AI导哈

AI交易模拟器

一个多智能体AI交易模拟器，由自主交易员代理研究市场、制定计划、按日程执行模拟交易，并通过实时Gradio仪表盘监控。

  • by CarolinaCapilla
  • Python
  • 本地部署
  • 更新于
在 GitHub 查看 CarolinaCapilla/trader-floor-ai-simulator

使用说明

title: "Trader Floor AI Simulator" emoji: "📈" colorFrom: blue colorTo: green sdk: gradio sdk_version: "5.47.2" app_file: app.py pinned: false

Trader Floor AI Simulator

A group of AI traders with different strategies research the market, decide and trade a simulated portfolio on a daily schedule, with a live dashboard to watch them.

This started as the autonomous traders project from Ed Donner's agentic AI course. What I changed:

  • Restructured the flat scripts into a package (domain, services, integration, agents, scheduler, UI)
  • Replaced the account, market and notification MCP servers with in-process tools, and ran traders sequentially after concurrent MCP servers kept conflicting
  • Kept the researcher's MCP servers (fetch, Brave search, libSQL memory) and silenced npm output so their stdio stays valid JSON
  • Dockerised it and deployed it to Railway with a persistent volume, first-run database setup and a daily schedule

Stack: Python, OpenAI Agents SDK, MCP, Gradio, SQLite, Docker, Railway

🚀 Quick Start

Local Development

# Clone and install
git clone https://github.com/CarolinaCapilla/trader-floor-ai-simulator.git
cd trader-floor-ai-simulator
uv pip install -e .

# Set up environment variables (copy .env.example to .env)
# Add your API keys: OPENAI_API_KEY, POLYGON_API_KEY, etc.

# Run the UI
python app.py

# Or run the scheduler
python trading_floor.py

Deploy to Railway (Recommended for Scheduled Trading)

railway login
railway init
railway up

See RAILWAY.md for complete deployment guide with:

  • Persistent database via volumes
  • Daily scheduled trading runs
  • Environment variable setup
  • Monitoring and troubleshooting

🏗️ Features

  • AI-Powered Trading Loop: Each trader researches, generates a plan, and executes trades
  • Multi-Agent Composition: Trader agents leverage a dedicated Researcher agent via MCP
  • Web Interface (Gradio): Live dashboard with portfolio values, charts, logs, holdings, and transactions
  • Pluggable Market Data: Use Polygon MCP (paid or realtime) or a local market server
  • Persistent State: SQLite-backed accounts and logs in accounts.db
  • Push Notifications: Optional Pushover integration for post-trade summaries
  • Configurable Scheduling: Run continuously or a fixed number of iterations

🏗️ Architecture

Core components:

  • Trader agents (traders.Trader): Orchestrate research and trading using tools
  • Researcher agent: Exposed as a tool to each trader via MCP servers
  • MCP servers (mcp_params.py):
    • accounts_server.py for account tools and resources
    • push_server.py for push notifications (optional)
    • Market data via:
      • Polygon MCP (uvx --from git+https://github.com/polygon-io/mcp_polygon@v0.1.0 mcp_polygon) when configured, or
      • Local market_server.py
  • Scheduler (trading_floor.py): Runs all traders every N minutes (or a fixed number of iterations)
  • UI (app.py): Gradio dashboard polling live account data and logs
  • Database (database.py): SQLite tables for accounts, logs, and market

📋 Prerequisites

  • Python 3.12+ (Hugging Face Spaces defaults to Python 3.10 unless you set a runtime. If using HF Spaces, either:
    • Add a runtime.txt with python-3.12 OR
    • Remove dependencies that require newer stdlib behavior.)
  • uv for fast installs/runs (or pip)
  • Optional: Polygon API key (for market data MCP)
  • Optional: Pushover credentials (for notifications)

