Usei o Codex a semana inteira e gastei 2% da cota

Author: Elton Machado · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Usei o Codex a semana inteira e gastei 2% da cota · Channel: Elton Machado · 11:29 · Watch on YouTube

Topics: AI agents, Software engineering with AI, AI models

Summary

The creator explains how he used a community-built project called Codex ChatGPT Web to route Codex requests through a ChatGPT account instead of using as much of his Codex quota. He says this kept his Codex usage around 2–4% for a week, while noting that the ChatGPT account has its own usage limit. The video walks through installing the tool, logging in, adding web models to Codex, and configuring an MCP tunnel and connector so the assistant can access local project files. He finishes by demonstrating the workflow and sharing an optional setting to hide the launcher.

Key points

The creator reports spending about 2–4% of his Codex quota after using it throughout the week.
Codex ChatGPT Web is described as a community-developed project available for Windows, Mac, and Linux.
The tool routes requests to ChatGPT logged into the user’s account, shifting usage away from the Codex quota.
The ChatGPT account still has a usage limit; the cost or limit is moved rather than eliminated.
After installation and login, the user runs a test and installs the web models in Codex.
An MCP tunnel is configured to let the assistant access the user’s computer and project folder.
The API, tunnel, and MCP application must be configured with matching names and verified.
The creator recommends starting a new conversation before using the web options in Codex.
The creator says selecting “zero risk” during setup prevents the launcher from appearing in settings.

Chapters

Start	Chapter	Summary
00:00	How Codex ChatGPT Web shifts usage	The creator introduces the project and explains that it routes Codex requests through a ChatGPT account, while noting that this account also has a limit.
02:33	Installing and testing the web models	He demonstrates installing the application, logging in, running its tests, and adding the web model option to Codex.
05:14	Setting up the MCP tunnel and connector	He configures an outbound tunnel, creates an API, and sets up and verifies an MCP application so the assistant can access local files.
09:23	Using Codex and final settings	The creator demonstrates requests flowing between Codex and GPT Web, recommends starting a new conversation, and explains how to hide the launcher.

Quotes

I just changed where the bill comes from. — 00:00

The bill didn’t disappear, it just moved somewhere else. — 02:01

And this path has a name, it’s called MCP. — 05:45

It’s a little annoying to install, but man, it will definitely be worth it. — 10:25

Concepts and entities

Codex ChatGPT Web, Codex, ChatGPT, MCP, VS Code, Elton Machado

Full transcript (original language)

00:00 I’ve used Codex for the whole week, 7 days, and I spent at most, at most, 3, 4%of my total quota. And it’s not a prompt trick, it’s not some miracle model installed on the PC, and I didn’t stop working for a single day, and I used it a lot, both in running the channels and for help here in the studio. And I just changed where the bill comes from. Codex is still the same Codex on the same screen, doing the same thing. Only now, the one answering is ChatGPT, which you already

00:31 pay for every month. And in the next few minutes, I’ll install all of this from scratch in front of you. The project is called Codex ChatGPT Web. It was developed by the community. Today it already has over 10k stars. So it’s a validated and functional project. And it really works. It is very simple to install. I will show it to you here. It is available for Windows as well as Mac and Linux. You can download the installer directly here or install it

01:01 via the terminal. That’s the part I’m most comfortable with. I like to install it that way. Let me explain what it does, because when you understand all this, the rest becomes obvious. Your project folder is your workbench and Codex is the assistant working at this bench. It reads your files, writes code, and runs commands. That doesn’t change at all, but only one thing changes. Who does this assistant talk to when it needs to think? Normally it asks the Codex models, and that’s when it gets expensive, and that’s when a piece of

01:31 your quota is gone. And this repository installs a program that sits in the middle of the process. It opens a ChatGPT logged into your account in its own browser. Then, when you choose a model like web, Codex delivers your question to this program. It asks there and returns with the answer. Think of a guy who takes your question, walks down the hall to the room where ChatGPT is. He asks and comes back with the answer on paper. You don’t even see him walking, but the question was asked in that room, so the one who paid was that account. And that’s where the savings

02:01 come from. It’s not that the question became cheaper; it’s that it started coming from another account. You have two pots, one is your Codex quota and the other is your ChatGPT limit. It’s the one you already pay for and almost never touch. That’s why my Codex quota stayed at an average of 2 to 4%usage after a whole week. It’s not that I worked less, it’s that almost everything came from another pot. Now here is the part nobody tells you: the second pot also has a bottom. If you hammer away all day, you’ll hit the ChatGPT limit too. The bill didn’t

02:33 disappear, it just moved somewhere else . And it doesn’t matter where you use Codex; if you use it in the terminal, it works. If you use the extension inside VS Code, it also works. It’s the same program and the same steps. And I’m going to show you the simplest way to set it up. As I mentioned, you can download the installer here; if it’s Windows, you’ll click, it will download , just click and install and proceed normally, like installing ChatGPT. Once the download finishes, just click on it to install, because it’s software that wasn’t developed by Microsoft. So, your

03:07 Windows will give you this warning, but you can run it anyway. “Just for me” and proceed. As soon as it finishes, I’ll return; it finished, I’ll click run, or finish, actually. So, as soon as it opens, you’ll choose the language as English. You can click on the first option, keep proceeding here. It asks you to follow his GitHub and open his X. So, we’ll do it, it doesn’t hurt, right? Give him some credit. And this

03:39 is his Twitter, right? In case you want to follow the whole project, everything is on his profile. So here you can close it, it’s already turned green. Open launcher. So the first thing we’ll need to do is log in. We’ll click on open. It will ask you to log in with your ChatGPT account. I’ll do it with Google. So, I confirmed my login. Click allow. Let’s see if the first step is already working. Yes, it has already

