Preface¶

DeepSeek Harness (DSH) uses a plugin-based architecture, aiming to improve workflow efficiency through extensible functionality. When handling complex tasks, the capabilities of a single agent are often limited, and multi-agent collaboration is an effective way to improve efficiency. The ETeams plugin simulates real team structures to address the challenges of multi-agent goal decomposition and task assignment, enabling AI to collaborate in the roles of “leader” and “members.”

What It Is¶

ETeams is a multi-agent team collaboration plugin for DSH. The plugin transforms large language model capabilities into team collaboration capabilities through a mode in which the “leader decomposes goals and members execute them in turn.” It supports custom roles, independent model configurations, and multi-session collaboration. All state data is stored in a local SQLite database, with no network access and no uploads.

Core Features¶

• Custom System Roles: It supports defining a dedicated system prompt for each role to shape different assistant personas.
• Differentiated Model Configuration: Different session models can be assigned to different members of the team to enable resource scheduling.
• Multi-Session Collaboration: Members work in independent sub-sessions, supporting real-time monitoring of task progress and decision-making.
• Local State Storage: It uses built-in node:sqlite to store team state. Data is kept locally only, with no network access and no uploads.

Installation and Enabling¶

Install the plugin using the command line:

dsh plugin --profile desktop add dsh-eteams

After installation, restart the application to load the plugin.

Typical Usage¶

1. Add Members: Enter /eteam in the conversation. Through interview-style interaction, a role handbook is generated, and the role is added to the global role library.
2. Form a Team: Go to the “Teams” page in the panel, select members from the role library to add to the team, and designate one person as the leader.
3. Create Tasks: Clearly state the goal in the conversation. The leader decomposes the tasks, and after approval, they are automatically assigned to members for execution. Progress can be tracked in real time in the panel.

Notes¶

This plugin is a dual-side single bundle (host-side + browser-side). Before use, note that the plugin runs with the permissions of the current DSH process. Review the source code and license before installation. State is stored in local SQLite, so pay attention to the storage path and permissions.

Conclusion¶

For more technical details and the source code, visit: GitHub