The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” During development or research, developers often face pain points such as scattered AI conversation logs, difficulty in forming structured knowledge, and inability to effectively trace sources or reuse research materials. The DSH Research Graph plugin is designed to address these issues.

What Is This

DSH Research Graph is a DeepSeek Harness Web plugin maintained by benz-ai-x. It belongs to the “Memory” category. The plugin organizes AI conversations into research topics and generates traceable knowledge cards and reusable research materials.

Core Features

  1. Research Graph
    The plugin provides a visual research graph for organizing discussions. You can inspect branch and merge states, and use dragging, collapsing, and relayout. Cards display turn counts, and the inspector reports session statistics.

  2. Source Tracing
    By searching discussion text, you can quickly return to the exact original turn, making it easy to locate and review context.

  3. Knowledge Capture
    It supports saving conclusions, conditions, and open questions. You can edit and iterate on knowledge through new revisions.

  4. AI Analysis
    The plugin supports extracting selected discussion content and comparing 2-3 materials. You can review before saving.

  5. Follow-up Research
    You can continue research starting from historical turns or a selected card revision, while retaining the materials already used as context.

  6. Tasks and Delivery
    It supports inspecting existing subagent records and exporting research results in Markdown format.

Installation and Enablement

Prerequisites

  • DSH Host version: 0.1.7-rc.2 (Web mode)
  • Node.js version: ^22.19.0 or >=24.0.0
  • System environment: dsh and pnpm commands must be available in PATH

Installation Steps

  1. Stop the currently running dsh web process.
  2. Run the install command:
dsh plugin --profile web add @benz-ai-x/dsh-research-graph@0.1.7-rc.2
  1. Start DSH Web:
dsh web
  1. Find the login URL in the terminal output, open it, enter a non-empty Session, and select Research Graph in the interface.

Typical Usage

After entering the Research Graph interface, the workflow is usually:
1. Read a source conversation or knowledge card.
2. Use the AI analysis feature to extract key information or perform comparisons.
3. Save conclusions as a knowledge card.
4. Start follow-up research based on that card or historical records.
5. Finally, export the research output as Markdown.

Applicable Scenarios and Considerations

This plugin is suitable for scenarios requiring product research, technical investigation, and continuous knowledge management.

Important Notes

  • Data Location: Research records are stored on the DSH Host, while workspace details (such as expanded nodes and view state) are kept in the browser. Topic arrangement must be explicitly saved.
  • Model Calls: Browsing, manual capture, and export operations do not consume model quota; AI generation and new discussions invoke models as needed.
  • Agent Capabilities: The plugin does not include presets or an Agent Team launcher; it only provides a research workbench.
  • License: The plugin uses the MIT license.

Ecosystem and Support

The DSH community catalog (such as skillhub.cn) is not officially affiliated with DeepSeek or Huanfang. The plugin is hosted on GitHub, where you can find the user guide, troubleshooting, and development documentation.