04:11 returned the answer. So now we go back to setup again. We’ll click here on run test. It will run the automatic test and returned the answer. So the second step is already green. So now we’ll click here on install models. Look, notice that it’s telling us to close Codex. So, we’ll close it and open it again. I’ll close it completely here and I’ll open it again. If you have Codex open right now, close it and open

04:42 it again. It will all turn green. So, both Codex and GPT Web. Once everything is green, you can open Codex again here . You’ll notice it will have an option here, look, for web. So, we’ve already passed that part. Now let’s check if it’s already working in this web model. So we can select GPT 5.6 Sol here. This one that isn’t checked, it’s also web. So you can use it in ultra and it will

05:14 work too. But I’ll be selecting GPT 6 Sol, extremely high intensity, via message. Look, it already returned the answer too. So we are up and running. Half of the process is already finished . Now we’re going to the next stage. And this stage is the most important one. Most people skip it because it’s tedious, but this is what will actually save your tokens. So, pay close attention, follow the step-by-step; for those of you with little experience, I’m sure it will work out. So here, just to provide context, regarding the

05:45 analogy I made at the beginning, so far the guy in the hallway only brings the text back. He hasn’t touched anything of yours. So, for him to read your folder, edit files, and run commands, he needs a path back to your computer. And this path has a name, it’s called MCP. And that’s what we’re going to set up now. And that’s where the tunnel comes in. It is an outbound tunnel from your computer to the service, without you having to open any router ports, nothing. It’s like a badge that lets GPT enter your workshop instead of just talking from the outside. So now we’re

06:17 going to click here on MCP. You can see that the yellow dot has already moved down here. So here you will click on opens, click on create tunnel, and for the name, you’ll put Codex Native 2. Everything I’m putting here, I’ll leave in a document in the description for you to use. Here under organization, you can leave it marked as personal, as it comes by default. And here in workspace, you will click and select your workspace and click on create. It

06:47 gave a message that the tunnel was created. So you will copy your tunnel ID and paste it into your notepad or WhatsApp so you don’t lose it. Then you’ll go back to Codex Web and click on create APK. Up in the top right corner, click on create a new API. Under name, you’re going to put codex native 2K. For projects, use default; for expiration, set it to never. It will give you this warning, but that’s no problem. And here in permissions, select the middle option, scroll down,

07:17 go to tunnels, select use, it will check all of them and it will be out. Click on create API. You’re going to copy your API and come back here to codex. web, click next, where it says API, you’ll paste your API and tunnel was the one I told you to copy and save . You can go back here to tunnels and copy your ID. So, it has the option to copy. Then you go back to Codex Web and

07:48 paste your ID and click connect. Now you’re going to click here on open chat options. It will open this screen. You’re going to scroll all the way down to developer mode and check these options here. These last three options. Mine are already checked, but you can check them. Make it look just like mine . After you’ve selected developer mode, you’re going to explore plugins here. You’re going to click on plus, create

08:20 app, create MCP application. For the name, we’ll use the same one, Codex Native 2. In the description, we’ll paste local codex access via tunnel. Here in connection, it’s not URL, we’ll click on tunnel. You’re going to select the Codex native 2. For Auth, you don’t need it, put no authentication, check the box and click create. In my case,

08:52 it already appears. Because I already created the connector with the same name. In this part, the name has to be identical, you can’t change it or it won’t connect. In this case, I won’t do it because I already have it. So I would have to go through the whole process again. After you configure everything correctly and click create, you’ll go back here to MCP and click verify. Yours will appear here in the corner, since I’ve already verified, mine is down here, but I’ll click again . It will perform the scan and identify

09:23 it. If everything turns green up here, that’s great, it worked. You’re good to go, it will work. Now, if any part of the process freezes, it was probably a name you entered incorrectly during configuration. Just rewind the video, watch it calmly, step by step, everything I did, and it will work. It turned green, just finish up. Now it’s working perfectly. You don’t need to use it through here. If you want, you can even hide it; you don’t need to use it at all. For me, it doesn’t matter, I

09:53 just leave it open like this. You use it directly through Codex. So now it’s working normally. It will have these web options here. So you can go ahead and use it. Remember to always open a new conversation before using it. Then it will be working normally. What happens, look? You sent it over there in Codex. What’s happening? It is sending the information here to GPT Web . So it sent the information, it will read it and bring back the answer. It will go and return it here to Codex. It returned the message. So, it’s working.

10:25 If you come here and ask it to explore some work of yours, a folder, create a script, it will work. It will first send the information to GPT Web, then it will bring the answer here. It’s a little annoying to install, but man, it will definitely be worth it. It will save you some tokens. Test it out a lot , then leave it here in the comments. I’d like to know if it worked for you too. I use it in my operation, man, and it helps me a lot. It saved a lot of tokens and you can see. I always use it in the, ah, one more piece of information. In case you don’t want the

10:57 launcher, this web, to keep showing up in the settings, instead of selecting the first option, you select zero risk, do the same process, and then the launcher won’t appear. Got it? And that was today’s video. If you have any questions, you can comment down below and I’ll help you. I hope this video helped you in some way, just as it helped me. As I always say, I bring what I use, so I’m using it in my day-to-day, it’s being very useful for me, it’s saving a lot of tokens, especially for my channel operations, constantly making scripts, channel

11:29 analysis. So, man, it’s a huge help, it saved a lot. So much so that I had the most expensive plan on ChatGPT, the one for 500 and something, right? No, not the most expensive one, the middle one. Now I’ve gone back to the basic one, using the 100 one, I think, if I’m not mistaken, and it’s serving me well because it’s saving me a lot of things. So, if this video helped you in any way , and you’d like to support the channel by leaving a comment or sharing it with someone, I’d be very grateful. So, until next time.