Serverless Inference

    Windsurf

    This tutorial walks you through configuring Windsurf to use an OpenAI-compatible API to access LLMs hosted by Kimchi. Because Windsurf does not natively support custom third-party model providers, you'll install the Roo Code extension to route requests through the Kimchi base URL.

    Overview

    By the end of this tutorial, you'll be able to:

    • Use Kimchi-hosted models (for example, glm-5-fp8) inside Windsurf
    • Route AI requests from Windsurf to Kimchi infrastructure via the Roo Code extension

    This tutorial is intended for developers who use Windsurf IDE.

    📘
    Windsurf's native BYOK (Bring Your Own Key) only supports Anthropic models. To use an OpenAI-compatible endpoint such as Kimchi, you need the Roo Code extension described in this tutorial.

    Prerequisites

    Before starting, ensure you have:

    1. A Kimchi API key — Go to Kimchi → API Keys and click Create API key. Copy the key and store it securely.
    2. Windsurf IDE — Download from windsurf.com
    3. Roo Code extension — Install from the Windsurf Marketplace

    Step 1: Open the Roo Code panel

    In Windsurf, click the Roo Code icon in the Activity Bar (left sidebar).

    If this is your first time opening Roo Code, the setup wizard will appear — follow its prompts to proceed to the provider configuration screen. Choose Use without an account.

    Step 2: Configure the third-party provider

    In the Roo Code Choose a model provider panel:

    1. Choose 3rd-party Provider from the provider list.
    2. Set API Provider to OpenAI Compatible.
    3. Set Base URL to https://llm.kimchi.dev/openai/v1.
    4. Set API Key to your Kimchi API key.
    5. Select glm-5-fp8 model from the list.
    6. Save the configuration.

    Step 3: Verify your configuration

    1. Open the Roo Code chat panel.
    2. Send a test prompt — for example, Hello. If you receive a response, the connection to Kimchi is working.

    Next steps