Each has packages for x64 and ARM64. To check your architecture, run:

uname -m

x86_64 means x64, and aarch64 or arm64 means ARM64. The guide says other distros may work but aren’t formally supported. I’m on Debian testing (Forky), which isn’t on the list, and it worked for me after fixing one dependency issue, but your experience may differ.

Step 1: Download the package

Open the official guide and download the right package: .deb for Ubuntu or Debian, .rpm for Fedora, or the install script for Arch. I downloaded the .deb for x64.

Step 2: Install it from the terminal

Ubuntu and Debian

The official guide recommends installing the downloaded .deb with apt, which also pulls in any missing dependencies:

cd ~/Downloads
sudo apt install ./chatgpt_amd64.deb

For ARM64, use chatgpt_arm64.deb instead.

In the video I used sudo dpkg -i instead. dpkg unpacked the package but stopped because of missing dependencies, which on my system were Mesa Vulkan drivers. Fixing the broken install pulled them in, and after that the app installed cleanly. I show this step on screen in the video. If you hit the same thing after using dpkg, this is the usual fix on Debian-based systems:

sudo apt --fix-broken install

The simpler path is to use apt install ./package.deb from the start, so apt handles the dependencies for you.

Fedora

cd ~/Downloads
sudo dnf install ./chatgpt.x86_64.rpm

Arch Linux

On Arch, the guide provides an install script that sets up OpenAI’s signed package repository. It runs a full system upgrade and asks for confirmation:

curl --proto '=https' --tlsv1.2 -fL -o install-arch.sh \
  https://persistent.oaistatic.com/codex-app-prod/linux/install-arch.sh
sudo bash install-arch.sh

If the install fails on your distro, AI Income Lab is a good place to ask while you work through it.

Step 3: Launch the app and sign in

Open ChatGPT from your applications menu, or run:

chatgpt

The first launch took a little while on my machine, but then the app opened. Sign in with your ChatGPT account. From there you can switch between Codex and regular ChatGPT chats in the same app.

The guide also notes that native Wayland support is experimental. In a Wayland session the app uses XWayland when it’s available, and you can try native Wayland by fully quitting the app and launching it with chatgpt --ozone-platform=wayland. Some window and shortcut features may not fully work there yet.

Step 4: Open a project and set permissions

I opened my project folder and gave Codex full access, which is what I usually do on my own machine. That’s a personal choice, not a default you should copy blindly. Full access lets the agent run commands and change files without asking. Read OpenAI’s permissions docs and pick the level you’re comfortable with for each project.

Step 5: Pick a model and use your skills

The app opened on an older model from my earlier settings, which was no longer available, so I switched to the model I use most from the model picker. Model names change often, so just choose from whatever the picker currently offers on your plan.

My existing skills showed up in the app too. I ran my AI SEO skill and asked it for a full audit of my TubeAnalytics site. It came back with a list of fixes, and from there I could tell it to implement them, the same way I would in the terminal.

Step 6: Check usage limits and Fast mode

Inside the app, the usage command shows your 5-hour and weekly usage limits and when they reset. I’m on the Plus plan, and I could see both limits from inside the app. In the terminal version of Codex, /status shows the same 5-hour and weekly figures.

There’s also a Fast mode toggle. I turned it on to try it, but Fast mode also uses more of your usage allowance, so I switched it back off and stayed on standard speed. If you’re close to your weekly limit, leave Fast mode off.

Step 7: Voice dictation and the desktop pet

The version I installed (26.8) had a small on-screen pet you can move around and start a voice chat from. It felt a bit buggy, and I didn’t find much use for it, so I closed it.

Voice dictation is the feature I actually liked. I couldn’t find an easy way to dictate prompts in the terminal, so being able to talk instead of type is a real time-saver. Features may change between releases, and the guide notes that some, like Computer Use, aren’t in the Linux preview yet.

Desktop app vs terminal: which should you use?

The desktop app used around 1 GB of memory on my machine while running a task, plus a fair amount of CPU. That’s reasonable for a desktop app, but the terminal is still lighter, and I still prefer it for most coding. If you want a lightweight terminal for this, I’ve also covered Ghostty on the channel.

The desktop app is better when you want voice dictation, a visual view of your projects, or to jump between Codex and regular ChatGPT. Many people will end up using both.

Common mistakes to avoid

• Ignoring the supported distro list. If you’re on an unsupported release, expect to troubleshoot.
• Downloading the wrong architecture. Run uname -m first.
• Using dpkg and stopping at the dependency error. Use sudo apt install ./chatgpt_amd64.deb, or fix broken packages afterwards.
• Giving full access to every project. Match permissions to how much you trust the task.
• Leaving Fast mode on all the time. It uses more of your limits.
• Forgetting to update. The install sets up a signed OpenAI repository, so update with sudo apt update && sudo apt install --only-upgrade chatgpt on Debian or Ubuntu.

Want help turning this setup into real projects? Join AI Income Lab for the Codex lessons and templates I use.

FAQ

Is there an official Codex app for Linux?

Yes. Codex is part of the official ChatGPT desktop app for Linux, which OpenAI currently lists as a preview.

Which Linux distros does the ChatGPT app support?

The official guide lists Ubuntu 24.04 and 26.04 LTS, Debian 13, Fedora 43 and 44, and current Arch Linux, on x64 and ARM64. Others may work but aren’t formally supported.

How do I update the ChatGPT app on Linux?

Use your package manager. On Debian or Ubuntu run sudo apt update then sudo apt install --only-upgrade chatgpt. On Fedora run sudo dnf upgrade --refresh chatgpt, and on Arch update with sudo pacman -Syu.

Does the Linux app support Wayland?

Native Wayland support is experimental. The app uses XWayland when it’s available, and you can launch with chatgpt --ozone-platform=wayland to try native Wayland.

Should I still use Codex in the terminal?

If you want the lightest setup, yes. I still prefer the terminal for most work, and use the desktop app when I want voice dictation or a visual overview.

Conclusion

Getting Codex running on Linux with the official ChatGPT app takes a few minutes: check your distro, download the right package, install it with apt, dnf or the Arch script, then sign in and open a project. Set sensible permissions, keep an eye on your usage limits, and try voice dictation. Watch the video above to see the full install, including the dependency fix.

If you’d rather not figure this out alone, join AI Income Lab for the courses, templates and help that go with these builds.