ChatGPT Desktop App for Linux: Preview Tested
OpenAI launched a ChatGPT desktop app for Linux in preview on 11 August 2026, and if you run Ubuntu, Debian or Fedora on your work machine, this is the first official native client you have ever been offered. It bundles ChatGPT, ChatGPT Work and Codex into one app, installs from a .deb or .rpm package, and signs in with the ChatGPT account you already pay for. There is no separate Linux price. There is also no Computer Use, which OpenAI has confirmed is missing from this preview.
That last point matters more than the launch headline. If you were hoping to hand your Linux desktop over to an agent that clicks around your screen, the preview does not do that yet.
What OpenAI actually shipped
The Linux release is a native desktop client, described by OpenAI as available in preview. TechCrunch, AIWeekly, OMG!Ubuntu and Gigazine all put the launch date at 11 August 2026. Inside the app you get three named products: ChatGPT for general conversation, ChatGPT Work for structured workflows, and Codex as a coding agent that runs alongside the chat window.
The framing OpenAI uses in its own community announcement is worth quoting plainly, because it tells you what the app is for. It calls the client a workspace for managing projects, working with files, using browser workflows, and running Codex alongside ChatGPT. That is a developer workstation tool with a chat window attached, and it explains why Linux got Codex on day one instead of some cut-down chat-only build.
Signing in is required. Projects, local files and Codex all sit behind a ChatGPT account, so there is no anonymous or offline mode documented anywhere in the launch material.
Supported distributions and hardware
Four desktop distributions are supported at preview, and nothing else is documented.
| What | Supported at preview (August 2026) |
|---|---|
| Ubuntu | 24.04 LTS and 26.04 LTS |
| Debian | 13 |
| Fedora | 43 and 44 |
| CPU architecture | x64 and ARM64, native packages for both |
| Package format | .deb for Debian and Ubuntu, .rpm for Fedora |
| Variant | Desktop editions only |
Arch, openSUSE, RHEL and Pop!_OS are not on the supported list. OpenAI has not said whether they are coming, so treat anything you read about them as speculation until the docs change. This list was accurate on 17 August 2026 and preview support lists move fast, so check OpenAI's Linux documentation before you plan a rollout across a team.
The ARM64 build deserves a mention. Native Arm packages at launch means a Raspberry Pi 5 class machine or an Arm cloud desktop is in scope from day one, which is unusual for a preview and saves you the emulation dance.
What it costs
The Linux app has no price of its own. It uses your existing ChatGPT account, whether that is Free, Plus, Team, Enterprise or Work, and every source covering the launch says the same thing. No Linux-specific pricing, no Linux-specific quota, no separate desktop licence.
That is genuinely good news for a small business. If two people on your team already run Plus and one of them uses Fedora, the Linux client costs that person nothing extra. Nobody is being charged a platform tax for having chosen the wrong operating system.
What the launch material does not tell you is the current dollar figure on each plan, or the message and file limits attached to it. The Linux announcements simply do not carry those numbers, and I am not going to guess at them here. Check OpenAI's pricing page before you commit budget, because plan limits at this level of the market change several times a year and any figure I printed today would be stale by the time you read it. If you are still deciding which assistant to standardise on, our comparison of ChatGPT and Claude for business emails covers the day to day quality difference in more useful detail than a price list will.
The Computer Use gap
Computer Use is not available in the Linux preview. OpenAI states this directly in its "What's new" documentation, and independent Linux preview guides published in August 2026 repeat it.
Here is why that gap is the single most important line in the release notes. Computer Use is the feature that lets the model take control of the desktop itself, clicking through applications the way a person would. It is the piece that turns an assistant into an automation tool. Without it, the Linux app is a very capable chat and coding client with file access, and the automation ceiling is lower than on macOS or Windows.
There is a partial consolation. Trending coverage from August 2026 reports that while native Computer Use is absent, the in-app browser works and Chrome automation works. So browser-based workflows are alive. Anything that lives outside a browser tab, a desktop accounting package, a local design tool, a terminal-driven build pipeline you wanted driven by clicks, sits outside what this preview can touch.
For most small business automation, browser workflows are where the work actually is. Invoices, CRM records, supplier portals and dashboards are all web apps now. If your automation ambitions stop at the browser edge, the gap will barely bother you. If you were counting on OS-level control, wait.
What Codex brings to the desktop
Codex is a coding agent, and the Linux app runs it alongside ChatGPT in the same workspace. It works with local files, projects and browser workflows, according to OpenAI's own documentation and its community post.
The practical read for a small team: this is the first time an OpenAI client on Linux can see your project directory and act on it without you pasting code into a browser tab. For anyone maintaining a website, a small internal tool or a set of scripts, that removes the most annoying part of the workflow, which is the copy-paste round trip between editor and chat window.
What OpenAI has not published is where Codex stops. There is no documented detail on sandboxing, on which languages get first-class treatment, on execution limits, or on whether a runtime persists between sessions. Those are exactly the questions a cautious team asks before pointing an agent at a production repository, and the honest answer today is that the launch material does not answer them.
Honest cons
Every review needs the part where the product stops looking good. Here are the real ones, all of them traceable to what OpenAI and the launch coverage actually said.
Computer Use is missing. This is the headline limitation, confirmed in OpenAI's own documentation. The Linux client is behind the macOS and Windows clients on automation capability, and no date has been given for closing that gap.
