> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prisminference.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Cursor

> Point Cursor's OpenAI-compatible provider at Prism with a custom base URL.

Cursor exposes an **Override OpenAI Base URL** setting, so you can route its
chat and agent requests to Prism. Cursor sends these requests through its own
backend for prompt building, so this is a bring-your-own-key setup rather than a
direct connection.

If you have no key, ask the workspace owner for an existing `PRISM_API_KEY`.

## Configure

**1. Open model settings**

Open **Cursor Settings** (`Cmd + ,` on macOS, `Ctrl + ,` elsewhere) and go to
**Models**.

**2. Set the OpenAI API key**

In the **API Keys** section, turn on **OpenAI API Key** and paste your Prism API
key. Cursor sends this key to whichever base URL you configure, even though it
is labeled OpenAI.

**3. Override the base URL**

Turn on **Override OpenAI Base URL** and set it to:

```text theme={"theme":{"light":"github-light","dark":"github-dark"}}
https://api.prisminference.com/v1
```

**4. Add the Prism models**

In the **Models** section, click **+ Add model** and add each Prism model ID:

* `deepseek-v4.1-flash`
* `gemma-4-31b`

Make sure both are toggled on. Send the bare model ID on the wire; do not prefix
it with a provider name.

**5. Select a model**

Open the model picker in chat or the agent panel and select
`deepseek-v4.1-flash`.

## Notes

* The override applies to every model that uses your key, so turn it off to
  switch back to Cursor's hosted models.
* Tab completion always uses Cursor's built-in models and does not use Prism.
* Avoid custom names that resemble a native Cursor model; Cursor may route those
  back through its own servers.
* Prism discards request inputs and outputs after processing. See
  [Zero data retention](https://docs.prisminference.com/zero-data-retention.md).

## Verify

Send a short prompt in a new chat and confirm the response arrives. The request
should appear in your Prism usage once it completes.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.