Foreword

DeepSeek Harness (DSH) provides a Web UI interface for managing agent tasks. When reading/writing project files or running tools, frequently switching windows to enter commands interrupts the workflow. The dsh-voco plugin tightly integrates voice input with the backend DSH Agent, allowing users to complete tasks through natural language and receive results as voice feedback in the Web UI.

Plugin Positioning

dsh-voco is a voice conversation plugin for the DSH Web UI. It is maintained by developer lgquan, currently has 2 stars, and uses the MIT License. By integrating cloud-based speech recognition and text-to-speech technology, it enables voice, text, and backend Agent tasks to collaborate within the same session.

Core Capabilities

  • Speech recognition and synthesis: Uses SiliconFlow’s free model XingChenAGI/XingChenASR-V3.2-Ultra for cloud speech-to-text; uses Edge TTS zh-CN-XiaoxiaoNeural for Chinese voice replies.
  • Task collaboration: Normal conversations are handled by the frontend, while work that requires reading/writing projects or running tools is delegated to the backend Agent.
  • Session persistence: After page switching, disconnection and reconnection, or DSH restart, the plugin can restore the voice session binding and maintain context continuity.
  • Instant confirmation and logging: Before delegating a task, it announces a confirmation message in real time, and the complete task report remains in the DSH task interface.
  • Optional memory extension: Can integrate @flowingspring/dsh-workspace-memory to include the primary voice session and completed delegated subtasks within the same Workspace long-term memory scope.

Installation and Configuration

Before installation, prepare Node.js 22.19 or higher (for DSH CLI), a running DSH Web profile, and a SiliconFlow API Key.

  1. Install the plugin:
    dsh plugin --profile web add @flowingspring/dsh-voco
  1. Start the Web UI:
    dsh web
  1. Configure the API Key:
    The plugin uses the environment variable SILICONFLOW_API_KEY. The secret is written to the DSH credential store and is not stored in plaintext in configuration files.
    • Recommended method: Enter and save it in DSH Web UI: “Settings → Plugins → Plugin Configuration”.
    • Other methods: Set the environment variable in the same terminal where you start dsh web, or configure it in the .env file under the DSH user directory.

How to Use

  1. Open the DSH Web UI and create or select a session.
  2. Click the microphone button and allow the browser to use microphone permissions.
  3. Speak directly. After about 1.5 seconds of silence, recognition is automatically submitted, or press and hold the microphone for about 400 milliseconds and release to submit.
  4. The speech recognition result is sent to the backend Agent for processing. After the Agent completes the task, the system announces the result via voice and saves the complete report in the DSH interface.

Notes

  • Not a browser extension: The plugin is not an independent browser extension, nor does it require a separate backend service; it depends on the DSH process.
  • Data cleanup: Uninstalling the plugin does not automatically delete saved API Keys, DSH sessions, or voice histories. If you need to completely remove the Key, manually edit the credential file or environment variable.
  • Dependency checks: If recognition fails or waits a long time, check network connectivity, API Key validity, account quota, and network latency.
  • Version requirements: Ensure Node.js 22.19 or higher is used.

Ecosystem Context

DSH follows the “everything is a plugin” philosophy. This plugin is listed in the community directory SkillHub and has no official affiliation with DeepSeek or High-Flyer.

Conclusion

dsh-voco integrates voice interaction into DSH workflows, making it suitable for scenarios where you need to operate the terminal with both hands or prefer to drive tasks with natural language. For more details and source code, visit GitHub.