Feature request: support the existing ChatGPT plan harness on self-hosted/server deployments

The issue names no source files or tests. Start by tracing the described Desktop flow and comparing it with the self-hosted agent-langgraph-agui harness settings, especially PICKED_HARNESS_* and auth-store handling. Done means a supported, documented server setup covers sign-in, secure credential storage, token refresh and re-authentication, upgrades, and credential rotation while preserving built-in Bot behavior.

Summary

OpenBot Desktop can already let built-in Bots such as General Assistant use a signed-in ChatGPT subscription through the existing OpenBot-owned harness, without requiring an OPENAI_API_KEY.

I would like the same capability to be supported for self-hosted/server deployments.

This is not a request to invent a new ChatGPT subscription integration or to treat a consumer subscription as a generic OpenAI API key. The required model path already exists in OpenBot Desktop. The request is to make that existing OpenBot capability configurable and maintainable on a normal self-hosted deployment.

Why this is different from #41 and #349

I found the earlier requests:

• #41 — ChatGPT subscription OAuth for model access
• #349 — API-only LLM access makes OpenBot unnecessarily limited

Those were closed on the basis that a generic consumer-subscription OAuth path should not be added to the server/template.

Since then, OpenBot itself has implemented and merged the relevant functionality for Desktop:

• #451 added end-to-end model-provider setup, including a signed-in ChatGPT plan backed by the Codex model/harness.
• #626 made built-in Bots use the selected ChatGPT or Claude subscription through the owned harness while retaining their existing prompts, permissions and server tools.

So this request is narrower and materially different:

Please expose the existing OpenBot Desktop subscription-capable harness as a supported option for self-hosted/server deployments.

The model integration already exists in OpenBot. The missing part on the server is the supported setup and lifecycle management.

Current behaviour

OpenBot Desktop

A user can effectively do:

OpenBot Desktop
  -> Sign in with ChatGPT
  -> OpenBot-managed ChatGPT/Codex auth store
  -> OpenBot agent-langgraph-agui harness
  -> Built-in Bots such as General Assistant

No separate OpenAI API key is required for that model path.

Self-hosted/server OpenBot

A normal self-hosted deployment currently expects model access through API keys or compatible endpoints.

The existing Desktop-managed plan settings/harness are not exposed as a supported server configuration flow.

Requested server capability

Please support something equivalent to the Desktop path for a self-hosted deployment, for example:

Self-hosted OpenBot
  -> Sign in with ChatGPT / provision ChatGPT auth store
  -> official OpenBot agent-langgraph-agui harness
  -> built-in OpenBot Bots
  -> user's eligible ChatGPT/Codex plan

The exact UX does not need to match Desktop. A server-friendly setup command or documented configuration flow would be sufficient.

For example, a supported flow could:

1. Perform the ChatGPT/Codex sign-in.
2. Store the refreshable auth store securely on the server.
3. Start the official agent-langgraph-agui harness.
4. Mount/provide the auth store to that harness.
5. Configure the required PICKED_HARNESS_* / harness endpoint settings.
6. Preserve token refresh and re-auth behaviour.
7. Keep the built-in Bot prompts, OpenBot tools, permissions, policy and audit path exactly as they work with the Desktop-owned harness.
8. Document upgrade and credential-rotation behaviour for self-hosted deployments.

Why this matters

A large portion of self-hosted users are likely to already pay for ChatGPT and would naturally prefer to use their existing eligible subscription/plan where OpenBot already supports that model path, instead of maintaining a separate API account and metered API spend just because the deployment runs on a server rather than a desktop.

It is also surprising from an operator's point of view that:

OpenBot Desktop + built-in Bots + ChatGPT plan

is supported, while:

OpenBot on a self-hosted Linux server + the same built-in Bots + the same OpenBot harness

is not a supported configuration.

Because the harness and Bot integration already exist, this appears primarily to be a packaging/setup/lifecycle gap rather than a request for a new inference architecture.

Desired outcome

A documented, supported way to run the same ChatGPT-plan-backed built-in Bot architecture used by OpenBot Desktop on a self-hosted Linux/server deployment, without maintaining a custom OpenBot fork.

That would keep self-hosted deployments stock and upgradeable while giving them feature parity with Desktop for model-plan sign-in.