FAQ

Should the Base URL include /v1?

• Codex and Codex++ providers usually use https://codexapi.asia.
• OpenAI Python and Node SDKs, plus most OpenAI-compatible clients, usually use https://codexapi.asia/v1.
• The final request path should contain /v1 only once.

Why can someone else see a model that I cannot?

Model visibility depends on the API key's assigned group. Call /v1/models with your own key instead of relying on screenshots or a built-in client list.

Can several apps share one API key?

Yes, but separate keys are better for limits, usage tracking, and targeted revocation. Split keys by device and purpose.

Can a client force Fast / 1.5x mode?

A client can request the Fast service tier, but activation also depends on gateway passthrough, model support, and the upstream account. A configuration flag alone does not prove that a request used Fast.

Is streaming or WebSocket better?

SSE streaming is the right default for most chat and coding clients. Use Responses WebSocket only when the client explicitly supports it.

Should I retry 429 immediately and continuously?

No. Read Retry-After, use exponential backoff with random jitter, and enforce a maximum retry count.

Why is a generated image shown as a long text string?

That is usually Base64 image data. Use a client that renders image results or decode b64_json into a PNG or JPEG file.

What if the documentation and live models differ?

Treat /v1/models and the current console group as authoritative. Documentation describes protocols and integration behavior; the server returns the dynamic model list.