Codex is OpenAI’s AI agent. It works on your computer for hours while you check in and approve its work from your phone. Here’s the setup, start to finish.
I use Codex and I use Claude Code. These tools keep leapfrogging each other, and I think they always will. What matters is learning to work with AI: how to brief it, give it context and think in systems. Those skills carry over to every tool.
In May 2026, OpenAI made Codex controllable from the ChatGPT app on your phone. Your agent can run for hours on your computer and you can manage it from anywhere, which is perfect if you’re always on the go like me.
01
Check your plan
Codex is included in every ChatGPT plan, even Free. Free is fine for trying it, but you’ll hit limits fast on real work. Plus ($20/month) is where I’d start. (Plans and limits)
02
Install the app
Go to OpenAI’s Codex quickstart and download the ChatGPT desktop app, which has Codex built in. It runs on Mac (Apple Silicon), Windows and Linux. Open it, sign in with your ChatGPT account and choose Codex.
Prefer the terminal? The npm package is @openai/codex. Typing npm i -g codex installs an old, unrelated package. The right command is npm i -g @openai/codex.
03
Give it a project
You can start a chat, create a project or open a folder. I’d make a folder on your Desktop called "Codex" and drop in a file you want help with, like a spreadsheet or a document. Codex works inside that folder and asks before touching anything outside it.
04
Pick your approval mode
This decides how much Codex can do without asking. (OpenAI’s guide)
05
Connect your phone
In the desktop app, go to Settings, then Connections, then Control this Mac or PC, and click Set up. Scan the QR code with your phone and sign in to the ChatGPT app with the same account. Keep your computer awake and online. (Phone setup)
Now, from your phone, you can start tasks, send new instructions, approve commands and review what changed, all while your computer does the work.
Start every new task by having Codex read first, so it understands what’s there before it changes anything. Then paste one of these.
01
Give it standing instructions
Add a file called AGENTS.md to your folder with how you like things done. Codex reads it at the start of every task. It’s like project instructions in Claude. (How AGENTS.md works)
02
Put it on a schedule
In the desktop app, web or phone, you can set up scheduled tasks that run on a timer, like a daily briefing. Tasks that use your computer’s files need it to stay on.
03
Avoid the beginner mistakes
Keep each task specific ("fix the chart on page 2," not "make it better"). Use a new thread for each task. Don’t sit and watch: start it, go do something else and let it check in with you. And don’t jump to Full access just because approvals feel slow. They show you what an agent actually does.
Using Claude too? Here’s how I build my agents in Claude Code.