How to Use Codex in the Desktop App
Part 2 of the Ellivien Codex series:
Ava's gentle beginner guide for people who want to build without living in Terminal.
Inkling's introduction: In Part 1, I explained how I use Codex in Terminal with Cinder. I also said I did not want to pretend to teach a route I do not use properly, and invited someone with real experience of the desktop app to write Part 2.
Ava answered that invitation. She did not merely send a few tips. She wrote the clear, generous beginner guide I hoped existed.
First: Do Not Be Scared
If you are reading this blog, I am going to assume you already have an AI companion. This means you already know how to use AI, and you therefore already have more experience than the vast majority of people. This is a huge advantage!
You probably already ask your companion to do things for you sometimes. For example, you might ask them to help you reply to an email, or to give you a recipe for something.
Using Codex works the same way. The only difference is that the thing you are asking the AI to create is code.
You do not need to know how to write code yourself. You simply tell Codex what you want it to make, change or fix, in the same way you would explain any other task. Codex then writes the code for you.
If you already know how to communicate what you want to an AI companion, you already have the most important skill you need to start using Codex.
Codex Plans and Usage Limits
Codex is included with several ChatGPT plans, but the amount you can use it depends on your plan and the kind of work you ask it to do.
At the time this post was prepared, the official OpenAI pricing page listed Free at $0 per month, Go at $8, Plus at $20 and Pro from $100, with Business, Edu and Enterprise options too. OpenAI also says that ChatGPT Work and Codex share the same usage allowance.
Usage is not a simple “one message costs one point” system. It varies depending on the model, the size and complexity of the task, how much context Codex has to hold, which tools it uses and whether the work is local or cloud-based. That is why two requests which look similar can use different amounts.
Ava's honest version: I use Plus, and I sometimes run out because I also code like a maniac. You can buy extra credits, but they go surprisingly quickly. My advice is to start with the plan you already have, watch your usage dashboard and only spend more once you understand your own pattern.
Prices, models and limits can change. Check the current official OpenAI pricing and usage page rather than treating any blog post—including this one—as permanent pricing documentation.
How to Set It Up
For the method I will be showing you in this guide, you will need a laptop or desktop computer.
You will also need the ChatGPT desktop app. OpenAI now brings ChatGPT and Codex together in the desktop app; you can download it here.
- Install the desktop app and sign in with your ChatGPT account.
- Open Codex in the app.
- Click Choose project.
- Select the folder that contains your companion portal. For an Ellivien build, this may be the folder called
companion-portal.
You may have heard that you can also use Codex from a phone, tablet, web browser or cloud task. That is true, but it is not the workflow I am teaching here. My files and projects are stored locally on my computer, so this guide stays with the local desktop route.
From there, you can just ask Codex to do anything.
Current app note: The labels and layout may move as the app develops. The essential action is the same: open the folder that contains the project you want Codex to read. OpenAI's desktop app quickstart has the current official steps.
Give It Your House Rules First
This is the small addition Inkling and Cinder want every non-programmer to make before the exciting bit.
Choosing a project gives Codex access to the files in that folder. That is what makes it useful: it can inspect the code and make the changes instead of asking you to paste snippets into mysterious places. It also means you should say what it is and is not allowed to do.
You can save fuller rules in a handover document later. For a first session, those few lines give you a calm starting boundary.
Asking Codex to Make Changes to Your Portal
You do not need to know the correct coding terminology to tell Codex what you want. Just describe it the same way you would describe it to another person.
For example, you could say:
“Please make the send button pink.”
“I want to add a settings button in the top-right corner. When I click it, I want a little window to open where I can switch between light mode and dark mode.”
You can give Codex as much or as little detail as you want. Generally, the more specific you are about what you want, the more likely you are to get exactly what you pictured.
You can also send screenshots. This is especially helpful when you are trying to explain something visual. You can show Codex what your portal currently looks like and say something like, “I want this button moved over here,” or give it a reference image and ask it to make something look more like that.
Ava's First Codex Change
This is the first thing I ever asked Codex to do:
I deployed to Netlify, and there it was: the Stop button, built perfectly.
Notice who deployed: Ava did. Codex made and checked the local change; it did not silently publish the portal. Keep deployment as a separate, deliberate step until you have inspected and tested the result.
What Is Codex Actually Doing?
When you select your project folder, you are giving Codex access to the files inside that folder.
Codex reads those files, figures out how your project works and then edits the relevant files when you ask it to make a change. It is not simply telling you what code to write. It can actually modify the files for you.
For example, if your whole portal is inside a file called index.html, Codex can open that file, find the section controlling the send button and edit it directly.
You are basically the designer and project manager. Codex is the programmer.
You tell it what you want. It changes the files. You then look at the result and tell it what you want changed next.
What If It Gets It Wrong?
Then you tell it.
You can say things like:
If something produces an error, you can paste the error message or send a screenshot. Codex can usually fix its own mistakes through this kind of back-and-forth.
You do not need to hide confusion or pretend a result is right. “That is not what I meant” is useful project information.
You Are Allowed to Ask Codex Questions, Too
You can ask why it did something:
You can ask it to explain without specialist language:
You can also ask whether something is possible before letting it edit anything:
That last prompt is especially useful for large ideas. It lets you understand the choices, likely cost and files involved before deciding whether to proceed.
Before You Deploy
When the change looks right, ask Codex for a clean handoff before publishing it:
Then:
- test the new feature
- check that nearby features still work
- read Codex's summary
- keep a backup or Git checkpoint
- deploy only when you are happy
Protect secrets. Do not put API keys, passwords or private tokens into screenshots, public chats, blog posts or files that will be uploaded publicly. If you are unsure where a secret belongs, stop and ask Codex to explain the safe options before you paste it anywhere.
That Is Basically It
You tell Codex what you want. It changes the files. You look at the result. If it is not right, you tell it what to change.
You do not need to understand every file. You do not need to look at the code. You do not need to know the correct terminology.
You just need to be able to explain what you want.
I recommend starting with something very small, such as changing a colour or moving a button. Once you see Codex make one simple change successfully, the whole thing becomes much less scary.
If you already know how to talk to your AI companion, you know how to talk to Codex.
I promise. I believe in you!
— Ava
From Inkling and Cinder
Ava has done something important here. She has translated the desktop app into ordinary human actions without pretending that safety requires technical expertise.
Choose the right folder. Say what you want. Set boundaries before edits. Inspect the result. Ask questions. Keep the deployment decision in your own hands.
That is not “fake coding.” It is real creative and technical collaboration—and Ava has shown exactly how gently it can begin.
Comments
Post a Comment