Desktop Clients
Most clients that support a custom OpenAI provider can connect to Codex Relay. Field labels vary, but the required values are the same.
OpenAI-compatible clients
Use these settings for Cherry Studio, Cline, Roo Code, NextChat, and similar clients:
| Field | Recommended value |
|---|---|
| Provider | OpenAI Compatible / Custom OpenAI |
| API Key | Your Codex Relay key |
| API Base | https://codexapi.asia/v1 |
| Model | An exact ID returned by /v1/models |
| Endpoint | Prefer Responses; use Chat Completions when required |
If the client automatically appends /v1, use https://codexapi.asia instead. The final request path should contain /v1 only once.
Validate a new configuration
- Disable the client's built-in official login so it uses the custom provider.
- Refresh models and compare the list with
/v1/models. - Send a short text-only request first.
- Then test streaming, tools, or image features.
- If it fails, record the HTTP status and request ID, and remove the API key before sharing diagnostics.
The app only shows built-in models
Select a custom model option and enter the exact model ID manually. If the app does not allow custom models at all, use a compatible client instead.