How to Install Codex CLI: A Step-by-Step Setup Guide

If you have already used Codex in ChatGPT, the Codex CLI will feel familiar. It brings the same AI assistance directly into your local development environment, where it can work with your repository, shell, tests, and other tools.

Installation takes only one command, but the setup choices that follow are worth reviewing carefully. In this article, you’ll learn how to install Codex CLI on Windows, macOS, and Linux and complete a clean first-time setup.

Before You Install

Use the standalone installer if you do not already manage developer tools through Homebrew or npm.

• Windows: PowerShell on a supported Windows system. WSL2 is optional.
• macOS or Linux: A terminal with curl available.
• Account: A ChatGPT account or an OpenAI Platform API key.
• Workspace: A project folder you own and trust.

Installation

1. Run the installer

Open a terminal. Next, run the command for your operating system.

macOS and Linux

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

Windows PowerShell

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

You can also run this via CMD. It executes the PowerShell command.

The Windows installer adds Codex to your PATH; consequently, it is ready for use. It also offers to launch the application.

Note: If the installer asks to start Codex immediately, press N. Do this unless the terminal is already inside a project folder.

Alternative install methods

Do you prefer a package manager? Use one of these options instead:

# Homebrew
brew install --cask codex

# npm
npm install -g @openai/codex

Similarly, Homebrew is suitable for a Brew-managed Mac setup. Use npm if Node.js already manages your global developer tools.

First Run

2. Open a project folder

Move into the repository where you want Codex to work. Then, launch it.

# Windows PowerShell
cd C:\Users\you\code\my-project
codex

# macOS or Linux
cd ~/code/my-project
codex

Codex treats the current directory as its starting workspace. By default the System32 directory would be the default one.

3. Choose how to sign in

The first launch offers three sign-in routes:

• Sign in with ChatGPT: Uses the Codex allowance from your ChatGPT plan.
• Sign in with Device Code: Best for SSH, containers, or machines without a browser callback.
• Provide your own API key: Usage is billed through the OpenAI Platform account linked to that key.

OpenAI’s current Codex pricing page lists Free, Go, Plus, Pro, Business, Edu, and Enterprise plans. API-key usage is billed separately. On the other hand, usage in a subscription is shared between the web app and Codex.

4. Authorize Codex in the browser

If you choose ChatGPT, Codex prints an authentication URL. It also opens this URL in your browser. If the browser does not open, copy and paste the URL yourself.

Next, sign in to the browser using the ChatGPT account you want Codex to use.

Check the account on the confirmation screen before you continue.

As a result, when the browser shows the success page, close the tab. Then, return to the terminal.

5. Review the first-run notice

Back in the terminal, Codex confirms your account. It also reminds you that autonomy is a choice. Do not ignore this notice. Press Enter after you read it.

You can inspect or change the active boundary later. Use the /permissions command to do this.

6. Trust the right folder

Codex now prints the directory it is about to trust. Read the path carefully. Do not just look at the highlighted option.

Trust allows project-local configuration, hooks, and execution policies to load. If the path is wrong, quit the program. Move into the correct repository and run codex again.

7. Configure the Windows sandbox

On native Windows, Codex may ask how to create its sandbox. Choose the default sandbox if you can approve the administrator prompt.

OpenAI’s Windows sandbox documentation calls the elevated mode the preferred native setup. It uses lower-privilege sandbox users. It also uses filesystem boundaries, firewall rules, and local policy changes. The non-admin mode exists for machines where elevation is unavailable. However, it offers weaker protection.

8. Send a safe first prompt

Once the prompt appears, start with a read-only task. This confirms that Codex can see the repository. It does not ask the tool to modify anything.

Explain this repository's structure. Do not change any files. Then list the three commands I should run to verify it locally.

Useful Commands After Setup

• /status: show the model, directory, account, and active settings.
• /permissions: inspect or change what Codex can read, edit, or run.
• /model: choose the model and reasoning effort available to your account.
• /init: create an AGENTS.md file for repository-specific instructions.
• /review: review changes before you commit them.
• /exit: leave the CLI cleanly.

Keep the default workspace boundary unless the task needs more access. The permissions documentation explains the read-only, workspace, and full-access profiles.

Updating Codex CLI

For a standalone installation, rerun the platform installer. OpenAI uses the same command for installation and updates.

# macOS or Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh

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

Update package-manager installs through the same manager:

# Homebrew
brew upgrade --cask codex

# npm
npm install -g @openai/codex@latest

Finish with codex --version to confirm the installed build.

Conclusion

Installing Codex CLI comes down to three moves:

1. An install command
2. Single account approval
3. A workspace and sandbox decision

The only genuinely bad setup is the convenient one in the wrong directory. Launch from the repository, keep the sandbox boundary, check /permissions, and begin with a read-only instruction.

Once that is clean, run /init and let the project carry its own rules.

Read more: Top 10 Skills for Claude Code and Codex CLI