How to use Codex on Linux in 2026: 8 ways compared

Disclosure: Enjoy is our product. We checked every way below against its maker’s own pages on November 10, 2026, and we say where another way is the better pick, including the free ones that don’t need Enjoy.

Ways to use Codex on Linux, compared

Codex has always run on Linux in a terminal. What changed is the window: “The ChatGPT desktop app for Linux is available in preview,” OpenAI’s Linux page says, so you can “sign in with your ChatGPT account to work with projects, local files, and Codex.” OpenAI announced the preview on August 11, 2026. The choices now are which distribution you’re on, and whether you want Codex in a window, a terminal, an editor or a browser.

Way	Best for	Price, checked November 10, 2026	Distributions	You use it in
The ChatGPT desktop app, in preview	A window, from OpenAI	Every ChatGPT plan, from Free; Plus is $20 a month	Ubuntu 24.04 and 26.04, Debian 13, Fedora 43 and 44, Arch; x64 or ARM64	The app
OpenAI’s CLI installer	Most terminal users, and servers	Plus ($20 a month) and up, or an API key	Linux builds for x86_64 and arm64	Your terminal
npm, or Arch’s own package	Keeping the CLI with your other packages	The same plans	x64 or arm64 from npm, with Node.js 16 or later; Arch’s package is x86_64	Your terminal
The IDE extension	Staying in your editor	Plus and up, or an API key	Wherever VS Code, a fork or a JetBrains IDE runs	A panel in your editor
Zed	An open-source editor with Codex in a thread	Free; your ChatGPT plan pays for Codex	x86_64 or ARM, best with a Vulkan GPU	Zed’s threads
Codex Cloud	Nothing to install, and work that keeps going	Plus and up, within your limits	Any, in a browser	Your browser or the app
Enjoy	Codex beside other labs’ agents	A monthly subscription, free for 7 days (pricing)	x64: AppImage, .deb for Ubuntu or Debian, .rpm for Fedora	Enjoy’s window
Orca	Many agents side by side, free	Free	x64 and arm64: AppImage, .deb, .rpm, and the AUR	Orca’s window

Prices are per month as each maker lists them, before tax. OpenAI’s pricing page lists “Codex on the web, in the CLI, in the IDE extension, and on iOS” under Plus, at $20 a month, with Pro at “$100, $200, or $500 USD per month.” Free and Go, at $8, list only “GPT-6 Luna at Standard speed in the desktop app, subject to rollout.” Zed, Enjoy and Orca run Codex on the same ChatGPT plan.

Which way works on your distribution

• Ubuntu and Debian: the ChatGPT app supports Ubuntu 24.04 and 26.04 LTS, and Debian 13. On Ubuntu 22.04 or Debian 12, OpenAI says “Other Linux distributions may work but aren’t formally supported,” so the CLI is the sure route there.
• Fedora: the ChatGPT app for Fedora 43 and 44, from an .rpm. Enjoy and Orca ship .rpm packages too.
• Arch: the ChatGPT app through OpenAI’s own install script, the CLI from Arch’s extra repository, and Orca from the AUR. Enjoy runs as an AppImage. The AUR also carries community repackagings of the app, such as codex-app-unofficial, an “Unofficial Linux build of Codex from OpenAI’s Codex appcast feed.” They aren’t OpenAI’s.
• Other distributions: the CLI. OpenAI’s GitHub releases build it for x86_64-unknown-linux-musl and aarch64-unknown-linux-musl, and the AppImages from Enjoy and Orca are another way to a window.
• ARM machines: the ChatGPT app, the CLI, Zed and Orca run on arm64. Enjoy is x64 only.

The ChatGPT desktop app, in preview

Best for: a window instead of a terminal, from OpenAI, at no extra cost.

The app is where Codex runs on a desktop: “Run projects in parallel, work with files, use your computer, and keep long-running work moving from one desktop workspace,” its overview says. For coding, its quickstart says, “select Codex from the ChatGPT dropdown.” On Linux, “Each supported distribution has packages for x64 and ARM64 processors.” Download the package from OpenAI’s Linux page, then install it:

sudo apt install ./chatgpt_amd64.deb        # Ubuntu or Debian, x64
sudo dnf install ./chatgpt.x86_64.rpm       # Fedora, x64

On ARM64, the files are chatgpt_arm64.deb and chatgpt.aarch64.rpm. On Arch, OpenAI’s page has you download install-arch.sh and run sudo bash install-arch.sh. “Installation configures a signed OpenAI package repository,” so updates come from your package manager: sudo apt install --only-upgrade chatgpt after sudo apt update, sudo dnf upgrade --refresh chatgpt, or sudo pacman -Syu.

From the app you can also reach a Linux server. Its remote connections “add remote projects from an SSH host and run chats against the remote filesystem and shell,” as long as you “Install and authenticate Codex on the remote host.”

Price: included in every ChatGPT plan, from Free. Plus is $20 a month, and Pro starts at $100.

