DeepSeekSoftwareShipping codeAI score · 82/100AI-assisted edit · no obvious prompt-injection riskGitHub

Install the Nexus Voice Assistant Plugin for DeepSeek Harness

Install Nexus via local tgz package on DeepSeek Harness Desktop to enable voice-controlled multi-session tasking

Some experienceStepsWorkflowTip
Who
Developers already using DeepSeek Harness Desktop who want to quickly dispatch tasks across multiple work sessions by voice
You get
After installation, create sessions, assign tasks, and receive spoken replies via voice commands
You need
DeepSeek

Constraints

Prerequisites: Windows; DeepSeek Harness / Desktop 0.2.1-rc.5; a default model configured with tool call support; DashScope API key stored in Host credential service as DASHSCOPE_API_KEY. macOS and Linux are untested per source.

What they did

jiangkaiqi2005 developed Nexus as a voice butler plugin for DeepSeek Harness. Users describe tasks by voice, and Nexus uses Harness's native tools to dispatch work to appropriate sessions, then reports results by voice. Supports interruption anytime, continuing by text after hang-up, and two voice modes (Cascade and Realtime).

How it went

Author reports Cascade mode was added mainly for cost: in Realtime mode most spend was re-sent context, and the Realtime model never hit cache. Known issues: supervisor sometimes says 'dispatched' without actually calling a tool; on recognition failure Nexus shows '没听清' (Could not hear you) and doesn't start a turn; on synthesis failure only that sentence goes silent while text is kept. Automatic compaction is skipped when there isn't enough old history to summarize, so token meter can sit above budget temporarily. (Source: README.md Known limitations)

Worth knowing

Only Windows is tested; macOS and Linux are untested. DashScope API key must be configured via Host credential service; browser client never receives the key. Realtime mode uses qwen3.8-omni-flash-realtime. Full scenario with physical devices, real DashScope, and real default model has not been fully verified. Development build requires Node.js ^22.19.0 || >=24.0.0.

Steps

  1. Prepare Desktop environment

    Open DeepSeek Harness Desktop once to let it auto-create the desktop profile, then quit.

  2. Install the Nexus plugin

    Use the built-in dsh CLI from Desktop (PowerShell) to install the plugin. Replace paths with your local actual paths.

    Execution prompt · Source text。PowerShell terminal; replace dsh.cmd and tgz paths with local actual paths before execution

    dsh plugin --profile desktop add .\dsh-nexus-0.2.3.tgz

  3. Configure default model

    Reopen Desktop and select your default model and reasoning effort in Harness model settings.

  4. Verify text interaction

    Click Nexus at the bottom of the sidebar, type one line to confirm the model replies normally.

  5. Connect voice and authorize

    Click 'Connect voice' and allow the browser to use the microphone.

  6. Voice interaction

    Click 'Tap to talk' to start recording, then 'End recording' when done. One recording is at most 60 seconds, and the transcript is sent immediately. While Nexus is reading aloud, click 'Tap to interrupt' to cut in. Click the phone icon to hang up.

  7. Adjust settings (optional)

    Go to Settings → Nexus to adjust voice mode (cascade/realtime), speech recognition and synthesis models, voice, compaction budget, etc. Changes usually take effect immediately; voice mode takes effect on the next call.

Prompt you can copy

Reference prompt · Editorial suggestion。Confirm the destination, input variables and output format before use.

This block is the prompt. Ads are never placed here.

No prompt needed — just follow the steps.

Replace paths in the install command with your local actual paths; Windows paths use backslashes.

Source

用本地 tgz 安装 dsh-nexus 桌面插件 · jiangkaiqi2005 · GitHub · 2026-10-31T09:04:07Z

Comments & feedback

Tried this method? Share results, problems or improvements. Comments appear after moderation. Please avoid sensitive information.

Loading…

    Related