Introduction

The plugin ecosystem of DeepSeek Harness (DSH) emphasizes “everything is a plugin.” In long-running project development or complex agent debugging, chat records are often scattered across chat windows, making them hard to manage and reuse. The dsh-noteboard plugin provides a solution: it adds a workspace-level infinite canvas in the session view, extracts conversation content into structured Markdown notes, and supports sending notes back to the AI for processing via quote bubbles.

Setup and Enabling

This plugin depends on the DSH environment. Before installing, ensure your environment meets the following requirements:
- DSH version: >=0.2.0-rc.2
- Node.js version: >=22.19.0

Install it through the official plugin registry:

dsh plugin --profile web add dsh-noteboard
dsh web

The npm installation uses prebuilt artifacts, so no local build scripts are needed, and pnpm’s build authorization confirmation is not triggered.

Core Features

Infinite Canvas and Note Management

The plugin provides an infinite canvas with support for panning, zooming, grid snapping, and switching multi-canvas layouts. Notes are stored as Markdown files; each file includes frontmatter metadata and supports seven colors and both light and dark themes.

Capturing from Conversations

In the conversation view, selecting a segment of text displays the capture toolbar. You can choose to “Save to canvas” directly or choose “AI distill” to generate a concise note.

Each note records its source session. Clicking the “Open conversation” button on the note jumps directly back to the original location and highlights the corresponding text fragment.

Sending Back to AI

Notes can be dragged into the input box and sent to the AI as quote bubbles. The system validates the length of quoted content, and sending is blocked if cumulative quote bubbles exceed 32,000 characters.

Organization and Tools

You can focus, filter, and search by title, content, or tags using tag pills, and recover notes that are off-canvas. The plugin registers 12 model tools and 5 built-in workflows for automated note data processing.

Typical Use Cases

  1. Capture and Distill
    Select key content in a conversation and choose “AI distill.” The plugin generates a note according to the configured distillation requirements (title, tags, and body length).

  2. Batch Extraction
    Use the noteboard-extract workflow to filter content from sessions by conditions and save it as notes.

  3. Quoting and Jumping
    Click a note’s “Source” or “Open conversation” to jump to the original text. After processing, drag the note back into the input box, and the AI can perform follow-up operations based on these structured quotes.

AI Tools and Workflows

The plugin provides a rich set of AI tools:

12 model tools:
- canvas_add_note: Create a note
- noteboard_query: Search or read the full text of a note
- noteboard_create: Create notes in bulk (up to 200)
- noteboard_update: Modify notes, with support for bulk coloring by color groups
- noteboard_layout: Rearrange, align, or distribute notes by tags
- noteboard_sessions: Find source sessions
- noteboard_read_session: Read the original session text
- noteboard_add / noteboard_remove: Add and remove notes
- noteboard_save_as: Save the canvas layout
- noteboard_history / noteboard_restore: History and restore

5 built-in workflows:
- noteboard-extract: Filter and extract notes by conditions
- noteboard-organize: Categorize by topic, add tags, and rearrange
- noteboard-compare: Compare options, weighing pros and cons
- noteboard-actions: Break ideas down into action items
- noteboard-synthesize: Merge duplicate notes or synthesize conclusions

Notes and Limitations

  • File Format: Canvas files use JSON Canvas 1.0 format; content and layout are separate and can be managed with a text editor.
  • Single-User: The plugin assumes a single-user environment and does not support cross-process file locking or real-time collaboration.
  • No UI Edges or Groups: Connection lines and grouped UI are currently not provided, but the file format reserves space for them.
  • Operation Mechanism: File changes are not listened for; the entire state is reread before each RPC operation.
  • History: The most recent 50 operation records are retained by default.
  • Configuration Changes: Old prompt configurations are no longer effective; use “Additional distillation requirements.”

Conclusion

dsh-noteboard combines DSH’s conversational flow with structured knowledge management. Through notes and canvas, it enables the accumulation and reuse of information, making it suitable for developers who need to organize ideas and track decisions during long-term sessions.

GitHub Repository | Plugin Registry