Point Cursor chat at the chat door.

1. Open Cursor Settings, then Models.
2. OpenAI API Key: rea_YOUR_REA_KEY.sk-YOUR_OPENAI_KEY. Turn the OpenAI key on. A saved key with the switch off is not sent.
3. Turn on Override OpenAI Base URL. Set it to https://api.reacontrolplane.com/v1. No trailing slash after v1.
4. Add a custom model with the exact id, such as gpt-4o-mini. Select that model in chat. Do not reuse a built-in name. A built-in name stays on Cursor.

This field holds one vendor key. The part after the dot has to match the model. gpt-* needs the OpenAI key. Put an xAI key there only if the custom model id starts with grok-. One field cannot carry both.

Tab completion does not use this URL. It stays on Cursor. Claude and Gemini names are not sent here either. Those use Cursor's own Anthropic and Google keys, and those fields have no base URL. With the OpenAI key on, Cursor also sends its other non-Claude models to this URL, and it may reject them. Turn the OpenAI key off to use those again. The base URL stays saved.

Cursor builds the prompt on its servers, then sends the call. REA sees it only after that.

If Models says the account is not on that plan, Cursor will not show these fields. REA cannot turn them on. Until it does, Cline inside Cursor uses this same chat path.

LoopGuard, PreFlight, and TraceGuard run on this path. They are included on every plan. The model table is on Set up. You bring your own model key. REA does not host the model.