The AI Hedge Fund is a Python-based proof of concept designed to explore how artificial intelligence can be applied to make trading decisions in a hedge fund context. The project is explicitly positioned as educational and research-focused, with no capability to execute actual trades. The system is currently undergoing significant evolution, with plans to rebuild it into a persistent, always-on AI hedge fund that functions as a first-class entity capable of backtesting, paper trading, and optional live execution, while reimagining investor agents as pluggable, backtestable alpha models.
The core architecture employs a multi-agent system where different AI agents embody the investment philosophies of renowned investors and analysts. These include agents modeled after Aswath Damodaran focusing on disciplined valuation, Ben Graham emphasizing value investing with margin of safety, Bill Ackman pursuing activist investment strategies, Cathie Wood representing growth and innovation investing, Charlie Munger seeking wonderful businesses at fair prices, Michael Burry hunting for deep value opportunities, Mohnish Pabrai applying Dhandho investing principles, Nassim Taleb analyzing tail risk and antifragility, Peter Lynch seeking ten-baggers, Phil Fisher conducting deep research, Rakesh Jhunjhunwala representing Indian market perspectives, and Stanley Druckenmiller pursuing macro opportunities. Beyond these named investor agents, the system includes specialized agents for valuation calculations, market sentiment analysis, fundamental data analysis, technical indicator analysis, risk management, and portfolio management that makes final trading decisions.
The repository shares contributors with several major open-source projects including home-assistant/core, prisma/prisma, and pytorch/pytorch.
Users can interact with the system through both a command-line interface and a web application. The command-line approach supports direct execution with optional flags for local LLM execution via Ollama and date range specification for backtesting. The web application provides a user-friendly visual interface as the recommended approach for users preferring graphical interaction. The system requires at least one LLM API key from providers including OpenAI, Groq, Anthropic, or DeepSeek to function. Installation uses Poetry for dependency management, and the project maintains clear contribution guidelines emphasizing small, focused pull requests. The codebase is licensed under the MIT License and explicitly disclaims any liability for financial losses, positioning itself strictly as an educational tool rather than investment advice.