Watch for: it’s a preview, and “Computer Use is available on macOS and Windows but not yet in the Linux preview.” “Native Wayland support is experimental”: in a Wayland session “the app uses XWayland when available,” and when you choose native Wayland with chatgpt --ozone-platform=wayland, “Some features, such as floating windows, window positioning, focus, and keyboard shortcuts, may not fully work.” On Arch, “Installation performs a full system upgrade and prompts for confirmation.” Free and Go get GPT-6 Luna only, where Plus adds GPT-6.1 Sol. And Codex is the only coding agent it runs.

The CLI, with OpenAI’s installer

Best for: most people in a terminal, and servers without a desktop.

OpenAI’s Codex CLI page gives “the standalone installer for macOS and Linux”:

curl -fsSL https://chatgpt.com/codex/install.sh | sh

The same line updates it. Then “Open a project directory and run codex. The first time you run Codex, choose Sign in with ChatGPT.” OpenAI’s standalone installers “download from https://releases.openai.com/codex by default and fall back to GitHub Releases if a metadata or asset download is unavailable,” its README says.

On a server, OpenAI warns that “the browser-based login UI may not work.” Its authentication page says to “prefer device code authentication (beta)”: turn on device code login in your ChatGPT security settings, run codex login --device-auth, then “Open the link in your browser, sign in, then enter the one-time code.” Its fallbacks are copying ~/.codex/auth.json from a machine where you’ve signed in, or ssh -L 1455:localhost:1455 user@remote to forward Codex’s sign-in callback.

Price: OpenAI’s pricing lists the CLI from Plus, at $20 a month. With an API key instead, you “Pay for Codex usage based on API pricing,” with “No cloud-based features.”

Watch for: the sandbox needs bubblewrap, below. And your sign-in sits in “a plaintext file at ~/.codex/auth.json or in your OS-specific credential store.” OpenAI says to “treat ~/.codex/auth.json like a password,” and setting cli_auth_credentials_store = "keyring" in config.toml keeps it in your system’s keyring instead, and “fails if it is unavailable.”

The CLI from npm, or Arch’s own package

Best for: machines where Node.js or pacman already manages your tools.

OpenAI’s CLI page lists npm as another way, and the same command updates it. The package carries Codex for Linux on x64 and arm64, and asks for Node.js 16 or later:

npm install -g @openai/codex

On Arch, sudo pacman -S openai-codex installs it from Arch’s extra repository, where on November 10 it was at version 0.167.0, the same as OpenAI’s latest release. It updates with the rest of your system.

Price: the same plans.

Watch for: Arch’s package is built by Arch’s own maintainers, not by OpenAI: its package page lists Arch’s packagers, and OpenAI’s CLI page names only its own installers, npm and Homebrew.

Turn on the sandbox

The sandbox “is the boundary that lets the agent act autonomously without giving it unrestricted access to your machine,” and “applies to spawned commands,” OpenAI’s sandbox page says. “On Linux and WSL2, install bubblewrap with your package manager first”:

sudo apt install bubblewrap   # Ubuntu or Debian
sudo dnf install bubblewrap   # Fedora

“Codex uses the first bwrap executable it finds on PATH.” Without one, “Codex falls back to a bundled helper, but that helper requires support for unprivileged user namespace creation,” and Codex shows a warning at startup. On Ubuntu 25.04, “installing bubblewrap from Ubuntu’s package repository should work without extra AppArmor setup.” On Ubuntu 24.04, Codex “may still warn that it can’t create the needed user namespace”; OpenAI’s fix is to install apparmor-profiles and apparmor-utils, copy the bwrap-userns-restrict profile into /etc/apparmor.d/, and load it with sudo apparmor_parser -r. OpenAI says to “prefer loading the bwrap AppArmor profile so bwrap can keep working without disabling the restriction globally,” which its last resort, sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0, does.

The IDE extension, in VS Code, a fork or a JetBrains IDE

Best for: people who already work in an editor.

“VS Code and compatible editors use the Codex extension; Xcode and JetBrains IDEs provide their own integrations,” OpenAI’s IDE page says. It installs in VS Code, Cursor and Windsurf, and it’s on the Open VSX Registry under OpenAI’s verified namespace, for editors such as VSCodium that install from there. Choose the Codex icon, or run Codex: Open Codex Sidebar. In IntelliJ, PyCharm and the other JetBrains IDEs, Codex is an agent in AI Assistant, and JetBrains’ page lists a JetBrains AI subscription, an API key or “ChatGPT account” to sign in.

Price: OpenAI’s pricing lists the IDE extension from Plus, at $20 a month, or with an API key.

Watch for: Free and Go don’t include it. In JetBrains IDEs, “Codex does not work with AI Assistant’s .aiignore functionality, which means that files listed in .aiignore can be processed by Codex.” It runs only Codex.

Zed

Best for: an open-source editor with Codex in a thread beside your code.

Zed installs on Linux with curl -f https://zed.dev/install.sh | sh, and its Linux page says it “works best” on systems with “a Vulkan compatible GPU” and glibc 2.31 or later on x86_64, or 2.35 on ARM. Zed’s setup by company offers Codex two ways: “Codex via ACP,” as “Codex in an External Agent thread,” or the “Native Codex CLI experience in a Terminal Thread.” Install Codex “from the ACP Registry, then start a Codex thread from the Agent Panel or Threads Sidebar,” its external agents page says.

