ChatGPT Desktop for Linux: What the Codex Preview Changes

The ChatGPT desktop app for Linux is now out in preview, with Codex, OpenAI's coding agent, built into it. The story is doing the rounds on Hacker News today, pointing at OpenAI's Codex page, and the install details sit in OpenAI's Linux app docs.

I run Linux as my daily driver, so my question was not "is a GUI nice." It was narrower: which distros does this actually cover, what does the installer put on my system, and does it beat the terminal I already have?

---

🐧 What actually shipped, and what didn't

This is a preview, and the supported list is short and specific. OpenAI names desktop versions of these distributions only:

Distribution	Versions supported	Architectures	Package
Ubuntu	24.04 LTS, 26.04 LTS	x64, ARM64	.deb
Debian	13	x64, ARM64	.deb
Fedora	43, 44	x64, ARM64	.rpm

Two things stand out. First, ARM64 builds exist from day one for every supported distro, which matters if you are on an ARM laptop or an ARM cloud desktop rather than a plain x86 box. Check yours before downloading:

uname -m     # x86_64 → x64 package, aarch64/arm64 → ARM64 package

Second, the gaps are real:

• Ubuntu 22.04 is not on the list. Neither is Mint, Arch, openSUSE, or anything else. The docs say other distributions "may work but aren't formally supported."
• Computer Use is missing on Linux. OpenAI says that feature works on macOS and Windows, and a future release will bring it to Linux.
• Native Wayland is experimental. In a Wayland session the app uses XWayland when available. You can force native Wayland by quitting fully and running chatgpt --ozone-platform=wayland, with the caveat that floating windows, positioning, focus, and keyboard shortcuts may misbehave.

That --ozone-platform flag is a Chromium flag, so this is a Chromium-based shell rather than a native GTK or Qt app. Budget your RAM accordingly.

---

📦 The installer adds a package repo, and that's the line to read twice

Installation itself is ordinary:

cd ~/Downloads
sudo apt install ./chatgpt_amd64.deb      # Ubuntu / Debian
sudo dnf install ./chatgpt.x86_64.rpm     # Fedora

The part worth pausing on is what happens next. Per the docs, the package configures the signed OpenAI package repository during installation, so later updates arrive through apt update && apt install --only-upgrade chatgpt or dnf upgrade --refresh chatgpt.

Warning: that is genuinely convenient, and it also means you have added a vendor-controlled software source with root-level install rights to your system's update path. On your personal laptop, fine. On a shared office machine, a client's box, or anything that also runs a build, decide that deliberately rather than by accident.

After installing, look at what landed:

grep -ri chatgpt /etc/apt/sources.list.d/ /etc/yum.repos.d/ 2>/dev/null

Key takeaway: the news isn't that Codex reached Linux. The CLI has run on Linux for a long time. The news is that OpenAI now wants a supported, self-updating package sitting inside your distro's package manager.

---

⚡ When the GUI helps, and when the CLI still wins

Most developers I know here work on a laptop that is also the family machine, a cheap VPS abroad, or a company box they don't fully control. The app fits some of those and not others.

Your setup	Desktop app	Codex CLI
Ubuntu 24.04 or 26.04 laptop	Supported preview	Works
Ubuntu 22.04, Mint, Arch	Not supported	Works
Code lives on a remote VPS, you SSH in	No, the GUI runs locally	Yes, runs where the code is
Low-RAM machine already running Chrome and Docker	A Chromium shell costs you memory	Light by comparison
Metered or mobile connection	Heavier client to update	Smaller footprint
CI, cron jobs, scripted runs	Not the tool	The tool

If your work happens over SSH on a droplet in Singapore, a local desktop client does nothing for that workflow. Keep the CLI. If you build locally on a supported distro and you want the project view, worktrees, and the same account tying your editor, terminal, and chat together, the app is worth an evening.

---

💰 The real gate here is the account, not the app

The package is a download. Using it is not: the docs say you sign in with your ChatGPT account to work with projects, local files, and Codex.

I won't quote plan prices, because the pages I read for this post don't list them. Check openai.com/pricing yourself. What I will flag is what actually bites for a Sri Lankan builder:

• Billing is in USD. A subscription that reads as a small number in dollars is a different decision in rupees. If you are sizing that, our LKR exchange rate tool gives you today's rate and a 30-day chart, so you can see whether you are budgeting at a good moment.
• Card acceptance is its own obstacle. Plenty of local debit cards fail on international recurring billing. Sort that before you plan a workflow around the tool.
• Read the permissions docs before pointing it at client code. An agent with local file access on a machine holding someone else's repository is a decision you should make on purpose.

OpenAI's own marketing page leans on customer quotes, including a claim from Harvey's mobile lead that Codex cut early iteration time by 30–50%. Treat vendor testimonials as vendor testimonials. Measure it on your own repository for a week.

---

🛠️ How I'd trial it without breaking a working setup

1. Confirm your distro and architecture: cat /etc/os-release and uname -m. If you are on 22.04, either upgrade on purpose or stay with the CLI.
2. Install on a machine you can afford to reset, or a VM, before your main work box.
3. Check which repository got added, using the grep above.
4. Point it at a throwaway repository first. Not client work, not production credentials.
5. If windows behave strangely under Wayland, relaunch with chatgpt --ozone-platform=wayland and see whether it improves or gets worse.
6. Compare against your existing CLI workflow on the same task. If the GUI doesn't win on something concrete, uninstall it and move on.

---

💡 What this means for you

If you are on a supported distro and already paying for a ChatGPT plan, this is a low-risk trial and the project view may genuinely suit how you work. If you are on 22.04, on Arch, or doing most of your work over SSH, nothing here changes your setup today, and the CLI remains the sensible default.

The wider signal is more interesting than the app. AI coding tools started as terminal utilities and are now shipping as distro packages with their own signed repositories and update channels. That is a vendor asking for a permanent seat inside your operating system, not just inside your editor. Say yes if it earns it. Just notice that you are being asked.