🛠️ Installation

  1. Clone the repository

    git clone <your-repo-url>
cd trader-floor-ai-simulator

  2. Install dependencies

    # with uv (recommended)
uv sync

# or with pip
pip install -r requirements.txt

  3. Set up environment variables Create a .env in the project root:

    # Server and scheduling
RUN_EVERY_N_MINUTES=60
RUN_EVEN_WHEN_MARKET_IS_CLOSED=false
USE_MANY_MODELS=false
MAX_ITERATIONS=1

# Market data
POLYGON_API_KEY=
POLYGON_PLAN=free   # free | paid | realtime

# Research tools
BRAVE_API_KEY=

# Model providers (use any you plan to call)
DEEPSEEK_API_KEY=
GOOGLE_API_KEY=
GROK_API_KEY=
OPENROUTER_API_KEY=

# Push notifications (optional)
PUSHOVER_USER=
PUSHOVER_TOKEN=

🚀 Usage

Run the Gradio Dashboard

uv run app.py
  • Opens the dashboard in your browser. It auto-refreshes logs (~0.5s) and portfolio/chart/holdings/transactions (~120s). Adjust timers in app.py if needed.

Run Traders on a Schedule

# run once (default MAX_ITERATIONS=1)
uv run trading_floor.py

# run N iterations
MAX_ITERATIONS=3 uv run trading_floor.py

# run continuously every N minutes
MAX_ITERATIONS=1000000 RUN_EVERY_N_MINUTES=60 uv run trading_floor.py
  • Press Ctrl+C to stop when running continuously.

📁 Project Structure

trader-floor-ai-simulator/
├── accounts.py               # Account model and operations
├── accounts_client.py        # MCP client for accounts server
├── accounts_server.py        # MCP server exposing account tools/resources
├── account.db                # SQLite database (ignored by git)
├── app.py                    # Gradio UI
├── database.py               # SQLite setup and CRUD helpers
├── market.py                 # Market data integration (Polygon or fallback)
├── market_server.py          # Local MCP server for market data
├── mcp_params.py             # MCP server parameters (accounts, push, market)
├── push_server.py            # Pushover MCP server (optional)
├── traders.py                # Trader and Researcher agent glue
├── trading_floor.py          # Scheduler to run traders periodically
├── requirements.txt / pyproject.toml
└── README.md

🔧 Configuration

  • Scheduling: RUN_EVERY_N_MINUTES, MAX_ITERATIONS, RUN_EVEN_WHEN_MARKET_IS_CLOSED
  • Models: Toggle USE_MANY_MODELS to run traders with multiple model backends
  • Market Data: Set POLYGON_PLAN to free, paid, or realtime; provide POLYGON_API_KEY
  • Push: Provide PUSHOVER_USER and PUSHOVER_TOKEN to enable notifications
  • Memory (MCP): Local memory DBs are created under memory/ automatically

🤝 Contributing

  1. Fork the repository
  2. Create a feature branch: git checkout -b feature/amazing-feature
  3. Commit your changes: git commit -m 'Add amazing feature'
  4. Push to the branch: git push origin feature/amazing-feature
  5. Open a Pull Request

📄 License

This project is licensed under the MIT License — add a LICENSE file if missing.

🐛 Troubleshooting

Common Issues

  1. Failed to spawn: accounts_server.py: Ensure file exists and mcp_params.py references accounts_server.py
  2. Unable to open connection to local database ./memory/<name>.db: 14:
    • Ensure the memory/ folder exists (the app now creates it automatically)
    • Use absolute LIBSQL_URL if running tools elsewhere
  3. Polygon API errors: Set POLYGON_API_KEY and POLYGON_PLAN
  4. UI not updating: Lower timers in app.py or trigger a new trader run

Debug Mode

Enable additional logging in your shell as needed (example):

export MAX_ITERATIONS=1
export RUN_EVERY_N_MINUTES=1

🔄 Future Enhancements

  • [ ] Migrate UI to Vue.js/Nuxt.js
  • [ ] Export reports (CSV/PDF)

📞 Support

  • Open an issue with logs, steps to reproduce, and environment details.

相关 MCP