The Rundown

In this guide, we'll teach you how to 10x your code output with Codex. We’ll get you setup with it in Cursor, then walk you through how to run it in the cloud using your phone.

You’ll even learn how to setup an agent that automatically reviews your code for you.

Who is this useful for

• Developers who want AI coding help but find terminal-based tools intimidating
• ChatGPT Plus/Pro subscribers who haven't tried Codex yet
• Anyone curious about AI coding agents

What You Will Build

You'll install Codex, connect it to a project, and have it create an AGENTS.md file. That's the configuration file that tells Codex about your project. Then you'll explore slash commands and set up automated code reviews.

What You Need to Get Started

• A ChatGPT Plus account ($20/month)
• Cursor or another supported code editor (VS Code, Windsurf, etc.)
• A GitHub account (optional)

Step 1 Choose Your Version

Codex comes in three versions:

• Codex CLI runs in your terminal.
• Codex Cloud runs in your browser at chatgpt.com/codex.
• Codex IDE Extension runs in your IDE as a chat window

For this guide, we're using the Cursor extension because it displays your agent chat separate from your terminal. This keeps everything nice and tidy. If you use VS Code or Windsurf, you will be able to follow along as well.

Go to https://developers.openai.com/codex/quickstart/ and select your IDE. Follow the brief download instructions.

Step 2 Codex Project Setup

Open a project in Cursor. Press CMD+Shift+P and search Codex. Select New Codex Agent.

You should see a window open that looks like this.

Tell the agent to create an agents.md file based on the project files. If you’re starting from scratch you can give it your PRD or spec sheet instead.

After it creates the agents.md file, you should tell it to initialize git and write a commit, if you haven't yet.

Step 3 Slash Commands

You can click the slash button to see commands.

Here are the important ones:

• /review Quick code review of your changes
• /status Shows context limit (%) and your 5-hour usage limit
• /mcp Shows connected MCP servers
• /context Toggle auto-context from recent files

Here’s a link to the official documentation for all slash commands

Step 4 Use Codex From Anywhere

Go to https://chatgpt.com/codex/settings/connectors and connect your GitHub account.

This lets Codex access your repositories and create pull requests from its work. Tasks run in isolated cloud sandboxes so your local machine stays free.

Now you can access Codex from anywhere by going to chatgpt.com/codex. Just select a repository and a branch and give it a prompt. It will run the code in a virtual environment.

You can edit your project-specific virtual environments so Codex can use your API keys or browse the web.

Step 5 Automatic Code Reviews

This is where Codex shines. You can have it automatically review every pull request in the cloud:

1. Go to https://chatgpt.com/codex/settings/code-review
2. Enable automatic code reviews on new PRs
3. Flip Auto code review to “Review my PRs” for your repo

Now whenever you open a pull request, Codex will automatically review it.

Going Further

Token-Saving Hack

Cloud-based Codex calls (from the web or other connectors) don't count against your 5-hour local limit. Use this workflow to save tokens:

1. Code locally using Codex in Cursor.
2. When you finish a feature, tell Codex: "Commit these changes to a new branch and submit a new pull request".
3. The cloud agent automatically reviews your PR for free. You can follow along in the Github app or website.