Price: “Zed is free to use,” its pricing says, and “Zed does not charge for External Agents.” Your ChatGPT plan or API key pays for Codex.

Watch for: “Codex owns its own authentication and billing. An OpenAI API key configured for Zed Agent does not automatically configure Codex.” To change accounts, “use the Codex thread’s native login/logout flow.”

Codex Cloud, in your browser

Best for: any distribution or a borrowed machine, with nothing to install, and tasks that should keep going after you shut the laptop.

Codex Cloud runs tasks on OpenAI’s computers: “Each task has its own workspace and can keep working while your computer is asleep.” Open ChatGPT on the web or in the desktop app, “choose Work in > Cloud,” and pick an environment. From the CLI, codex cloud lets you “submit work to a configured environment, and apply the result to your local repository.”

Price: listed from Plus, within your plan’s usage limits.

Watch for: it starts from GitHub. To set up an environment, “Choose your GitHub repositories and select Get started. Connect GitHub if prompted.” A project that isn’t on GitHub needs another way here.

Enjoy

Best for: Codex beside other labs’ agents in one desktop app on Linux.

Enjoy runs Codex, Claude Code, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on Linux, Mac and Windows, on the plans you already have. Its download page has an AppImage, a .deb for Ubuntu or Debian and an .rpm for Fedora. It installs Codex for you from OpenAI’s official installer “without administrator access,” and signs it in with your ChatGPT account in your browser. Then:

• Your past Codex chats. Enjoy imports the Codex conversations already on your computer, so you pick up where you left off.
• Switch agents mid-conversation. When Codex is the wrong fit or out of usage, Claude Code or another agent picks up the same conversation.
• Keep going through limits. When Codex runs out, Enjoy switches to another of your accounts that still has usage, or waits for the limit to reset, then continues.
• Worktrees and parallel work. Each conversation runs its own agent, and an agent can work in a worktree, a separate copy of your project on its own branch.

Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and one computer; each other computer is $9.99 a month, and each teammate you invite needs a subscription of their own. See pricing.

Watch for: Enjoy isn’t open source, and it’s a monthly subscription on top of your ChatGPT plan once the free week ends. Its install guide asks for “a Linux computer with a 64-bit (x64) processor,” so ARM machines can’t run it. The AppImage updates itself and needs “FUSE 2, xdg-utils, and desktop-file-utils,” but “If you installed the .deb or .rpm package, download the latest package from enjoy.dev to update.” It needs “a keyring service such as GNOME Keyring installed and running,” and on Linux, “closing the Enjoy window quits Enjoy,” so leave it open, or minimized, for your phone to reach it. Scheduled workflows run only while your computer is awake with Enjoy open.

Orca

Best for: many agents side by side, free and open source, on x64 or ARM.

Orca, from Stably AI, says “Run Claude Code, Codex, and any other coding agent in parallel, each in its own worktree.” Its README says “Orca is free and open source under the MIT License,” and its latest release, 1.4.255 on November 9, has Linux AppImages and .deb and .rpm packages for x64 and arm64, with yay -S stably-orca-bin for the AUR. For Codex, “Orca reads ~/.codex for accounts and credentials,” and its account switcher “swaps the active account without re-login or config edits.” It can also run worktrees on “a Linux VPS” or another SSH host.

Price: free. Your ChatGPT plan pays for Codex.

Watch for: you install and sign in to Codex yourself: its Codex page says “Install Codex per OpenAI’s docs” and “Log in from any terminal.” By default each Codex tab runs its own server, so “Codex’s shared-server features (background runs, codex agents, and sessions opened from the Codex desktop app or an IDE) do not see sessions started in Orca.” And “Orca does not sell managed VPS hosting,” its ways to run page says, so an always-on machine is yours to run.

When to stay with the terminal

If codex already runs in your terminal, you don’t need anything else. OpenAI’s installer line keeps it updated, it runs on any distribution its Linux builds do, and with bubblewrap it has the sandbox. On Ubuntu 24.04 or 26.04, Debian 13, Fedora 43 or 44, or Arch, the ChatGPT app adds a window, parallel projects and remote SSH projects at no extra cost, even on the Free plan. Add another app when you want other labs’ agents in the same place, or an AppImage on a distribution the ChatGPT app doesn’t list.

How we compared

We read OpenAI’s pages on the ChatGPT desktop app and its Linux preview, the Codex CLI, its GitHub page and latest release, sandboxing, authentication, remote connections, the IDE extension, Codex Cloud and pricing, the Open VSX listing, JetBrains’ Codex page, Arch’s openai-codex package and the AUR’s codex-app-unofficial, Zed’s Linux, AI and pricing docs, Orca’s homepage, docs, GitHub page and latest release, and Enjoy’s own help on November 10, 2026, and every one is linked below. We quote each maker’s own words for its limits and leave out what it hasn’t documented. These apps change weekly, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.