Introduction

In the conversation flow of DeepSeek Harness, if you need to explain or annotate specific content, you usually have to interrupt the current interaction, copy the text to an external environment, or issue a new instruction. This approach can easily disrupt conversation coherence and increase contextual noise.

The look-here plugin aims to solve this problem. It provides a selection-priority interaction model, allowing users to directly select content in the conversation for explanation, questioning, or annotation without interfering with the main session.

Plugin Positioning

This is a plugin designed for the DeepSeek Harness Web UI, maintained by ZanderKong. It places explanation, memory, and annotation features on top of text selection, helping developers keep the main session clean while reading and debugging, and build personal reading memory.

Core Features

  • Explain from a selection: Select conversation text and quickly get an explanation or ask a custom question via the nearby interaction point.
  • Keep side work separate: Explanations and follow-up questions use separate ctx.llm calls and do not add messages to the main DSH session.
  • Choose the answer model: Follow the current session, or specify a particular provider and model in the plugin settings.
  • Save reading memory: Save full answers or excerpts, and show previous explanations when encountering the same term again.
  • Build an annotation batch: Collect multiple anchored annotations, edit them in the drawer, and then drag them into the DSH composer.
  • Stay in control: Dragging annotations into the composer only updates the draft; look-here never automatically sends messages in the main session.
  • Match your environment: Supports Chinese and English UI, theme colors, configurable selection gestures, and reduced motion support.

Installation and Enablement

Requirements: Node.js 22.19+ (22.x series or 24+), DeepSeek Harness 0.1.0-rc.7, and a DSH web profile.

Run the following commands to install and restart the service:

npx @deepseek-ai/dsh plugin --profile web add github:ZanderKong/look-here
npx @deepseek-ai/dsh --profile web --dump-config
npx @deepseek-ai/dsh --profile web

After installation, restart the Web service. The dump-config output should contain a # == look-here layer, and the default UI can be accessed at http://127.0.0.1:3080.

Typical Usage

  1. Open a DSH conversation and select text in an assistant reply.
  2. Short press the selection point to explain, or long press to ask a question or add an annotation (the order can be changed in settings).
  3. Save useful explanations, add follow-up questions, or collect multiple annotations.
  4. Drag the annotation basket into the composer, review the draft, and send it yourself.

Use Cases and Notes

  • Intended users: DSH users who frequently need to explain, annotate, and follow up on long text or code.
  • Data and permissions: Selected text, surrounding conversation context, questions, and follow-ups are sent through the model provider configured in DeepSeek Harness. Plugin settings, saved explanations, and annotations are stored in the browser profile used by the DSH Web UI.
  • Security note: The plugin does not read or store model API keys, nor does it automatically submit composer drafts. Avoid selecting sensitive information unless your configured model provider permits processing it.

Conclusion

look-here provides a mechanism for deep reading and annotation in DSH. By separating side work and preserving control of the main session, it improves workflow efficiency. For more implementation details and compatibility notes, refer to the GitHub repository.