Introduction

The DeepSeek Harness (DSH) plugin ecosystem allows users to extend host capabilities. When managing multiple agent roles or coordinating tasks within a single window, traditional single-turn conversations or simple context injection often fall short. dsh-groupchat is designed to address multi-model collaboration, task allocation, and shared memory.

Plugin Overview

This is a plugin that provides multi-model, multi-role group chat functionality for DeepSeek Harness. It allows you to create independent group chats within a main conversation, supporting cross-group collaboration, task allocation, and member memory management.

  • Name: dsh-groupchat
  • Maintainer: fqxxzyw
  • Category: Workflow
  • License: MIT
  • GitHub: https://github.com/fqxxzyw/dsh-groupchat

Installation and Enablement

Use the DSH plugin installation command to add a local link:

dsh plugin --profile desktop add link:C:/plugins/dsh-groupchat

After installation, enable the plugin in the host interface under “Settings -> Plugins,” then restart the host and refresh the client.

Core Features

The plugin provides the following capabilities:

  1. Independent Sessions: Each conversation corresponds to one group chat. Group history, drafts, tasks, and memory are isolated from the host session.
  2. Custom AI Members: Supports configuring independent APIs, models, personalities, responsibilities, avatars, and bubble colors.
  3. Cross-Group Collaboration: Supports categorized @-mentions of members in the current group or other groups, enabling hierarchical relationships and approval workflows for calls.
  4. Task Allocation: Supports an “allocate first, then execute” mode with explicit nomination, responsibility matching, and task dependencies.
  5. Task Overview: Displays a task board below the message area, showing bot names, corresponding tasks, and completion status.
  6. UI Scaling: Provides a global font size setting (11–20 px) and scales buttons, icons, and spacing accordingly.
  7. Member Memory: Displays fixed memories and real-time conversation excerpts by member, with support for shared memory.
  8. Chat Experience: Supports Markdown, streaming output, and copy functionality.

Typical Usage

Initiate collaboration by @-mentioning a member or explicitly specifying a role:

  • Casual chat:
    > @Big Fat Fish Are you eating plain rice today too?
  • Task request:
    > @Big Fat Fish Help me explain the relationship between __init__, forward, and instance invocation in an RNN model. Provide a short runnable example.
  • Collaboration:
    > @Big Fat Fish @Review Expert Let’s review this plan together: @Big Fat Fish organizes the implementation steps, @Review Expert checks for omissions, then we compile the final suggestions.

Notes

  • This is a community plugin and is not affiliated with the official DeepSeek Harness.
  • Avatars are sourced from external image links, and community art assets are not covered by this plugin’s MIT code license.
  • Requires a DSH version that supports plugin clients and ctx.llm / ctx.webServer.
  • Automated testing cannot replace acceptance testing on an actual Windows host.
  • It relies on host services to access the API, so there is no need to re-enter keys in the plugin.

Conclusion

dsh-groupchat provides the infrastructure for managing multi-agent collaboration within a single interface. For scenarios requiring role-playing, complex task decomposition, or cross-role collaboration, this plugin can reduce the complexity of session management.