What it does GPT-vup is a Python-based virtual streamer framework that connects GPT-3.5 to Bilibili or Douyin live chats. It reads弹幕 (danmu), welcomes viewers, thanks gift-givers, and pipes responses through VTube Studio to animate a Live2D avatar. A producer-consumer model queues incoming events so the bot doesn’t choke when the chat explodes.

The interesting bit The plugin architecture is where the ambition shows: voice control via ctrl+T, scheduled storytelling/rap interludes, and even a Milvus vector-store plugin for context-aware answers scraped from Tieba. It’s essentially trying to build a full AI talent agency in one repo.

Key highlights

• Supports both Bilibili (bilibili-api-python) and Douyin (via a separate proxy-based弹幕 grabber)
• VTube Studio integration for lip-sync and triggered animations via pyvts
• Speech plugin: pyaudio + speech_recognition for voice-to-text commands
• Context plugin: Milvus 2.0 + MySQL pipeline for RAG-style memory from forum data
• Character personality configurable through utils/prompt_temple.py templates

Caveats

• Explicitly abandoned: README banner says “本库停止维护” (stopped maintenance); author points to successor langup-ai
• Windows 10 + Python 3.8 + global VPN only; no cross-platform support claimed
• Douyin setup requires running a separate Gitee project for弹幕 capture
• Context plugin is labeled “实验功能” (experimental) and needs Docker, MySQL, and manual script editing

Verdict Worth a quick archeological dig if you’re building a Chinese-platform VTuber stack and want to see how far glue code can stretch. Everyone else should probably look at the author’s maintained successor or more modern alternatives.