Quickstart
Use your existing SDK or client
Connect official OpenAI SDKs, desktop clients such as Cherry Studio, or terminal coding tools to Xisu AI by changing only the base_url and API key. Covers whether to include /v1, empty model lists, and clients still calling the official API.
This platform speaks the OpenAI-compatible protocol. Anything that lets you set a custom base_url works as-is — no special SDK, no plugin from us.
Only two settings change
| Setting | What to put there |
|---|---|
base_url / API Base / "endpoint" |
See below |
api_key / API Key / "token" |
The key you created in the console |
https://api.xisu.ai/v1Some clients want only the host, others want the /v1 suffix. Paste the whole string above. If the client appends /v1 itself and the path ends up with it twice, drop that part from what you pasted.
Common snags
- You set the address but it still talks to the official API: most clients need a restart, or a refresh in their model list, before they re-resolve the endpoint.
- The model dropdown is empty: the client is calling
/v1/models. That call also fails when the key is invalid or the balance is negative — work through Error codes and rate limits first. - Tools with the official domain hard-coded: those can't be pointed anywhere else. You need a client that supports a custom endpoint.
What we expose
See Endpoints and authentication. There are no image-generation, audio or video endpoints — if a client needs them, that part of it will not work here.
If you use the Anthropic SDK or Claude Code, you don't need the OpenAI-compatible layer at all — point them at /v1/messages; see The Anthropic-native endpoint.