NexusAPI supports common AI clients and OpenAI-style API calls. Different clients append different paths automatically, so choose the address from the guide for the tool you are using.
Before you start
Whichever method you use, make sure that you have:
- Created an API key in the console.
- Chosen a token group that includes the target model.
- Copied the exact model ID from the Model Marketplace.
- Confirmed the model’s supported protocol before sending a request.
Choose your entry point
URL and protocol reference
| Scenario | Base URL to enter | Main protocol | Important note |
| --- | --- | --- |
| OpenAI SDK or a general text application | https://nexusapi.link/v1 | Chat Completions | Your SDK normally appends the request path. |
| Codex CLI | https://nexusapi.link/v1 | Responses | Use the provider settings from the Codex guide. |
| Claude Code | https://nexusapi.link | Anthropic Messages | Do not append /v1 again in the Claude Code setting. |
| CC Switch | Depends on the selected app | Depends on the app | Claude and Codex use different URL rules. |
| Image models | https://nexusapi.link/v1 | Images | Use only a model explicitly marked for image generation. |
Before relying on an endpoint, check the API and model capability matrix. It explains what NexusAPI currently documents as an external capability; the Model Marketplace remains the source of truth for a specific key and model.
Pick the protocol first
The same domain does not mean that every model accepts every request format.
Do not use Responses as a universal replacement for ClaudeIf a current Anthropic-type Claude channel returns status_code=500, not implemented through /v1/responses, switch a custom application to /v1/chat/completions, or keep Claude Code on its native Messages configuration. Repeating the same request will not make this protocol mismatch recover.
Why URLs can look different
Most OpenAI-style applications use:
Some clients require the site URL without /v1, or add /chat/completions, /responses, or /messages themselves. Follow the tool-specific guide rather than copying an address from another client.
Avoid duplicate pathsIf a client already appends /v1, entering a URL that includes it can produce /v1/v1/... and a 404 error. Check the actual request URL first when you see a 404.
Confirm that the setup works
- The client returns a complete response.
- The matching request appears in the console usage log.
- The model, token group, and charge match your expectation.
If it fails, keep the error text and Request ID, then use the self-check guide. 