DeepSeek Harness (DSH) uses a modular plugin architecture. yyxcnasd/amadeus-for-dsh is a client-side plugin designed to materialize the Amadeus system (Kurisu Makise) from Steins;Gate 0. It provides Live2D character portraits, Japanese voice synthesis, and long-term memory management, helping users build a personalized intelligent assistant in DSH.

Core Features

  • Red flip-phone UI: Provides a red-themed flip-phone panel. The upper screen displays the Live2D portrait, supporting breathing, blinking, speaking mouth movements, and gesture animations; the lower screen is the chat area.
  • Real-time voice: Uses Edge TTS (free) by default; supports switching to VOICEVOX, Aqua-TTS, an OpenAI-compatible interface, or the VOICEVOX public API. Supports Japanese voice output and Chinese streaming display, with a stable voice used by default.
  • Long-term memory: Provides cross-session memory, recording historical conversations and performing fact extraction. Data is persisted to disk and the context can be recovered after restart.
  • Interaction features: Supports proactive incoming calls (with a retro ringtone and screen shake effect) and idle chit-chat. Supports event announcements and voice input (browser speech recognition preferred, falling back to the Whisper backend STT).
  • Word-level mouth synchronization: Uses word-level timestamps in TTS responses to achieve fine-grained synchronization between mouth movements and speech.

Installation and Enablement

The plugin follows DSH’s standard installation process. After installation, you can load it by restarting the corresponding profile.

  1. Install the plugin (using the desktop profile as an example):
    dsh plugin --profile desktop add github:yyxcnasd/amadeus-for-dsh

The installation process will automatically handle dependencies and configuration files.

  1. Load the plugin:
    After installation is complete, restart DSH on the profile you are using. The flip-phone panel will appear in the right sidebar.

  2. Update and uninstall:

    • Update: Re-run the same add command to complete the update.
    • Uninstall: Run dsh plugin --profile <p> remove amadeus-for-dsh (the data directory will be retained).

Dependency Setup

The plugin depends on Python and the edge-tts library to implement voice features. Using Linux as an example, it is recommended to use a virtual environment to isolate dependencies.

  1. Create a virtual environment and install edge-tts:
    python -m venv ~/.local/share/amadeus/venv
    ~/.local/share/amadeus/venv/bin/python -m pip install edge-tts
  1. Configure the environment variable (optional):
    The plugin resolves the Python interpreter in the following order: AMADEUS_PYTHON environment variable > python > python3. If using a venv, you can explicitly specify it:
    export AMADEUS_PYTHON="$HOME/.local/share/amadeus/venv/bin/python"

Notes

  • Copyright statement: The character image, voice, and lines are copyrighted by MAGES./Nitroplus. The Live2D model and voice assets are fan-made, for personal learning only, and commercial use is prohibited.
  • Platform differences: Auxiliary installation tools such as PowerShell, .bat, and install.ps1 are intended only for Windows. Linux and macOS users should directly use the dsh plugin add command.
  • Version compatibility: The plugin supports DSH 0.1.x and DSH Desktop 2.0.x. If the DSH 2.0 generation no longer supports the new installation method, please use the corresponding version’s release package.
  • Browser recognition: The voice input feature prefers browser speech recognition and falls back to the Whisper backend STT if it fails.

Further Reading