How to Use ChatGPT Codex: Getting Started Safely

You might have heard people say they got a website tweak, a spreadsheet script or a whole small tool built by an AI that works on code for them. Codex is OpenAI's version of that idea. If you run a business, freelance or just like tinkering, it's a good way to see what an AI coding assistant can and can't do.

I'll walk through the ways to start with ChatGPT Codex, what a first task looks like, and the safety settings OpenAI documents. It all comes from OpenAI's developer documentation, so menus may differ on your version.

What ChatGPT Codex is

Codex is OpenAI's coding agent. You describe a job in plain language, and it can read the files in a project, make changes, run commands and show you what it did. OpenAI's documentation presents it in several places, not one single app:

Where you use Codex	About it	Good for
ChatGPT desktop app	Codex is selected from the ChatGPT dropdown; available for macOS, Windows and Linux	Working on projects and local folders
Codex CLI	A command you run in your terminal	People comfortable with the command line
IDE extension	Codex beside the code in your editor	Developers who already use an editor
Codex Cloud	Coding tasks that run in cloud environments connected to your GitHub repositories	Longer jobs running in the background

How to start using ChatGPT Codex in the desktop app

For most non-developers, the desktop app is the gentlest way in. OpenAI's quickstart describes these steps:

1. Install the ChatGPT desktop app for your operating system.
2. Open it and sign in with your ChatGPT account.
3. Choose where to work: start a chat, create a project, or open a folder. The docs say the assistant can read and modify files in the folder you choose, so pick carefully.
4. In the ChatGPT dropdown, select Codex for software development with codebase context and developer tools.
5. Describe your task and any files or context it needs, then send your first message.

The documentation adds that you can also use Codex with an API key, though some features might not be available that way.

How to use the ChatGPT Codex CLI

If you're comfortable in a terminal, the Codex CLI is another route. The steps in OpenAI's docs are:

1. Install Codex using the method for your system. The docs list a standalone installer for macOS and Linux, plus Windows, npm and Homebrew options.
2. Open a project folder in your terminal and run codex.
3. The first time, choose Sign in with ChatGPT or another sign-in method that's offered.
4. Describe what you want, for example asking it to explain the project, make one focused change or help debug a problem.

Inside a session, OpenAI lists commands such as /init to create an instructions file, /status to show your current settings, /permissions to choose what Codex may do, /model to pick a model and reasoning level, and /review to review changes and find issues.

Starting a ChatGPT Codex task in the cloud

Codex Cloud suits jobs you want to run without tying up your own computer. The docs describe this flow:

1. Open ChatGPT on the web, mobile or in the desktop app and sign in.
2. Create or select an environment. A new one starts by choosing your GitHub repositories.
3. Let Codex inspect the repositories, install dependencies and tools, and test the workflow. It will ask for anything missing.
4. Review the setup and test results, then publish the environment.
5. Start a task, then review the changed files and results, ask for follow-up changes, and commit or open a pull request when you're happy.

Each task gets its own workspace, according to OpenAI's docs, and can keep going while your computer is asleep.

What a good first ChatGPT Codex task looks like

Start small. A first job could be asking Codex to explain what a project does, fix one clearly described bug, or tidy a short script. My guide to using AI prompts to speed up coding work covers how to word a request so you get something usable.

• Say exactly what you want changed and where.
• Say what "done" looks like, for example the page loads without an error.
• Ask it to explain its changes in plain English.
• Work in one small area at a time, rather than asking for a rebuild.

The ChatGPT Codex permissions and sandbox OpenAI documents

This is the part to read before you let any AI coding tool loose on real files. OpenAI's security page explains that Codex runs inside a sandbox and asks for approval before certain actions:

• On launch, Codex checks whether your folder is under version control. For those folders it recommends an Auto mode that can edit files in the workspace and asks first for anything outside it or for network access.
• For folders that aren't under version control, it recommends read-only.
• Codex may also start in read-only until you say you trust the folder.
• Some folders inside a workspace, including .git, .agents, .codex and .aws, stay read-only by default.
• A full-access mode with no sandbox and no approvals exists, and OpenAI labels it as not recommended.

OpenAI also warns to be careful when you turn on network access or web search, because a malicious page can contain instructions that an agent might follow. If you don't need the internet for a task, leave it off.

Keeping your work safe when you use ChatGPT Codex

• Use Git, or at least a backup, before you start. OpenAI recommends creating Git checkpoints before and after a task so you can undo changes.
• Review the code or changes before you run or publish them. An AI can sound confident and still be wrong.
• Never paste passwords or API keys into a prompt or a shared file.
• Keep client data and customer lists out of any folder you point it at, unless you've checked your data settings.
• Start in read-only if you only want it to explain something.

How a small business could use ChatGPT Codex

• Making small fixes to a website's code or styling.
• Writing a short script to tidy up a spreadsheet export.
• Explaining what an old piece of code does before you hire someone to change it.
• Drafting a simple internal tool, then having a developer check it.
• Reviewing changes and spotting issues before they go live.

If you'd like a second pair of eyes on which jobs are safe to hand to AI and which need a developer, that's the kind of work I do as an AI consultant.

What people confuse with ChatGPT Codex

• Chatting in ChatGPT versus Codex: a normal chat gives you text to copy. Codex works with the files in your project.
• Codex versus Codex Security: Codex Security is a separate area for scanning code for vulnerabilities, and OpenAI describes its cloud version as a research preview.
• Codex versus the OpenAI API: the API is how developers build AI into their own software. Codex is a ready-made coding assistant.

Frequently asked questions

How do I start using ChatGPT Codex?

Install the ChatGPT desktop app, sign in with your ChatGPT account, choose a project or folder, and select Codex from the dropdown. You can also use the Codex CLI or the cloud route.

Do I need to know how to code to use Codex?

No, you can describe tasks in plain language, but you should still review what it changes. OpenAI recommends Git checkpoints so you can undo work.

Is Codex safe to run on my files?

OpenAI documents a sandbox and approval system. It recommends read-only for folders without version control, and warns to be careful with network access and web search.

What is the difference between the Codex CLI and the desktop app?

The CLI runs in your terminal, while the desktop app gives you a visual workspace with projects and chats. Both sign in with your ChatGPT account.

Can Codex work while my computer is off?

OpenAI says Codex Cloud tasks have their own workspace and can keep working while your computer is asleep.