DSH Research Graph

English | 简体中文

An independently maintained, open-source AI research workbench plugin for DeepSeek Harness (DSH), running inside DSH Web. Organize conversations into research topics and traceable Knowledge Cards for product research, technical research, and ongoing knowledge management.

Capability map

---
config:
  flowchart:
    nodeSpacing: 16
    rankSpacing: 32
---
flowchart LR
  G["DSH Research Graph"] --> O["Organize research"] & K["Capture knowledge"] & R["Extend research"]
  O --> O1["Cross-workspace topics"] & O2["Conversation graph<br/>Branches · Merges<br/>Subagent records"]
  K --> K1["History search<br/>Sources · Digests"] & K2["Knowledge Cards<br/>AI extraction · Review · Revisions"]
  R --> R1["Synthesis · Frozen materials"] & R2["Follow-up discussions<br/>Markdown export"]
  classDef brand fill:#2563eb,stroke:#2563eb,color:#fff
  classDef group fill:#eff6ff,stroke:#93c5fd,color:#1e3a8a
  classDef feature fill:#f8fafc,stroke:#cbd5e1,color:#0f172a
  class G brand
  class O,K,R group
  class O1,O2,K1,K2,R1,R2 feature

Workbench screenshot

Version 0.1.5-rc.2.8 with demo data. Click the image for the full-resolution screenshot.

Key capabilities

Capability	What you can do
Research graph	Organize discussions, inspect branches and merges; drag, collapse, and relayout
Source tracing	Search discussion text and return to the exact original turns
Knowledge capture	Save conclusions, conditions, and open questions; edit through new revisions
AI analysis	Extract selected discussion and compare 2–3 materials; review before saving
Follow-up research	Start from a historical turn or chosen card revision, retaining the materials used
Tasks and delivery	Inspect existing subagent records and export Markdown research results

Key facts

Item	Details
Plugin release	0.1.6-alpha.2, npm next
Required Host	DSH 0.1.6-alpha.2 · Web, Node.js ^22.19.0 || >=24.0.0
Data	Research records live on the DSH Host; working positions stay in the browser and Topic arrangements require explicit saving
Model calls	Browsing, manual capture, and export make no model call; AI generation and new discussions use models on demand
Agent capabilities	DSH provides models, tools, presets, and agent execution; this plugin includes no preset or Agent Team launcher. Details
Stack and license	TypeScript · React · MIT

Quick start

Requires DSH 0.1.6-alpha.2, with both dsh and pnpm on your PATH. Stop any running dsh web, then run:

dsh plugin --profile web add @benz-ai-x/dsh-research-graph@0.1.6-alpha.2
dsh web

Open the login URL printed in the terminal, enter a non-blank Session, and select Research Graph: read a source → save knowledge → continue the discussion. See installation options if you run DSH through npx or from source.

Docs and support

User guide · Troubleshooting · Development and release · Release acceptance · Report an issue