dsh-talk-map: DSH Visual Conversation Map

Introduction¶

As a plugin-based platform, DeepSeek Harness (DSH) allows users to enhance workflows through extensible tools. In long-running or multi-session development and collaboration, users are often limited by “working memory” and find it difficult to maintain contextual connections across multiple conversations. The dsh-talk-map plugin translates this pain point into spatial management capabilities, using an infinite canvas and card interactions to provide a more intuitive way to navigate conversations.

What Is This¶

This is a visual conversation map plugin designed for DeepSeek Harness. It turns each conversation into a card on the canvas and provides a whiteboard-like interaction experience. The plugin aims to replace traditional “working memory” with “spatial memory,” helping users manage an unlimited stream of conversations. The project is maintained by Tasihi89 and is released under the MIT License.

Core Features¶

The plugin mainly addresses session organization and context inheritance:

1. Spatial Memory and Infinite Canvas: Provides an infinitely extending whiteboard with 16px grid snapping. Card positions are persisted, allowing users to arrange conversations according to their spatial intuition without being automatically scrambled by refreshes or interactions.
2. Double-Click Interactions and Card Management:
   • Double-click blank space: creates a new conversation at that location and enters it directly.
   • Double-click a card: opens the corresponding conversation.
   • Drag a card: moves and places it freely on the canvas.
3. Connection Branching and Context Injection: Drag a line from the right-side handle of a card to blank space to preview and edit a context summary. After confirmation, a new conversation is forked while inheriting the context of the previous conversation.
4. Automatic Summaries: Once a conversation enters an idle state, the plugin automatically generates a three-part summary containing the summary, key findings, and next steps.
5. Bilingual Interface: Supports Auto / Chinese / English switching. The map interface language can be changed in the header language pill, and the selection is remembered.
6. Lineage Visualization: Native DSH fork/subagent generation relationships are displayed as dashed lines, while user-defined injection connections are displayed as solid lines.

Installation and Enablement¶

The installation process requires no build step; add the plugin directly using the official command:

dsh plugin --profile web add github:Tasihi89/dsh-talk-map

After installation, restart dsh web. A map button will appear at the bottom of the sidebar.

Typical Usage¶

Use Cases and Notes¶

• Target Users: Suited for people who prefer locating information through visual and spatial cues, and especially as an ADHD-friendly conversation management tool.
• Data Storage: Canvas data (positions, connections, and summaries) is stored via DSH’s official storage-domain in $DSH_HOME/storages/talk_map.json.
• Context Generation: Summaries are generated using the default model configured in DSH through local LLM channels, with no external requests.
• Compatibility: The plugin is built against deepseek-harness 0.1.0-rc.6. Since it is in the rc stage, APIs may drift. Before upgrading, it is recommended to check the structural contracts in src/client/dsh.ts and src/host/dsh-host.ts.

Summary¶

dsh-talk-map provides DSH with a visual session management layer. By combining cards, connections, and an infinite canvas, it transforms scattered conversations into a structured knowledge graph, making it easier for users to trace back, fork, and inherit context in complex conversation flows.

Resources:
* Plugin directory
* GitHub repository