OpenAI's Codex CLI: an open-source coding agent that runs in the terminal

Four ways to install it

The repository's documentation lays out installation commands for each operating system. On Mac and Linux, you install it with this curl script:

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

On Windows, you use a PowerShell command:

powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

By default, this installer script pulls files from releases.openai.com/codex, and automatically falls back to GitHub Releases if there's a problem. Setting the environment variable CODEX_INSTALLER_USE_RELEASES_OPENAI_COM to false forces it to use GitHub Releases from the start instead.

The official README in the repository also lists npm and Homebrew installation commands.

npm install -g @openai/codex

brew install --cask codex

If you'd rather download the executable directly, the GitHub Releases page offers Apple Silicon (codex-aarch64-apple-darwin.tar.gz) and Intel (codex-x86_64-apple-darwin.tar.gz) builds for macOS, plus x86_64 and arm64 archives for Linux. Once you unzip the archive, you'll find a single executable file with the platform name attached to it — the repository recommends renaming it to codex for convenience.

Getting started

1. Once you've installed it using any of the four methods above, the codex command becomes available in your terminal.
2. Type codex into the terminal and run it, and you'll see a screen asking you to choose a login method.
3. Selecting "Sign in with ChatGPT" lets you log in with your account. OpenAI supports two options: "Sign in with ChatGPT" and API key authentication.
4. If you don't have a ChatGPT subscription, you can connect an API key instead, though the repository notes this requires some additional setup.
5. Once you're logged in, you can start giving it tasks right away — just describe in natural language what code you want fixed or what new file you want created.

How Codex CLI, IDE, app, and web differ

Even under the same Codex brand, where each version runs and how you access it varies quite a bit.

Format	Runs on	Access path
Codex CLI	Local terminal	Install via curl, npm, or brew
Codex IDE	Inside VS Code, Cursor, Windsurf	IDE installation page
Codex app	Desktop	codex app command or app landing page
Codex web	Browser (cloud)	chatgpt.com/codex

If you want to use it right inside your code editor, install the extension from the IDE page. If you'd rather work in a separate window without opening a terminal, get the desktop app. And if you want to hand off work entirely through a browser with no local setup at all, Codex web is the cloud-based option — the key difference from CLI being that the work runs on OpenAI's servers instead of your own machine.

Same name, different ways to run it

OpenAI's official documentation treats Codex CLI, the IDE extension, the desktop app, and Codex web as distinct access methods. This article is specifically about Codex CLI, the terminal-based version — usage figures for other Codex products or enterprise services shouldn't be conflated with CLI statistics.

You can find the full source code and installation files for Codex CLI at the github.com/openai/codex repository.

Editor's take

Codex CLI's strength is that it works directly with files and commands right in your terminal. That said, the fact that it runs locally shouldn't be taken to mean anything specific about the scope of data processing involved.

Before rolling it out at work, it's worth checking which access method — CLI, IDE, app, or web — actually fits your team's workflow.

Since the official documentation clearly separates CLI, IDE, app, and web, teams can just pick whichever matches where the work happens. CLI suits terminal-centric development, the IDE extension fits work done inside an editor, and Codex web works well for browser-based delegation.