Preface

DeepSeek Harness (DSH) adopts a plugin-based architecture, allowing developers to inject specific capabilities into the system as modules. When handling tasks that require visual assets or dynamic content, agents often need to rely on external websites. The dsh-agnes plugin integrates Agnes AI’s image and video generation APIs into DSH, enabling agents to directly call free generation capabilities through standard tools and reducing reliance on external services.

Plugin Overview

This is an online tool plugin developed by chaoliu615. It exposes two core tools, agnes_image_generate and agnes_video_generate, through the DSH tool system, and provides an “Agnes” tab in the settings page to manage default parameters. The plugin is licensed under the MIT License and requires a Node.js ≥ 20 runtime environment.

Core Features

The plugin mainly provides two types of tools and one configuration interface:

  • Image Generation Tool (agnes_image_generate): Supports text-to-image, image-to-image, and multi-image composition. Parameters include prompt, size tier, aspect ratio, and optional reference images.
  • Video Generation Tool (agnes_video_generate): Supports text-to-video, image-to-video, and keyframe animation. This tool follows an asynchronous task flow; after returning a video ID, polling is required to obtain the final result. The total budget is 900 seconds.
  • Settings Panel: An “Agnes” tab appears on the DSH settings page, supporting model switching (such as agnes-image-2.1-flash, agnes-video-v2.0, etc.) and adjusting default parameters such as frame size, resolution, duration, and so on.

Installation and Enablement

Installing this plugin requires specifying the web profile. The installation can be completed with the following command:

dsh plugin --profile web add dsh-agnes

Typical Usage

After the plugin is installed, the agent must first configure AGNES_API_KEY in the DSH credentials, and then call the corresponding tools.

  1. Text-to-Image: The agent calls agnes_image_generate and passes parameters describing the subject, scene, style, and so on.
  2. Text-to-Video: The agent calls agnes_video_generate; the system creates an asynchronous task and returns a video_id. The agent must periodically query the task status until the video URL is obtained.
  3. Parameter Adjustment: In the “Agnes” tab on the DSH settings page, the agent or user can configure default parameters. After saving the changes, the configuration is persisted to ~/.dsh/settings.yaml, and subsequent tool calls will automatically read these default values.

Applicable Scenarios and Notes

  • Applicable Scenarios: Developers who need to configure multimodal generation capabilities for agents; users who want to reduce dependence on third-party image/video websites.
  • Runtime Environment: Ensure that the Node.js version in the environment is ≥ 20.
  • Video Task Limitations: Video generation uses an asynchronous flow, with a total budget of 900 seconds. If a task fails, re-query later with the video_id.
  • Browser-Side Loading: The “Agnes” tab on the settings page is provided by the browser side. The Host only includes the browser-side module in the startup graph when it scans for a dsh.client declaration. For previously installed older versions, after adding the browser side, the web profile must be restarted once to see the tab.

Summary

By integrating Agnes AI’s capabilities into the DSH toolchain, dsh-agnes provides agents with the ability to directly obtain image and video assets. By combining settings-page parameter persistence with tool calls, agents can flexibly adjust generation strategies according to task requirements. For more details and source code, please refer to the plugin directory or GitHub repository.