Introduction

One of the core design principles of DeepSeek Harness (DSH) is “everything is a plugin.” When a main model (such as deepseek-v4-flash) lacks native visual input capability, a plugin is needed to bridge it to vision models. mimo-vision aims to address this need by providing text description capability for main models that do not have visual capabilities.

What It Is

mimo-vision is a native plugin designed for DeepSeek Harness (DSH), and its package name is mimo-vision. It is not an independent process, but a “first-class citizen” in the DSH capability model, leveraging DSH’s infrastructure to manage dependencies, files, credentials, and subprocesses.

Core Features

  • Tool Registration: Registers the describe_image tool for the main model to call.
  • Vision Bridging: Sends images to the mimo-v2.5 series multimodal models and converts visual information into text descriptions.
  • Format Transcoding: Supports transcoding multiple image formats (SVG, TIFF, HEIC, etc.) to ensure that vision models can parse them correctly.
  • Routing Strategy: Supports routing to free or paid APIs. By default, it prefers the free route and falls back to the paid route on failure.

Installation and Enablement

Prerequisites: Node.js version must be ^22.19 or >=24, and DSH must be able to start normally.

Install the plugin:

dsh plugin --profile web add mimo-vision

Typical Usage

Use the describe_image tool to describe an image. The tool parameters are path (required) and question (optional).

Examples:
1. Tell the model: “use describe_image to describe D:\photos\cat.png, focusing on what breed of cat it is”.
2. The model will call the describe_image tool and obtain a text description of the cat’s breed.

Notes

  • Operating Mechanism: The plugin is not an independent process. All dependencies, files, credentials, and subprocesses run through the seams defined by DSH.
  • Credential Configuration: An OpenCode API Key must be configured. The key resolution order is OPENCODE_GO_API_KEY -> OPENCODE_API_KEY.
  • Uninstallation: Use dsh plugin --profile web remove mimo-vision to remove it. This also cleans up dependencies and the bundle layer.

Conclusion

mimo-vision provides DSH with lightweight visual capability through a native plugin. It uses DSH’s infrastructure to handle files and credentials, making it suitable for scenarios where a main model without visual capabilities needs to gain image description capability.