For Windows users, enabling this feature is actually quite easy. 1. Update Codex on PC to the latest version. You’ll then see the Codex mobile version, and configure it as shown in the image.

These Codex features – you enabled them for nothing

Codex has over twenty hidden features, but most people only use half after installation. This article sorts them by frequency of use, putting the most valuable ones first.

Not knowing these Codex features means you’ve installed it for nothing

I recently watched a complete Codex tutorial video with 13 chapters, from installation to plugins – a long, sprawling epic.

After watching, my biggest takeaway is this: 80% of this tool’s power is hidden in menus most people never open.

It’s not that the tool isn’t good – it’s that you’re not using it the right way.

Below I’ve reorganized the video content by actual frequency of use. This isn’t a from-zero-to-one tutorial; it’s a from worthwhile to not worthwhile filter.

1. Don’t enable Full Access for permissions – one wrong step and everything is lost

This is the part I most want to get to early.

Many people’s first step after installing Codex is to log in, select the work type, and enter the main interface. That looks fine.

But there are three permission options below the input box that many people don’t even look at before clicking “Next.”

These three options determine how smoothly you can use Codex – and how much trouble you can get into.

I always use Auto review. Day-to-day coding and file modifications pass through automatically; only truly dangerous operations (like deleting system files) trigger prompts. Both efficiency and safety are achieved.

Full Access looks convenient, but if something is accidentally deleted, there’s no undo.

2. Model and speed: for simple tasks, don’t set the highest reasoning depth

Next to the input box, the current model and reasoning depth are displayed, e.g., 5.5 medium.

Two parameters can be adjusted:

Reasoning depth: Low, medium, high, extra high. Higher is slower and consumes more tokens, but code quality is indeed better.

Output speed has two gears:

• Standard: standard speed, normal token consumption
• Fast: 1.5 times speed increase, but token consumption doubles – under GPT-5.5, Fast mode consumes twice the tokens of Standard mode

My usage: for modifying a style or writing a small function, Low + Fast is fine; for refactoring architecture or writing a complete module, High + Standard ensures quality.

Additionally, in the model list, besides GPT-5.5, you can also select 5.4 or others. Don’t blindly trust the newest – for simple tasks, using an older model saves money.

3. Annotate: screenshot and modify UI – 10x faster than typing

This feature is particularly clever – you circle a button in the preview area, type “change to blue,” and it directly modifies it.

The operation flow is simple:

• After Codex generates the code, click the 【Annotate】 icon in the preview area
• Circle the selected UI element
• Enter your modification suggestion, e.g., “remove,” “make bigger,” “change color to blue”
• Click send, and Codex automatically modifies the code based on your request

The time saved isn’t in modifying the UI itself, but in not having to describe “the button is too small, the color is wrong, the spacing is too narrow” in words.

But there’s one pitfall: The Codex preview area is not exactly the same as a real browser. Features like localStorage may not work in the preview area.

I’ve encountered a situation: “Clicking the plus sign has no effect” on a page generated by Codex. By closing the preview and opening the same page in an independent browser, everything worked fine.

Conclusion: External applications should be tested in an independent browser; don’t treat the preview area as a production environment.

4. Git: two paths – manual and automatic, each has its strengths

Codex has built-in Git support with two operation methods:

Method 1: Terminal commands (precise and universal) Use the shortcut Command + J to open the terminal panel and run familiar commands:

git init .
git add .
git commit -m "Initial commit"

Pros: accurate, universal, you have full control over every line.

Method 2: Built-in Git (fast commit) Click 【Environment Info】 → 【Changes】 in the left sidebar → view unstaged changes → click the plus sign to enter modification suggestion → click the commit button to fill in a message → continue.

Of course, if you think something is wrong with a change, click the + sign, make your request, and let Codex modify it.

What scenarios is it suitable for? You changed a style or added a small function, and you’re too lazy to write a commit message – just click commit, and it’s done in three seconds.

Note: Changes to the AGENTS.md file are not automatically committed as Git commits; you need to manually run the command once.

5. Session management: Fork and Archive – don’t wait until the code is a mess to use them

Fork: one fish, multiple dishes – open two paths simultaneously

When you want Codex to handle two different requests at the same time without interfering with each other’s code, use Fork.

Fork has two options:

Important: Fork does not automatically roll back code. It only copies the conversation history; the code state does not change.

When using local mode, if you need to roll back the code to the starting point of the fork, manually run git reset --hard

Archive: not deletion, just storage

Click the icon next to the session, and it’s archived.

Archiving is not deletion. It only removes the session from the main list; you can still find it under 【Archived Sessions】 in the settings panel.

To completely delete a session, click delete here in the archive, and the session is truly deleted!

6. Plan Mode: not planning for large projects is like running blindfolded

Click the plus sign in the input box → select Plan Mode, and Codex will first make a plan, then take action.

For example, if you want to convert an HTML web page into a desktop application (like Tauri or Electron), and also introduce React and TypeScript – for projects of this scale, directly asking it to write code will likely lead to going off course halfway through.

After enabling Plan Mode, Codex will first ask a few key questions based on the project situation:

• How should data be stored? File or database?
• What is the delivery standard? Just runnable or production-grade?
• What is the testing plan? …

It builds the framework first, you confirm the direction, then it writes the code.

