Use Kimchi Router in Cursor | Kimchi Inference

This guide shows you how to use the Kimchi Model Router in Cursor. Tested with Cursor 3.22.12.

📘
Cursor ships its own Auto mode that picks among Cursor's models. The Kimchi Router is different: it is the kimchi/auto custom model you add in this guide, it routes among the Kimchi serverless catalog, and it bills through your Kimchi API key.

Prerequisites

A Cursor Pro subscription or higher. The free plan does not allow custom API keys or model changes.
A Kimchi API key from app.kimchi.dev/settings.

Configure Cursor

kimchi setup-tools does not add the kimchi/auto model to Cursor yet, so add it manually.

Add the Kimchi gateway

Open Cursor Settings and go to Models. Scroll down to API Keys and set:

OpenAI API Key: your Kimchi API key, with the toggle ON
Override OpenAI Base URL: ON, with https://llm.kimchi.dev/openai/v1

Add the auto model

Click View All Models, scroll down, and click Add Custom Model. Enter kimchi/auto as the name and click Add.

The kimchi/ prefix keeps the model separate from Cursor's built-in models with similar names.

Find kimchi/auto on the model list and make sure it is enabled.

Select the Router

Open the chat window and pick kimchi/auto in the model selector. Send a test prompt, such as Hello. If you get a response, the Router is working.

To see which models the Router picked, see Router analytics.

Supported features

Chat window
IDE window: inline edits and Agent mode
Image support

Limitations and known issues

⚠️
Cursor limits third-party models to a 200K context window. The cap applies even when the model the Router picks supports a larger window.
On Teams and Enterprise plans, Cursor also charges an additional $0.25 per million tokens on top of what Kimchi bills. On individual plans, this rate does not apply.

Next steps

Kimchi Model Router

How routing works, pricing, and Router analytics.

Cursor recipe

The full Cursor gateway setup with specific Kimchi models.