Quick summary
Codex on the ChatGPT desktop app has arrived on Linux in preview status — the first time Linux users get the same experience as Mac/Windows folks, without relying solely on a browser or the CLI.
The headline additions are a native window, direct access to local files, and genuine async background task execution — no more leaving a browser tab open like you had to on the web.
But let’s be honest: this is still preview. Supported distros are limited, bugs are par for the course, and you’ll need a paid ChatGPT plan tier to unlock the full feature set.
For devs who already work primarily on Linux, this is a meaningful step that closes a lot of the cross-platform tooling gap.
Screenshots of Codex actually running on Linux desktop
Easy to spot that this is genuinely preview — there’s a version badge in the top-right corner, distinct from the regular stable ChatGPT desktop build.
The layout looks similar to the macOS/Windows version, but the font rendering and title bar are unmistakably GNOME-styled — an instant tell that it’s running natively on Linux, not through some compatibility layer.
The background task bar is the most interesting part, since it shows Codex running background processes separate from the main UI — close the window and the task keeps running.
The night I alt-tabbed a dozen times just to fix one bug
I still remember that night — one single bug to fix, but I had to alt-tab between a ChatGPT web browser tab and the terminal a dozen times, copying code from the editor to paste in, then copying the answer back. Round and round, all night.
A friend on Mac has the native Codex app and can just drag the window next to their editor. On Linux, I was stuck with a browser tab as my only option. Once I had several tabs open at once, it got confusing figuring out which context was discussing what.
The real friction was that every switch broke my train of thought — not just a couple seconds lost, but the whole chain of reasoning I was chasing down for the bug got derailed. Doing this every day turned into a quiet, everyday frustration.
Where this Linux build of Codex sits among OpenAI’s tools
OpenAI already has Codex in several forms: Codex CLI for terminal-first users, Codex on the web for async task dispatch, and an IDE extension — comparable to what Code offers for people who want to watch code change live in the same window.
The ChatGPT desktop app itself was already available on Mac and Windows before this. Linux is the newest platform to join, arriving as a preview.
Linux showing up last isn’t surprising — the general desktop Linux user base is much smaller than Mac/Windows, so OpenAI put it at the back of the queue.
But for the dev/infra crowd who already work primarily on Linux (servers, containers, CI/CD), this fills a gap that’s been open for a while — previously the only options were the CLI or a browser tab, instead of a native app that plugs directly into a desktop workflow.
What’s different now compared to before the desktop app
Previously it was: keep a browser tab open, or switch to the terminal and invoke the CLI every time you needed to give Codex a task — context got lost easily, especially when switching between screens.
With the Linux desktop app, that’s changed a lot. The app sits in the system tray, ready to launch instantly like any other native app — no more relying on a browser tab that disappears on a machine restart.
| Factor | Before Desktop App | Now (Desktop Preview) |
|---|---|---|
| Access method | Browser tab or CLI only | Native app, installed directly on the machine |
| Notifications | Have to keep a tab open and watch it yourself | System notification when the job finishes |
| Running background tasks | Depends on browser/terminal session | Runs independently of other workflows |
| Linux desktop integration | No direct integration | Tied into system tray/desktop |
Simply put, it’s a shift from “a tool you have to go open” to “a tool that’s always on your machine.”
Features that come up often in real dev work
Four things get used the most in practice: direct access to local files/folders without copy-pasting code into a browser, async task execution that continues even after you close the window, desktop notifications when a task finishes, and managing multiple repos at once within a single app.
The most common scenario: kick off a PR review with Codex and go into a meeting — come back and the result is already sitting on screen, no tab left open.
Or fixing a bug that touches multiple repos at once — just switch back and forth within the same app, no need to juggle 5-6 separate browser tabs and sessions.
Another pattern is kicking off a long refactor overnight — wake up, turn on the machine, and the notification is waiting. No need to babysit the screen.
Altogether, these features really answer the needs of devs juggling multiple contexts at once — not just one-off chat Q&A.
How it stacks up against competitors in the AI coding assistant space
Linux has long been a sore point users have complained about, since Copilot Workspace and Cursor have always focused primarily on Windows/macOS. Claude Code, meanwhile, already runs through the terminal, so it never had to wait for a separate desktop app.
Codex on ChatGPT desktop Linux is still preview right now, so stability-wise it can’t yet compete with something that’s been out longer, like Cursor. On the front of independent agent work — running in the background across multiple repos/sessions at once — Codex and Claude Code are headed in the same direction, unlike Copilot Workspace, which still ties itself mainly to the GitHub flow.
Pricing and usage tiers change frequently across all of these, so check the official pages before deciding — I won’t nail down exact numbers here.
| Factor | Codex (ChatGPT desktop) | Cursor | Claude Code |
|---|---|---|---|
| Linux support | Preview (just launched) | Not Linux-focused | Supported via CLI |
| Usage model | Desktop app | Desktop app | Terminal/CLI |
| Independent agent work across sessions | Yes (preview) | More limited | Yes |
Pros and cons from real-world use in preview status
Actually being able to use a native app on Linux is the biggest game-changer — no more leaving a browser tab open like before. Async tasks that you let the agent run in the background are also a lot more convenient now, since the whole workflow lives inside one app.
But it really is still a preview. Currently supported distros are still limited to a handful — if you use a more niche distro, you may have to wait or find your own workaround. There are still a few small UI bugs to run into, and resource usage (RAM/CPU) isn’t as optimized as it should be yet. Some features that already exist on Mac/Windows are still lagging behind here too, waiting on future patches.
Pros
- +Genuinely usable native app on Linux, no need to rely on a browser
- +Async background tasks are more convenient and easier to manage than before
Cons
- −Supported distros are still limited, doesn't cover every distro family
- −Still has UI bugs and uses more system resources than it should
- −Some features still lag behind the Mac/Windows versions
Costs that don’t show up on the pricing page
The subscription page only shows package pricing — it doesn’t mention that the full Codex feature set requires a high-enough plan tier to unlock completely. Lower-tier plans may leave you with a limited experience.
The quota or compute used to run async tasks also has a monthly cap — heavy usage might hit that limit faster than you’d expect.
Another hidden cost is time: on distros that aren’t fully supported yet, you may end up spending time chasing down dependency issues or incompatibilities with your desktop environment yourself.
Most importantly, this is a preview build — app behavior can change at any point. Each update might mean the workflow you got used to needs re-adjusting. Always have a backup plan if you intend to rely on this for serious work.
Who it’s for, who it’s not for
Made for
- Devs who already use Linux as their main machine, with a ChatGPT plan that supports this feature fully
- Teams who want to experiment with async automation alongside their existing workflow
Think twice
- Users on a less common distro or desktop environment that hasn't been confirmed as fully supported — be prepared to chase down dependencies yourself
Skip this one
- Production work that demands high stability — preview builds can change behavior with every update, better to wait for the stable release
What to watch going forward
The likely next direction is OpenAI closing the gap between Linux and Mac/Windows further — both in terms of officially supported distros and the features that are currently still lagging behind.
Before rolling this out to your team for real, check three things: is the distro version your team uses on the supported list, are there any dependencies you’ll need to fix, and will your team’s existing workflow (CI/CD, scripts) clash with the preview build.
If you want to try it right now, it’s best to test it on a separate machine rather than your main work machine, and keep an eye on the changelog with every update, since preview behavior can shift at any time. For work that demands high stability, it’s safer to wait for the stable release first.