Super Agent Party Turns Your Desktop Into a Live-Streaming AI Agent
A self-hosted, all-in-one AI companion bundling 3D VTuber avatars, computer-use agents, IM/livestream bots, and an OpenAI-compatible local API.
PRO- Super Agent Party is a self-hosted AI companion combining VTuber avatars with computer-use agents, at 2.4k stars.
- Ships 3D VRM, Live2D via VTube Studio, and THA-based 2D talking heads, all streamable transparently to OBS.
- Task Center and desktop vision let agents drive mouse, keyboard, terminal, and a built-in AI browser.
- One-click bot deployment to QQ, WeChat, Feishu, Telegram, Discord, Slack, plus Bilibili, YouTube, and Twitch livestreams.
- Exposes OpenAI-compatible API and MCP server at localhost:3456 for external code and tooling.
- Cross-platform installers plus Docker; AGPLv3 license with separate commercial license required for closed-source use.
A Chinese open-source project called Super Agent Party has quietly grown into one of the more ambitious attempts to fuse two different AI trends into a single desktop app: the AI VTuber (think Neuro-sama) and the computer-use agent (think Claude's computer control). It bills itself as a self-hosted mashup of both, and at roughly 2.4k stars on GitHub has picked up a real user base of hobbyists, streamers, and indie builders.
The pitch is simple. Install one app, get a 3D or 2D anime companion sitting on your desktop, and that same character can drive your mouse and keyboard, run background tasks, chat in Telegram, or live-stream to YouTube while exposing an OpenAI-compatible endpoint for your own code. Everything runs locally, and you plug in whatever model backend you want.
One companion, many jobs
The desktop companion supports custom VRM models, actions, and 3D scenes, and works with Live2D extensions. A THA-based mode animates a 2D talking head from audio-driven blendshapes, and a bridge to VTube Studio handles Live2D models with custom actions and expression control. For streamers, the avatar can be piped transparently into OBS.
The more interesting piece for developers is the Task Center and computer-control stack. Through desktop vision combined with a toolchain for controlling the mouse, keyboard, and terminal, agents can accomplish tasks on a computer. The repo demos this with an agent playing Wordle. The Task Center lets agents perform advanced tasks in the background and supports MCP and Agent Skills, so you can drop in your own tool definitions.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
