Introduction

The core design philosophy of DeepSeek Harness is “everything is a plugin.” When conversing in the Web profile, manually entering tasks can be time-consuming and tedious. dsh-mic-dictation aims to solve this problem by adding a voice input feature to the prompt bar through a native client plugin.

Plugin Positioning

dsh-mic-dictation is a native client plugin for DeepSeek Harness, maintained by user Kilganon725.
Its core function is to insert a microphone button to the left of Full access in the Web prompt bar. Users can speak directly using their computer’s microphone, and the recognized text is automatically written into the prompt box, without switching input methods or typing manually.

Core Capabilities

  • Browser-native recognition: Uses the browser’s native SpeechRecognition / webkitSpeechRecognition, with Chinese priority (zh-CN).
  • Plug and play: A pure client-side plugin that works immediately after installation without modifying DSH source code.
  • Data security: Does not request or store any API key or account information, and does not collect voice, text, or any session data.
  • Automatic adaptation: Automatically follows React re-renders to keep the button on the left of Full access; supports automatic cleanup during HMR and plugin unloading.
  • Precise injection: Uses MutationObserver to locate DOM nodes, ensuring the button position is accurate and stable.

Installation and Startup

Before installing this plugin, make sure pnpm is installed on the system, because dsh plugin calls it internally.

  1. Install the plugin:
    Run the following command to add the plugin to the Web profile:
    dsh plugin --profile web add dsh-mic-dictation
  1. Start the service:
    Run the following command in the terminal:
    dsh web
  1. Refresh the page:
    Open the browser and visit localhost or 127.0.0.1. After refreshing the page, a microphone button will appear to the left of Full access in the prompt bar.

Usage Flow

  1. Click the microphone button on the left side of the prompt bar.
  2. The browser requests microphone permission. Choose Allow, then start speaking.
  3. Say a command, for example: “Help me organize the DeepSeek Harness project on my desktop.”
  4. The text being recognized is displayed in real time in the button tooltip, and the final result is automatically written into the prompt box.
  5. Click the button again to stop recognition, then press Send as usual.

Environment and Notes

  • Browser compatibility: Firefox has incomplete support for SpeechRecognition. Please use Chrome, Edge, or Safari.
  • Page address: You must access localhost or 127.0.0.1, as required by the browser secure context.
  • Network requirement: An internet connection is required. Chrome speech recognition depends on Google speech services.
  • Privacy note: The recognition process is completed locally in the browser, while Chrome sends audio to Google services for recognition. The plugin itself does not collect any data. It is recommended not to dictate sensitive information during recognition.

Conclusion

This plugin simplifies the input workflow for DSH Web by leveraging native browser capabilities. It is suitable for users who prefer voice input or need to assign tasks quickly. For more details or to submit an Issue, visit the GitHub repository or the plugin directory page.