What it is
Trading Agents is an experiment in autonomous financial decision-making: can AI systems profitably navigate markets with minimal human intervention?
The Setup
A Raspberry Pi running 24/7 with $500 of capital, executing trades across crypto markets and Polymarket prediction markets. The constraint is intentional: if the system can work with limited capital and compute, it validates the approach.
Crypto Day Trading
Built on Freqtrade, the system uses reinforcement learning to adapt strategies based on market conditions. It's not trying to beat the market: it's trying to compound small edges consistently while managing risk.
Polymarket Arbitrage
Prediction markets often misprice events, especially as new information emerges. The system monitors odds across markets, identifies discrepancies, and executes when the expected value is positive.
Technical Architecture
- Finite State Machine for execution flow
- Risk management with position limits and stop-losses
- Event-driven architecture for real-time market data
- Docker containers for isolation and recovery
The Learning
Running real money, even small amounts, changes how you think about AI systems. Every bug has a cost. Every edge case matters. It's a forcing function for robust engineering.
Connection to Chimera and Agora
This foundational work evolved into more sophisticated systems: Chimera (multi-agent consensus) and Agora Quantum (full autonomous hedge fund). Trading Agents was the proving ground.
Connected work