What is evo-trader?

{"i":0,"repo":"hhhmehmet/evo-trader","stars":6,"lang":"Python","ok":true,"char_count":1936,"text":"evo-trader is an experimental Python project that applies evolutionary ideas to cryptocurrency trading on Coinbase. It runs five independent agents, each following a different trading strategy, and lets them compete over 72-hour periods called generations. At the end of each generation, the best-performing agent's strategy is copied into five slightly modified descendants, which then compete in the next round. This cycle repeats indefinitely, so the system tries to improve its strategies over time through a process that resembles natural selection.\n\nThe project defaults to paper trading, meaning it uses real Coinbase market data but simulates the buys and sells without any real money changing hands. A fast demo mode lets you watch three generations run in under a minute using synthetic price data, which is useful for understanding how the evolution loop works without waiting days. Every generation's results are stored in a local SQLite database file, and if you stop and restart the program, it picks up the evolutionary history from where it left off.\n\nLive trading with real money is possible but deliberately hard to enable. The code can place real, authenticated orders on Coinbase, but it stays off by default. Turning it on requires setting two specific environment variables and passing an automated safety checklist that checks 11 things, including a loss limit, a kill switch, and mutation bounds. The project keeps live trading data in a separate database from paper trading data so the two never mix.\n\nA few supporting tools are included. A backtesting script fetches 30 days of real historical Coinbase price data and runs simulated generations against it in minutes, so you can evaluate strategies without waiting for real 72-hour windows. A dashboard built with Streamlit provides a read-only view of the database, showing per-generation rankings, individual agent performance, a family tree of how strategies evolved, and a log of agents that were eliminated.\n\nThe README is clear about the project's limitations: the included strategies are simple examples, backtest history showed net losses across 200 synthetic generations, and cryptocurrency trading carries substantial risk. The authors describe it as educational and experimental, not financial software, and they accept no responsibility for any losses from live trading."}