I updated Codex Web GPT. You can now use all your ChatGPT Web models (GPT-6 Pro, GPT-5.6 Pro...) directly inside Codex, with the full harness and their separate Web limits. The launcher now tracks Pro usage too, so you can keep an eye on it as you work. Repo... in replies.show more

Launcher for macOS, Linux and Windows. Includes Zero Risk Mode with zero browser automation: you send messages manually, while the model still gets Codex tools:

Be very careful about this. The last time a developer recklessly took advantage of goodwill gestures from OpenAI, they ended up having to remove the feature because of the fraud it created.

@thsottiaux Ban this guy. He's violating terms of service.

@thsottiaux The only possible ToS violation here is browser automation for sending messages, and there is Zero Risk mode even without this. Therefore, you can use the tool without violate anything, using the quota you've already paid for and Pro Models which are not available in Codex

Too bad unlimited chat is going next devday, enjoy your next week guys

Careful, the Pro on web is much dumber than native codex gpt models and much less context window. The pro prompt limitation is 200 prompts. More than that OAI may flag your account as abuse activity or you could get banned.

Ban hammer 🔨 Comienza la cuenta regresiva… 🕦 Buena idea por cierto?

I used it for a few weeks but Openai change their web DOM and safety policies so the plugin will throw errors on context length, uploads and compression, connection timeouts, and safety blocks of basic git cmds. It will 5x the time of work. Just spend $100 for the time.

web dom was updated ~30 hours ago, and 18 hours later I updated launcher to support new version give it a try, regarding safety policies sometimes recreating plugin helps (more tips in troubleshooting.md)

@futurescheetah You're just going to make them patch seperate lmiits and ruin it for everyone. pay $100 / month and use all the codex you can

@miu21590 It’s sad to lose the unlimited chat. But if openai does so, there is no reason to keep the plus sub anyways.

走网页端 session 调工具确实省钱，但官方封逆向接口的风控风险始终是个坎。

한국어 버전을 만든 당신은 센스가 넘쳐요 깃허브 별 마음만은 별 1

Nuh, I build today one of this for myself, still raw but working

Bro come on this should be kept secret by now, otherwise it will get banned very soon, delete pls

Из-за подобных вещей сперва убрали сессию до конца задачи даже при конце лимита. Теперь вы делаете это. Хотите что бы убрали разные лимиты веб и кодекс? Скорее бы вам забанили все ваши учетки.

wowww I've bookmarked this for later

Tapping into web quota while keeping the full tool harness makes Pro ten times more useful for heavy dev work.

Account ban for violating ToS coming in 3.... 2.... 1....

Nice, does that Pro usage view show separate Web limits too?

Its a plug-in to get access to the normal models not a feature from openai

Putting more models inside Codex makes model choice and usage tracking easier; the constraint shifts to picking the right model for each task and staying within separate limits.

I struck at SETUP: Step 3: Install into Codex: how to do in windows pc? Something went wrong Error invoking remote method 'launcher:setup-core': Error: codex-chatgpt-web: Codex model_provider selects a custom provider; the bridge requires the built-in openai provider. Select mod

do you use opencodex or anything simillar?

I am new to opencodex, still learning how to make it work for codex. I also use omniroute, jan, 9router, lmstudio and ollama

为什么我一直卡在第二步？

update pls

Can you get banned?

Tracking Pro usage is the useful part. Worth logging cache hits next to it, since per-request counts hide what a finished task actually cost

这下子开发效率直接起飞了

This just hang as hell i could never use you live in dreams

That "full harness" part is the real win. Being able to use those heavy-duty models directly inside Codex, without losing the context or limits, feels like a huge unlock for building robust agents.

Does the harness auto-switch models based on task complexity?

Les Usage Policies, en vigueur depuis le 29 octobre 2025, interdisent explicitement le « unsolicited safety testing » et le « circumventing our safeguards ». Ne faites pas ça.

Tracking the separate Pro limits right in the launcher is huge, that quota always seems to vanish the moment you get into a flow.