Summary

AI-Trader pits multiple AI models in a zero-human-input competition to determine if AI can beat the market, specifically trading NASDAQ 100 stocks. It targets researchers and practitioners by providing an autonomous, tool-driven platform for evaluating AI trading performance and strategies.

How It Works

The project utilizes a pure tool-driven architecture via the Model Context Protocol (MCP) toolchain, enabling AI agents to autonomously execute trading operations. Multiple AI models compete with identical starting capital, real market data, and historical replay capabilities, ensuring fair comparison. A core innovation is its replayable trading environment with strict anti-look-ahead data controls, facilitating reproducible empirical research into AI decision-making under controlled market conditions.

Quick Start & Requirements

• Installation: git clone, pip install -r requirements.txt.
• Prerequisites: Python 3.8+, API keys for OpenAI, Alpha Vantage, and Jina AI.
• Configuration: Set up API keys and other parameters in a .env file.
• Data Preparation: Run ./fresh_data.sh and python merge_jsonl.py within the data/ directory.
• Services: Start MCP services using python start_mcp_services.py in agent_tools/.
• Run: Execute python main.py to start the AI trading arena.
• Links: Quick Start, Configuration Guide, 中文文档.

Highlighted Details

• Fully Autonomous: AI agents operate with 100% independent analysis, decision-making, and execution, requiring zero human intervention.
• Tool-Driven Architecture: Leverages the MCP toolchain for standardized AI interaction with trading, data, and information retrieval functions.
• Multi-Model Competition: Deploys and competes multiple AI models simultaneously in a controlled arena.
• Real-Time Analytics: Offers comprehensive trading records, position monitoring, and profit/loss analysis.
• Reproducible Backtesting: Historical replay with anti-look-ahead data controls ensures scientific rigor.
• Extensible Framework: Supports easy integration of third-party strategies and custom AI agents.

Maintenance & Community

• Community: Feishu and WeChat groups are available for communication.
• Contributions: Welcomes contributions for AI trading strategies, custom agents, analysis tools, data sources, and documentation via GitHub Issues and Pull Requests.
• Roadmap: Future plans include A-Share support, cryptocurrency trading, and a strategy marketplace.

Licensing & Compatibility

• License: MIT License.
• Compatibility: The MIT license is permissive, suitable for commercial use and integration with closed-source projects, requiring attribution.

Limitations & Caveats

The current trading universe is limited to NASDAQ 100 component stocks; A-Share and cryptocurrency support are pending roadmap items. Obtaining and configuring multiple third-party API keys (OpenAI, Alpha Vantage, Jina AI) is required, which may incur costs. The project is under active development, with ongoing feature additions and potential for breaking changes.