Introduction

The core design philosophy of DeepSeek Harness (DSH) is pluginization, enriching functionality through extension components. When building agents or automated workflows, enabling a model to “understand” images is a common need, but DSH does not natively provide a visual analysis interface. dsh-vision is a plugin maintained by community developer zoahdev. It fills this gap by providing a vision_analyze tool, allowing DSH to analyze images and return textual descriptions through OpenAI-compatible vision models.

Core Features

This plugin mainly provides the following capabilities:
1. Provides the vision_analyze tool for performing visual analysis.
2. Supports images from local file paths or network URLs.
3. Interacts with OpenAI-compatible vision models.
4. Returns image analysis results in text format.

Installation and Enablement

Installing the plugin requires using DSH plugin management commands. Follow the steps below in order:

  1. Run the installation command in your terminal:
dsh plugin --profile web add dsh-vision
  1. Configure the environment. The plugin needs to connect to a vision model. You can set apiKey (or OPENAI_API_KEY) through environment variables or a configuration file, or point baseUrl to a local compatible endpoint (for example, Ollama http://localhost:11434).

Tool Parameters

When using the vision_analyze tool, you need to provide two main parameters:

Parameter Type Description
image string Image path or URL using https / data: protocol
question string The question about the image, default is “Describe this image in detail.”

Configuration Items

The plugin supports the following configuration items (usually set through environment variables or a configuration file):

Configuration Item Type Default Description
apiKey string (environment variable) API key for the OpenAI-compatible vision model
baseUrl string https://api.openai.com API endpoint
model string gpt-4o-mini Vision model ID
timeoutMs number 120000 Request timeout (milliseconds)

Typical Usage

When calling this tool in a DSH workflow or plugin, an example usage is as follows:

// 示例：分析一张本地图片
await dsh.vision_analyze({
  image: "C:/Users/Example/Desktop/screenshot.png",
  question: "这张图片里主要显示了什么内容？"
});

// 示例：分析网络图片
await dsh.vision_analyze({
  image: "https://data:example.com/image.jpg",
  question: "请用中文描述这张图片。"
});

Notes

This plugin relies on an external vision model service. If a valid API key or local endpoint is not configured, the tool will not work properly. Additionally, this is a community plugin and is not an official product of DeepSeek or OpenAI.

Summary

With dsh-vision, developers can integrate visual analysis capabilities directly into the DSH ecosystem. The plugin is lightweight and only requires configuring the necessary vision model parameters before use.

Plugin directory: https://www.skillhub.cn/plugins/zoahdev/dsh-vision
Source repository: https://github.com/zoahdev/dsh-vision