Preface

DeepSeek Harness (DSH) uses a plugin architecture to extend its capabilities. When building agents, developers often need to handle tasks such as image generation, video production, speech synthesis, and visual understanding, which usually involves complex scripts or external services. The @mokuyoaxis/dsh-iris plugin integrates the Iris workspace and multi-vendor model routing, providing DSH with a complete media production and visual understanding runtime and centralizing task management in the workspace.

Introduction

This is a multimodal production runtime maintained by mokuyoaxis. It provides image, video, speech, and visual understanding capabilities for Agents and the Iris workspace, and can directly optimize prompts in the DSH dialog. The plugin connects through the browser client to the settings page, session input area, and global floating layer, allowing users to complete the full chain from content generation to visual analysis without leaving the DSH environment.

Installation and Dependencies

Before installing, ensure the environment meets the following requirements:
- Node.js >= 22.0.0
- DeepSeek Harness version 0.2.0-rc.2 (older host versions require installing @0.1.4)
- Video processing requires ffmpeg and ffprobe installed on the system

Install the plugin to the Web profile using the following commands:

dsh plugin --profile web add @mokuyoaxis/dsh-iris@0.2.0
dsh web

The installation size is approximately 27MB (includes the sharp binary). The plugin runs on the host side, and vendor keys are stored in the host-side Iris configuration.

Core Features

The capabilities provided by the plugin cover the full chain from content generation to visual analysis:

  • Media generation: image generation, text-to-video, image-to-video, S2V digital human video, and text-to-speech.
  • Visual understanding: image Q&A, long-image OCR, object localization, cropping, and pixel difference analysis.
  • Video processing: video frame extraction and multimodal content summarization.
  • Tasks and workflows: asynchronous task tracking, recovery after restart, media playback, a standalone works library, and prompt optimization.
  • Input and output: browser upload, session attachments, host paths, and a multi-vendor model pool.

Typical Usage

In a DSH conversation or the Iris workspace, you can use the following commands:

  • Image generation: Use iris_draw_image: Draw a white cat sitting by a blue window.
  • Visual summarization: Summarize the image I just uploaded.
  • Video processing: Use Iris to summarize /absolute/path/demo.mp4, then synthesize the summary into speech.

Notes

  • Version compatibility: Iris 0.2.0 only supports DSH 0.2.0-rc.2. Older DSH versions require installing @0.1.4. The unscoped dsh-iris package on npm is another plugin; when installing Iris, the full package name @mokuyoaxis/dsh-iris must be preserved.
  • Permissions: The plugin runs with host process privileges. It is recommended to review the source code and license before installing.
  • Platform: The installation size note indicates that image processing depends on sharp; in some ARM proot environments, it may be necessary to run dsh-iris doctor for diagnostics.

Summary

Iris provides DeepSeek Harness with core media production and visual understanding capabilities. By unifying the workspace and tool routing, it lowers the implementation barrier for multimodal tasks in Agent development.

Related Resources