A pixel-style virtual office where AI agents work as your team members. Assign tasks via chat and watch your pixel team research, code, write, and collaborate — all in a retro pixel office visualization.
┌──────────────────────────────────────────────────┐
│ Frontend (React + Canvas) │
│ Pixel office visualization + Chat interface │
└──────────────────────┬───────────────────────────┘
│ WebSocket + REST
┌──────────────────────▼───────────────────────────┐
│ Backend (FastAPI + OpenAI) │
│ Agent orchestration + World state management │
└──────────────────────────────────────────────────┘
| Agent | Name | Specialty |
|---|---|---|
| 📊 Analyst | 小文 | Market research, competitive analysis, strategy |
| 🧪 Experimenter | 小实 | Experiment design, training, evaluation |
| ✍️ Writer | 小写 | Drafting, polishing, formatting documents |
| ⌨️ Developer | 小码 | Implementation, data processing, visualization |
- Real-time pixel office — watch agents walk, sit, eat donuts, and work at their desks
- Smart task routing — automatically assigns tasks to the best-suited agent
- Live chat — see agent responses with Markdown rendering
- Task queue — agents handle multiple tasks sequentially, getting "angry" when overloaded
- Idle animations — agents relax in the kitchen or watch TV when idle
- Office pets — a dog (豆豆) roams the garden, plus butterflies and birds
- Responsive design — works on desktop and tablet screens
cd backend
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txt
# Configure your OpenAI API key
cp .env.example .env
# Edit .env and add your OPENAI_API_KEY
# Start the server
python main.pycd frontend
npm install
npm run devVisit http://localhost:5173 in your browser.
- Type a task in the chat panel (supports English and Chinese)
- The system automatically assigns it to the best-suited team member
- Watch the pixel agent work in the office view
- Switch between Chat, Tasks, and Agents tabs to monitor progress
- "帮我调研竞品的最新动态"
- "Design an experiment to evaluate our new model"
- "帮我起草一份项目方案"
- "写一个 Python 脚本来处理 CSV 数据并画图"
The system uses OpenAI's GPT-4o-mini by default. You can modify the model in backend/agents/base.py.
To use without an API key, agents will return placeholder responses indicating what they would do.
- Backend: Python, FastAPI, WebSocket, OpenAI API
- Frontend: React, TypeScript, HTML5 Canvas, Vite
- Communication: WebSocket for real-time updates, REST for task submission