Hello OpenAI Codex team,

I maintain agent, an Apache-2.0 open-source personal coding agent. The public repository is named giovannijecha/agent on GitHub.

I am requesting written authorization and, if available, an agent-specific native OAuth public-client registration so users can voluntarily sign in with ChatGPT and use an eligible Plus or Pro Codex subscription directly from agent.

agent is a local Node.js command-line application with no backend, telemetry, embedded browser, or password collection. Authentication would remain on an OpenAI-hosted browser page. The client cannot keep a secret, so the preferred flow is device authorization; authorization code with PKCE and a loopback redirect is an acceptable fallback. Provider credentials would initially be memory-only.

The project will not bundle or invoke Codex CLI, the Codex SDK, or Codex App Server; reuse another application’s client identity; read another product’s credential files; or claim to be an OpenAI client. All protocol code would be independently implemented from a provider-published contract and identify the application truthfully as agent.

Please confirm whether this use is authorized and either provide the official registration process or state that no direct independent-client route is available. If it is available, please identify the documented authorization, refresh, revocation, entitlement, model transport, rate-limit, branding, and open-source distribution requirements. A written refusal is also useful because the repository fails closed until authorization is complete.

Thank you,
Giovanni Jecha
Maintainer, agent

Public project materials are available in the GitHub repository giovannijecha/agent:

• docs/OAUTH-REGISTRATION.md
• PRIVACY.md
• SECURITY.md

Could you consider editing your post so there aren’t so many unnecessary carriage returns? Looks like you’ve just cut and paste that from ChatGPT or similar and looks sloppy.

PS Welcome to the forum.

Thanks for pointing that out. I’ve cleaned up the formatting.

FYI

I struggled with this and I’d say use at your own risk.

I added a disclaimer to my bespoke coding agent:

Minimalist Go based CLI coding agent a la mode

See discussion linked here: Yesterday I wrote a new coding agent (CodeGoLLM) - #4 by merefield

I’d say using it outside of a coding-specific agent might be even more risky …