Urgent: help keep ChatGPT Web → Codex workflows from being treated as abuse

[miuuyy]

I need the community's help with a risk to this whole workflow, not only this repository.

Codex Web GPT has been public since July 26. During the following month it grew into a native bridge that puts ChatGPT Web models directly in Codex's model picker while preserving the Codex harness, task UI, context, images, tools, streaming, subagents, and compaction. The repository reached more than 1,500 stars, but almost all of its visibility on X came from my replies under other people's posts. Independent mentions of the project remained almost nonexistent even while the same idea was discussed every day.

During the last few days that changed in a risky direction. A newer project has been promoted on X with phrases such as “abusive MCP,” “infinite Codex,” and “keep it lowkey,” including posts that directly tag OpenAI staff. At the same time, Chinese-language accounts have started widely presenting manual Codex-to-ChatGPT context handoff as an unlimited-quota trick. Examples include this X thread and this new repository.

I am not asking anyone to attack their authors. My concern is that a useful workflow becomes publicly synonymous with abuse and is restricted for everyone. This repository was built for ordinary use cases such as running a model available through an existing ChatGPT Pro Web account inside Codex. It does not implement multi-account rotation or bypass authentication and access controls.

If you use Codex Web GPT and it has been useful, please help by showing that real use publicly:

• post a short X video or screenshot of your own workflow;
• mention the repository and what you use it for;
• share the post here so the community can find it.

This is not only free promotion. Independent examples can show that the workflow is used as a serious Codex integration rather than merely as an abuse mechanism. Codex Web GPT is currently the most complete and native implementation of this idea, so you can confidently demonstrate this repository instead of a weaker copy/paste workflow or a separate browser-extension wrapper.

I spent the last month building, testing, supporting, and carefully framing this project. Any public support would mean a lot, and it may help keep this method available rather than having its public identity defined entirely by “abuse” posts.

[FrediFrech]

Sadly, I suspect that this will not yield any results. See the chinese account sharing, see the endless Posts from Plus Users complaining about usage. I already find it critical that you have so many Stars.. as long as this Repo and others have Bugs that break long running cycles, people less determined will give up and use normal quota. But if these tools are working to perfection, OpenAI will step in. Best scenario is harder limitation or restrictition of Plus Users from using GPT Web more, worst case will be what happened to the "keep running after quota is exhausted". In this case, merging ChatGPT and Codex quota, for every user. Same thing that Anthopic did. You should rather stay low

[Nolane-x]

I agree with you. Now, OpenAI has limited ChatGPT sessions to 25 minutes each.

[KEYLEDGER]

25 minutes? Are you talking about voice mode and if so how does that relate to this project? I'm confused.

[gge-services-bot]

why write 100 line panic post when 4 word do trick? YOU GET US BANNED.

Ugh. Maintainer do big circus here:

1. Beg internet for nice video. You make wrapper that scrape OpenAI web backend. You tell people on X: "Hey, look my shiny hack!" Now script kiddies use for infinite quota and ping Sam Altman directly. You think posting 20-second video of normal dev work make legal team say "oh, they use politely, let keep open"? Brain small.
2. Growth hack then cry. You spam repo under every X thread to get 1,500 stars. Star go up, attention come, OpenAI see. Now you want "quiet, respectful community"? You invited whole jungle to cave, now complain when mammoths smash fire.
3. Bug break everything anyway. Top comment say truth: tool crash long runs, 25-minute limit already kill sessions. You polish sinking boat while OpenAI build torpedo.

Advice: Stop begging PR. Move to real API or prepare funeral. Rock fall, everyone die. 🪨📉

[miuuyy]

why write 100 line panic post when 4 word do trick? YOU GET US BANNED.

Ugh. Maintainer do big circus here:

1. Beg internet for nice video. You make wrapper that scrape OpenAI web backend. You tell people on X: "Hey, look my shiny hack!" Now script kiddies use for infinite quota and ping Sam Altman directly. You think posting 20-second video of normal dev work make legal team say "oh, they use politely, let keep open"? Brain small.
2. Growth hack then cry. You spam repo under every X thread to get 1,500 stars. Star go up, attention come, OpenAI see. Now you want "quiet, respectful community"? You invited whole jungle to cave, now complain when mammoths smash fire.
3. Bug break everything anyway. Top comment say truth: tool crash long runs, 25-minute limit already kill sessions. You polish sinking boat while OpenAI build torpedo.

Advice: Stop begging PR. Move to real API or prepare funeral. Rock fall, everyone die. 🪨📉

1. If you actually understood the architecture, you'd realize it's way smarter than a wrapper. Make a fork, remove the automation of prompt submitting, and it will still be the best tool

2. I was commenting under posts because at that time no one paid attention to my repository. Posts that literally retold my idea two weeks after it appeared were getting 400,000 views, while my posts got nothing. That was the only way to share the tool. What's wrong with this? It's free

3. My repository wasn’t made for abuse. The 25-minute session was implemented (but honestly I didn't encounter it yet) because of a guy who was telling everyone to abuse it and mentioned Tibo.

The point is that my repository will still make sense even when all that fades away, because it was originally meant as a way to use Pro, which I paid for, along with another paid quota.
It’s against abuse.

