# Use Radium with Cursor

Connect Radium's Tycho model to Cursor using Cursor's built-in OpenAI-compatible API support. Settings, model ID, verification, and troubleshooting.

## Use Radium with Cursor

Cursor has built-in support for OpenAI-compatible APIs, and Radium serves one, so connecting Tycho is four settings and no plugin.

This page covers Tycho specifically. Any Radium model reachable through the OpenAI-compatible endpoint can be added the same way by repeating step three with a different model identifier.

> **Scope**
> This connects Radium to Cursor Chat and Cursor Agent. Cursor Tab, the inline autocomplete, keeps using Cursor's own model regardless of what you select here. See [About Tab autocomplete](#about-tab-autocomplete).

### Before you start

- Cursor installed
- A Radium API key

### Open model settings

In Cursor, go to:

```text
Cursor Settings → Models
```

### Connect the API

Radium's API is OpenAI-compatible, so the connection goes under the OpenAI section of the Models panel, not Anthropic. Under API Keys, set:

| Setting | Value |
| --- | --- |
| OpenAI API Key | Your Radium API key |
| Override OpenAI Base URL | `https://api.radium.cloud/v1` |

Enable both toggles, **OpenAI API Key** and **Override OpenAI Base URL**. Neither takes effect while off.

> **The one to get right**
> This goes in the OpenAI fields even though you are not using OpenAI. Cursor does not have a generic custom-provider slot, and the OpenAI-compatible one is the correct place for any OpenAI-shaped endpoint, Radium included.

### Add Tycho

Still in the Models panel, select **Add Custom Model** and enter the identifier exactly:

```text
tycho-1.0
```

Enable the model once it is added. It will not appear in the picker otherwise.

### Verify the connection

Open a new Cursor Chat or Agent session and select `tycho-1.0` from the model picker. Send:

```text
Reply with exactly: RADIUM_OK
```

Expected response:

```text
RADIUM_OK
```

If you see that, the connection is live and Tycho is answering Chat and Agent requests.

### About Tab autocomplete

Cursor Tab, the inline completion as you type, continues to run on Cursor's own model no matter what is selected in the Models panel. Radium serves `tycho-1.0` for Chat and Agent sessions only. This is a Cursor limitation, not a Radium one, and there is currently no setting that changes it.

### When it does not work

- Confirm the base URL includes `/v1`. Without it, requests go to the wrong path.
- Confirm the model ID is exactly `tycho-1.0`. Cursor does not correct typos in a custom model entry.
- Confirm the **OpenAI API Key** toggle is enabled, not just filled in.
- Start a new Cursor session after selecting Tycho. An existing session can keep using whatever model it started with.
- Fully quit and restart Cursor if the new settings do not take effect. Cursor does not always pick up a provider change without a restart.

### Next

- [API quickstart](/quickstart), for calling Radium directly
- [Use Radium with Claude Code](/claude-code)
- [Tool calling](/tool-calling)