Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture, allowing community developers to extend its core capabilities. When building voice interaction scenarios, developers often need fine-grained session control, explicit user authorization flows, and secure draft management mechanisms. The jstn-1g/dsh-live-voice plugin is designed to provide DSH with a controlled voice interaction layer, ensuring that every voice interaction occurs under explicit authorization and that session environments are strictly isolated.

Plugin Overview

Plugin Name: jstn-1g/dsh-live-voice
Core Value: Provides conditional single-turn voice interaction, synthetic demos, and draft handoff capabilities.
Maintainer: Jstn-1g
License: MIT

Core Features

This plugin primarily includes the following verified capabilities:

  • Conditional single-turn voice interaction: Limited to a single conversational turn and requires explicit user confirmation.
  • Precise session isolation: Ensures each voice session runs independently and prevents state pollution.
  • Explicit draft handoff: Provides a clear button action to place transcribed text into the draft area.
  • Credential-free local synthetic demo: Provides local synthesis demos without relying on external credentials.
  • Manual turn control: Allows users to manually trigger or control the turn flow.

Installation and Enablement

Install the plugin through the DSH command-line tool. Make sure the dsh client is installed.

  1. Run the following command to install the specified preview version:
    dsh plugin --profile web add github:Jstn-1g/dsh-live-voice#v0.3.0-preview.5
  1. After installation, restart the Web profile and open a Live Session.

Typical Usage

The typical steps for using this plugin in voice interactions are as follows:

  1. In the editor control bar of a Live Session, click the microphone icon to open DSH Live Voice.
  2. In the popped-up interface, confirm that the target is displayed as Local deterministic synthetic demo.
  3. Start the voice interaction. After completing the turn, the interface will provide two transcriptions.
  4. Click the Use my transcript as draft button.
    • Note: This action only places the text into the current draft area and does not automatically submit it to the editor.

Applicable Scenarios and Caveats

This plugin is suitable for development scenarios that require controlled voice interactions in a Web environment.

Caveats:
* Unverified features: Credential-based Qwen models, physical microphones or speakers, and packaged Desktop environments are currently unverified.
* Concurrency limitation: Do not allow multiple Harness instances to write to the same persistent Session simultaneously, as this may cause data conflicts.
* Operational boundaries: This plugin is designed as a secure sandbox and will never automatically submit to the editor or run tools.

Summary

The jstn-1g/dsh-live-voice plugin provides basic voice interaction capabilities for DSH through explicit authorization, session isolation, and manual control. For developers who need to build secure, controlled voice interfaces, this is a practical technical solution.