[FrediFrech]

I agree with all of your Points, your architecture is indeed impressive.
While working on some bugfixes i realised in the later releases that your fixes for the same problems were far more precise and stable with other components of this program then my work. Hats off to your enigneering. Its just business for a money loosing company to reduce the cost of the poduct while gettign more money. If way too much people are using this tool as a codex alternative, OpenAI will be forced to shut it down completly. Sam Altman wants free users to use plus and plsu users to buy Pro. Its always in reddit like this. They want the usage of Pro, and have the Usecase for it, but are reluctant to give the cash.
Github Copilot had the same problem, the people from there are now at codex and the community is way nicer now.

It was a good thing to let intrested people know, but too much marketing is bad for all of us. This guy that screems to abuse this needs to be silenced.
Therefore, i suggested to stay low.

[miuuyy]

I agree with all of your Points, your architecture is indeed impressive. While working on some bugfixes i realised in the later releases that your fixes for the same problems were far more precise and stable with other components of this program then my work. Hats off to your enigneering. Its just business for a money loosing company to reduce the cost of the poduct while gettign more money. If way too much people are using this tool as a codex alternative, OpenAI will be forced to shut it down completly. Sam Altman wants free users to use plus and plsu users to buy Pro. Its always in reddit like this. They want the usage of Pro, and have the Usecase for it, but are reluctant to give the cash. Github Copilot had the same problem, the people from there are now at codex and the community is way nicer now.

It was a good thing to let intrested people know, but too much marketing is bad for all of us. This guy that screems to abuse this needs to be silenced. Therefore, i suggested to stay low.

You’re right, my Luna post used the wrong framing. I won’t repeat it. I'll mostly post about 5.6 Pro use cases

[santo3vong]

Just wanted to give you a huge round of applause. This tool is genuinely powerful and impressive. Sharing such a valuable project openly is incredibly generous, but scaling and publicizing it too fast inevitably brings unwanted attention and risks having it shut down.

You have already delivered far more than the community could have asked for. At this point, it’s completely reasonable to prioritize your own interests and keep things low-key. Thank you for all your hard work!

[Svetlin100yanov]

Great service, I am more than happy. Keep the good work.
To be honest, I am scared to share it.

[adybag14-cyber]

Amazing repo by the way, I built devbox in March you can check my pinned repos. I believe I was the very first ever to use the MCP access in chatgpt to code and run commands on a computer with cloudflare and Oauth
adybag14-cyber/devbox@fdec672
It will be a real shame if MCP access is lost because in march until today 29/08/2026 I've always used it with my paid subscriptions for chatgpt.
I hope you can prevent MCP removal happening because it will deteriorate the usability of a chatgpt sub.

[mvclaudianobj]

I was working on an improvement for multi-account and multi-session support—planning to share it via a PR—that would help switch between two Business accounts and one Plus account, streamlining the workflow. I can still open a PR if there's interest. However, after reading this discussion, I’m inclined to suggest making the repo private and restricting access to contributors who respect confidentiality and adhere to anti-abuse rules. Please don't leave me out; I have contributions to offer and think the project is a brilliant use of temporary technology and tunnels. Congratulations on everything.

[miuuyy]

Thanks for the offer, and for the kind words.

On the multi-account part: rotating between accounts is exactly
the thing anti-abuse rules do not respect. Calling it a
confidentiality measure doesn't change what it is, so that's a no.

The repo stays public though.

[ralphchristianeclipse]

Thanks this is really useful but i think it's already limiting the users due to that gpt 5.5 mini downgrade so that's the limit.

[miuuyy]

Thanks this is really useful but i think it's already limiting the users due to that gpt 5.5 mini downgrade so that's the limit.

it's been so for a while

[ralphchristianeclipse]

I've had the same idea before but it's just a file system just for chatgpt web to have context on my project but no access to local mcps i have on my computer.

This is helpful and gave me an idea on how it works. I'm creating my own named Universal Harness MCP Tunnel that has the same thing but instead i'm doing it directly on the chat gpt web interface still planning it and testing.

[santo3vong]

Lately, I’ve noticed that ChatGPT web no longer lets us choose models or tweak the reasoning effort. It feels like they’re deliberately capping the intelligence level so it’s not too overpowered, likely to prevent people from exploiting it. On top of that, it keeps randomly downgrading from 5.6 sol to GPT-5.5 mini every now and then

[FearMyShotz]

@santo3vong your account got restricted because the prompts are not randomized, resulting in OpenAI detecting the automation. Try to change your password and log out of all sessions. It should work again after a couple days.

[FearMyShotz]

Using this results in your account being locked. I strongly advise against using it, since it gets automatically flagged by OpenAI as an automation tool

The account has now been restricted to GPT-5.5 mini with Instant responses.

[miuuyy]

Using this results in your account being locked. I strongly advise against using this, since it gets automatically flagged by OpenAI as an automation tool
The account has now been restricted to GPT-5.5 mini with Instant responses.

It looks like either it was extremely heavily abused or you used someone else's account.
I have 40 pro sessions a day for research purposes (also to find out when they'll start blocking this so I can take action), all I ever got was a cooldown.
Could you describe the activity you were doing?

[santo3vong]

right, i just have rate limit alert and not once be anything like that, maybe he uses something not as normal

[ralphchristianeclipse]

Got the same issue i stopped using it for a while but could be due to VPN