Let four AI agents lose your money faster than you can

What it does

AutoHedge is a Python framework that wires four specialized AI agents into a linear trading pipeline: a Director generates strategy theses, a Quant runs technical analysis, a Risk Manager sizes positions, and an Execution Agent places orders. The goal is fully autonomous trading with “minimal human intervention.” Currently it only supports Solana via the Jupiter API; Coinbase and other CEXs are listed as “coming soon” or “roadmap.”

The interesting bit

The project treats LLM agents as institutional-grade pipeline workers rather than chatbots, with structured JSON outputs and “enterprise logging” for audit trails. Whether that discipline survives contact with live markets is the open question.

Key highlights

• Risk-first architecture: position sizing and risk assessment happen before any execution
• Modular agent design intended for custom strategies and new trading venues
• Built on the external Swarms agent framework
• Requires JUPITER_API_KEY, OPENAI_API_KEY, and a WALLET_PRIVATE_KEY to function
• 3,286 GitHub stars with tags name-dropping BlackRock, Goldman Sachs, and JPMorgan

Caveats

• Only Solana is actually supported; Coinbase is “in development” and other venues are merely “planned”
• The README claims “institutional reliability” but offers no backtests, performance metrics, or risk statistics
• “Experimental agents” label on the OpenAI integration suggests the LLM components may not be production-hardened

Verdict

Worth a look if you’re building agent-based trading infrastructure and want a reference architecture for pipeline separation. Avoid if you expect battle-tested execution across multiple exchanges or any evidence the system has traded real money successfully.