It is a preview, and OpenAI says so. Gigazine's coverage notes the app ships with some minor issues because release speed was prioritised. Preview software on a work machine is a choice you make deliberately, not one you make for a client-facing production workflow in week one.
The supported distribution list is short. Five distribution versions across three families. If your team standardised on Arch, openSUSE or RHEL, you are not invited yet, and OpenAI has not said when or whether you will be.
Nothing about the limits is documented. No published maximum project size, no file count per project, no clarity on whether local file access is read-only or read-write, no offline capability described. For a tool you are about to point at real client files, that silence is a legitimate reason to hold off.
Sign-in is mandatory for everything useful. Projects, files and Codex all require an authenticated ChatGPT account. There is no documented way to run this as a local-only tool, which matters if you handle data that cannot leave the machine.
Who should skip this
Skip the Linux preview if you handle client data under a contract that restricts what leaves your machine. The app requires sign-in for file and project access, and nothing in the launch material describes a local-only mode, so the safe assumption is that your workflow involves OpenAI's servers.
Skip it if your distribution is not on the list. Sideloading a .deb onto an unsupported system is a route to a support conversation nobody will have with you, since this is preview software.
Skip it if OS-level automation was the whole reason you were interested. Computer Use is the missing piece, and until it arrives on Linux you are buying a chat and coding client, which you can already get in a browser tab. For automation that has to work today, a dedicated tool is the better answer, and our comparison of n8n and Zapier for small business automation walks through what each actually costs to run.
Skip it if you are on a Free plan and expecting the desktop client to unlock something extra. It does not. The app uses your existing account and existing entitlements, so the experience follows your plan, not your operating system.
Who should install it today
Install it if you write code on Ubuntu, Debian or Fedora and you already pay for ChatGPT. Codex running against local files, in a native window, on the machine where the code lives, is a real workflow improvement over the browser. The cost of trying it is the download and ten minutes.
Install it if you run browser-based operations work and want the in-app browser and Chrome automation available without a tab-switching dance. Those parts are reported working in the preview.
Install it on a secondary machine if you are curious but responsible. Preview builds and production deadlines do not belong on the same laptop.
How this fits the wider 2026 picture
Linux getting a native OpenAI client at all is the story here. For years the pattern was macOS first, Windows second, Linux never, with Linux users pointed at a browser tab or a community wrapper. Shipping .deb and .rpm packages, with native ARM64 builds, is a real commitment of engineering time to a platform that developers use and mainstream consumers do not.
The competitive context is worth holding in mind. Assistants are being pushed hard on every platform right now, and adoption numbers at the top of the market have gone somewhere no one expected, as our write-up on Gemini reaching a billion users covers. Native desktop clients are one of the ways that fight gets fought, because a tool sitting in your dock gets opened and a bookmark does not.
For a small business, the practical takeaway is narrow and useful. You do not have to switch operating systems, buy anything extra, or change plans to get an OpenAI client on your Linux workstation any more. You just install a package.
FAQ
When did the ChatGPT desktop app for Linux launch?
OpenAI released it in preview on 11 August 2026. TechCrunch, AIWeekly, OMG!Ubuntu and Gigazine all reported that date. It is labelled a preview by OpenAI itself, so expect rough edges and changes without much notice.
Which Linux distributions are supported?
Ubuntu 24.04 LTS, Ubuntu 26.04 LTS, Debian 13, Fedora 43 and Fedora 44, desktop editions only. Both x64 and ARM64 architectures have native packages, delivered as .deb for Debian and Ubuntu and .rpm for Fedora. Arch, openSUSE and RHEL are not on the supported list as of 17 August 2026.
Does the Linux app cost extra?
No. It uses your existing ChatGPT account and subscription, and there is no Linux-specific pricing documented anywhere in the launch material. Your plan determines what you get, so check OpenAI's official pricing page for current figures before budgeting.
Can it control my desktop like the Mac and Windows apps?
Not in this preview. OpenAI's documentation states that Computer Use is not yet available on Linux. In-app browser use and Chrome automation are reported working, so browser-based workflows function, but OS-level control does not.
Is Codex included, or is it a separate product?
Codex is included in the Linux desktop app and runs alongside ChatGPT in the same workspace. It works with local files, projects and browser workflows. Access follows your ChatGPT account rather than a separate Codex purchase.
Should I use it for client work right away?
Be careful. It is preview software with undocumented limits on file handling and project size, and sign-in is required for anything involving your files. For client work under a data handling agreement, wait for the stable release or check the terms carefully first.
Verdict
Install it if you code on a supported distribution and already pay for ChatGPT, because Codex with local file access in a native window is worth the ten-minute setup. Hold off if you need desktop automation, run an unsupported distribution, or handle files that cannot leave your machine. The preview is a genuine step forward for Linux users who have been stuck in a browser tab for years, and it costs nothing beyond the subscription you already have.
About the author
This review was written by the team at The Tool Signal. We run AI tools in daily production work for small businesses, including on Linux workstations, and we review them the way we buy them: with the price in front of us and the limitations written down. Every fact in this piece about the Linux preview comes from OpenAI's own documentation and from launch coverage published in August 2026, checked on 17 August 2026. Where OpenAI has not published a number, we have said so rather than filling the gap with a guess. That is the only way a review stays useful six months after publication, and it is why we would rather tell you that OpenAI launched a ChatGPT desktop app for Linux in preview with real gaps in it than sell you a launch announcement dressed up as a recommendation.