Client setup
This endpoint speaks the OpenAI protocol, so anything that can talk to OpenAI can talk to this. In every tool below the job is the same: point it at our base URL and paste your key.
Each client also has a page of its own, with the full steps, the gotchas, and how to check it worked: Cline and Roo Code · Open WebUI · Continue · OpenAI SDK · LangChain
These are the standard places each tool keeps its OpenAI-compatible settings. Menus move between versions — if a screen does not match, look for whatever that tool calls "custom endpoint", "base URL override" or "OpenAI-compatible provider", and put the same two values there.
What every client needs
Three values. The same three, everywhere.
| VALUE | WHAT TO PASTE |
|---|---|
| Base URL | — |
| API key | — |
| Model | — |
Some tools want the base URL with /v1 on the end, some add it themselves. If a request comes back as not found, try it the other way round.
OpenAI SDK
The official libraries take a base URL argument. Nothing else in your code changes.
Loading
Loading
Cursor
An AI code editor. It has a built-in slot for a custom OpenAI endpoint.
- Open Settings, then the Models section.
- Paste your key into the OpenAI API Key field.
- Turn on "Override OpenAI Base URL" and paste the base URL.
- Add the model name by hand — it will not appear in the dropdown on its own.
- Click Verify. If it goes green, you are done.
Cursor's own agent features (Composer, Tab) run on its own servers and ignore this setting. Only the chat panel uses your custom endpoint.
Cline and Roo Code
Two popular VS Code extensions that let an AI edit files and run commands. Both ship an OpenAI-compatible provider.
- Open the extension's settings panel.
- Set API Provider to "OpenAI Compatible".
- Base URL: paste ours. API Key: paste yours.
- Model ID: type the model name exactly as listed below.
These agents send the whole file plus a long system prompt on every step, so they burn credits fast. Watch the balance in the console the first few runs.
Continue
An open-source assistant for VS Code and JetBrains. It is configured from a file rather than a settings screen.
Loading
LangChain
The chat model class takes the same two arguments. Everything built on top of it keeps working.
Loading
Open WebUI
A self-hosted chat interface. It accepts any number of OpenAI-compatible endpoints side by side.
- Settings, then Connections.
- Add an OpenAI API connection.
- URL: paste ours. Key: paste yours.
- Save, then pick the model from the model list at the top of a new chat.
When it does not connect
| WHAT YOU SEE | USUAL CAUSE |
|---|---|
| 401, invalid key | The key was copied with a space on the end, or that key was from a browser whose site data has since been cleared. |
| 404, not found | The base URL has /v1 twice, or none at all. Try it the other way. |
| 400, model not allowed | The tool sent its default model name. Type one of ours in by hand. |
| 402, out of credit | Balance ran out. Top it up on the home page. |
| An empty reply | Some tools set a very low output cap. Raise it, or leave that setting blank. |