Feature request: ChatGPT subscription OAuth for model access

[heavensnight0729]

Question

Is there a plan to support signing in with a ChatGPT subscription via OAuth and using that authorization for OpenBot model access, instead of requiring an OpenAI API key?

I understand that Google OAuth currently authenticates users into OpenBot, and that the default model integration uses an OpenAI API key. I am specifically asking whether ChatGPT subscription OAuth could be supported for model usage in a future release, or whether this is intentionally out of scope.

If it is not planned, is there a recommended OAuth-based alternative, such as a custom AG-UI agent or a supported provider gateway?

[zopeVaibhav]

The buildable version of this is probably worth splitting from the ChatGPT-sub framing.

On the sub itself: OpenAI doesn't expose an OAuth flow that trades a ChatGPT subscription (Plus, Team, Pro, Enterprise) for API access. The sub covers the web and app products; API usage is billed and keyed separately. So the literal ask can't be built on OpenBot's side without something OpenAI hasn't shipped.

The part that is buildable today: any AG-UI endpoint is a Bot, and any OpenAI-compatible endpoint fits OPENAI_BASE_URL. Either of these gets you OAuth without exposing a raw API key in .env:

• Custom AG-UI Bot that does OAuth against your IdP and holds the model credential itself. (docs/coworkers.md)
• OpenAI-compatible gateway (LiteLLM, OpenRouter, internal proxy) that terminates OAuth server-side and hands OpenBot a short-lived token. Configure via OPENAI_BASE_URL and BOT_MODEL.

Worth reframing the issue as: OAuth-authenticated model access via a configurable IdP or gateway. That version is scoped, doesn't depend on OpenAI shipping something new, and covers the real ask if the concern is "don't want a static API key sitting in .env."

If the concern is actually cost (reuse the sub I already pay for), that's a different conversation and worth calling out explicitly so maintainers can respond to the right question.