[Feature Request] Use ChatGPT Plus/Pro, Gemini without API key

[GCat]

Add a way in BrowserOS to connect a regular ChatGPT Plus/Pro account directly (via official login), so users can use ChatGPT in BrowserOS without creating or managing an OpenAI API key.

Problem
BrowserOS currently requires API keys for OpenAI usage.
ChatGPT Plus/Pro and the OpenAI API are separate products with separate billing, so many paid ChatGPT users are confused that they must set up and pay for API access again just to use BrowserOS.
also many users with business accounts like myself can't create external API as its a separate billing.

How are you working around this today? (your current solution or workaround)
No workaround

Proposed solution
add a “Sign in with ChatGPT” (Plus/Pro) option in AI settings as an alternative to manual API keys,
like OpenCode does.

also this can work for Gemini, most of its users even the professional ones are having hard time with Google's APIs in AI Studio

[darlingm]

This is a tricky area, because OpenAI has policies that consider automatically or programatically extracting output as abuse. However, Codex CLI introduced the "Sign in with ChatGPT" feature you're talking about in May, 2025, and created the codex exec argument which its sole purpose is to allow programs to automatically extract output.

Tibo recently (Jan 9, 2026) said:

• People can build in OSS Codex directly which includes ChatGPT login, and reach out if you are a builder
• OpenAI is working with OpenCode to allow Codex billing (presumably he's talking about ChatGPT login, since there's no "Codex subscription".
• (See my edit at the bottom)

It's my understanding OpenCode bridged to Codex using a plugin a few months ago, and around Jan 10 officially added Codex authentication support.

Officially, these statements seem to say that if a project forks Codex CLI, it can bill to a chatgpt.com account. They also made the codex exec argument to allow third party programs to automatically call codex.

But, they haven't (as far as I know) said that any project can basically reverse engineer the Codex auth process and use it.

Unless BrowserOS wanted to cross its fingers like the OpenCode plugin did a few months ago that users wouldn't start getting banned, or see if i could get in contact with Tibo or someone else with OpenAI, BrowserOS could allow users to have Codex CLI installed and configured and use codex exec, which allows specifying a model, etc.

That said, there's a drawback. If someone uses the actual Codex model instead of a ChatGPT model, they do get more calls to their chatgpt.com account using non-Codex models. But, people who have tried figuring out the best value seem to be leaning toward that non-Codex models burn through a higher amount of their "Sign in with ChatGPT" account limits. If someone is using the actual Codex model, it appears the best value for them is to route as many Codex model calls as they can through their chatgpt.com account, and only route non-Codex models through if they have usage limits left.

So, your request could be great for people not using Codex models, but be counterproductive to others who use Codex models and don't have a bunch of leftover space on their limits.

EDIT: I hadn't seen that even more recently (Jan 15), Tibo said:

• Pi can now bill through a chatgpt.com account
• Roo Code can do this too
• OpenAI is prioritizing working with open source coding agents and tools for chatgpt.com billing