DeepSeek Harness Plugin: dsh-wsl-vllm

Introduction¶

DeepSeek Harness (DSH) adopts an “everything is a plugin” architecture. When you deploy vLLM in WSL to provide an OpenAI-compatible interface, checking service status and interacting through chat directly inside Harness reduces context switching. The dsh-wsl-vllm plugin is designed for this purpose, connecting vLLM instances running locally or in Docker into the DSH ecosystem.

Overview¶

• Name: dsh-wsl-vllm
• Purpose: Model inference
• Maintainer: 173787247
• Description: Connects to the OpenAI-compatible service of local/Docker vLLM (default port 8000).
• License: MIT

Installation and Enabling¶

Run the following command in the DSH CLI to add the plugin:

dsh plugin --profile web add github:173787247/dsh-wsl-vllm

If the companion dsh-wsl-kit is installed, you can batch-link plugins with the following command:

bash dsh-wsl-kit/scripts/link-linux-plugins.sh

Core Features¶

The plugin provides two main tools:

1. vllm_status
   Checks the reachability of the vLLM service and confirms the currently loaded model.

2. vllm_chat
   Provides chat completion by sending requests directly to vLLM.

Configuration Notes¶

• Service address: Connects to :8000 by default. Native Windows environments generally do not support vLLM; running it in Docker is recommended.
• Environment variables: You can override the default configuration using environment variables:
• DSH_VLLM_BASE: Custom Base URL
• DSH_VLLM_MODEL: Specifies the default model name

Compatibility¶

• Plugin version: 0.1.0
• Minimum DSH version: ≥ 0.1.2
• Runtime permissions: The plugin runs with the permissions of the current DSH process. Ensure you have access to the vLLM service.

References¶

• GitHub repository
• Plugin catalog page