What you save is the time spent tearing everything down and starting over.

There is also a supporting feature called Side Chat: while Codex is executing the main task, you can ask lightweight questions in Side Chat without interrupting the main task’s progress. For example, while waiting for it to finish building, you can ask about a configuration parameter.

You can also directly ask questions in the Codex session about something you don’t understand.

The Steer function is also worth mentioning: Codex by default waits for one request to finish before accepting the next. If you have urgent supplementary information that Codex needs to know immediately, you can use the Steer function with the shortcut Ctrl+Enter. However, the default is queuing; you can modify the default behavior in the settings.

This way, if you suddenly remember a key point you forgot to describe during the conversation, you can directly type it and send it with Ctrl+Enter. Codex will incorporate the newly entered information into the current session and continue.

7. AGENTS.md: cross-session persistence – this is Codex’s real trump card

AGENTS.md is a configuration file placed in the project root directory.

Every time Codex starts a new session, it automatically reads the instructions in this file and executes them.

What does this mean?

It means you write the rules once, and Codex automatically follows them every time.

For example:

• “After completing each code modification, automatically make a Git commit”
• “Use camelCase for code naming; variable names must be self-explanatory”
• “The project must use TypeScript 3.0+, React 18+”
• “All API requests must include error handling and retry logic”

This is the key to making Codex your dedicated assistant across sessions.

Without it, you’d have to repeat the project background and specifications every time you start a new conversation.

Creating it is simple: If you only want it to take effect in the project, create a new AGENTS.md file in the project root directory, write your rules, then start a new Codex session, and it’s already active.

The better you write it, the deeper Codex understands your project, and the more accurate the results. This is a feature that gives you continuous benefits from a single investment.

8. Plugin ecosystem: five core plugins – enable them and you win

Codex’s capabilities are extended entirely through plugins.

In the sidebar, click Plugins. Checkmarks for installed ones, plus signs for uninstalled – click to install.

Here are a few most worth installing:

Image Gen (killer feature) Specifically for generating images. Supports text-to-image, upload reference images to generate variations.

Usage: type @Image Gen at the very beginning of the task, then describe the image you want. If you need a reference to the current project’s UI, click + and select 【Add Photos and Files】

Codex will generate the image according to your requirements.

Chrome Plugin After installation, Codex can control your Chrome browser.

For example: “Let Codex query the usage of the Agnes image generation API” Codex will automatically open a page in your Chrome, open the specified URL, and browse the page.

Of course, you can also ask Codex to do more complex things, like query the latest products on Product Hunt and summarize them.

Computer Use This is even more powerful – Codex uses an independent virtual mouse to operate your computer, completely separate from your own mouse.

Installation and usage are the same as Chrome.

You can ask it: “Help me create a new calendar event for June 68 at 10:00 AM with the title ‘Product Development Discussion’” It will silently do it in the background while you continue with other tasks.

Skill (Custom skills) In the built-in Skills list, you can see all the skills provided by installed plugins.

Even more powerful is you can create custom Skills. For example, let Codex write an article rewriting Skill:

“Write an article rewriting Skill specifically for rewriting existing markdown articles. Rules are as follows: [your rules]” Codex will generate a Skill named e.g., rewrite. Later, when starting a new conversation, reference it with /rewrite, and Codex will rewrite the provided markdown article according to your rules.

9. Automation: Let Codex do tasks for you on a schedule

You can directly add automation in the session.

Click the three dots above the session → add automation, and you can create a scheduled task.

Or directly select 【Automation】 from the left menu, then create via chat or according to your own needs.

Configuration items are few:

After creation, you can see your scheduled task under 【Automation】 in the left sidebar.

Click the button to run a test; Codex will create a new session to execute it.

What scenarios is it suitable for? Periodically scan project code quality, automatically generate weekly reports, check a website for updates every day.

10. Mobile version: control your computer from your phone

Codex has a mobile version that supports remote control of your computer via your phone.

Setup flow: Scan the QR code on your phone → navigate to ChatGPT’s Codex page → allow the phone to control the computer.

Once set up, you can issue commands on your phone: “Delete the schedule item ‘Discuss project design’ on May 28th from the calendar.” The Codex on your computer will receive the command and execute it. No need to open your computer, no need to open your browser – just tap on your phone.

Specific setup details here:

AFei Liang @afly813 · May 30 For Windows users, enabling this feature is actually quite easy. 1. Update Codex on PC to the latest version. You’ll then see the Codex mobile version, and configure it as shown in the image.

Quote AFei Liang @afly813 · May 30 Huge update from OpenAI! Codex’s Computer Use now officially supports Windows! Now you can have Codex directly control your desktop, operate various apps, and autonomously complete tasks on your Windows computer without staring at the screen. Even cooler: use the ChatGPT mobile app to remotely start tasks, monitor progress in real time, and give new commands anytime! Commands from your phone, executed on Windows – Mac

x.com/OpenAI/status/…1176

Tools aren’t valuable; how you use them is valuable

After all that, it’s just one sentence:

Codex is not a tool you just install and use; it’s an assistant that needs to be “trained.”

With the right permissions, your efficiency doubles. With a well-written AGENTS.md, every conversation saves half the time. With all the plugins installed, one tool does the work of three people.

Tools save you time. Where you spend that saved time is where the difference between you and others lies.