A project called Stonkfly simulates a full fly brain connectome. It processes Coinbase price data as visual input. The simulation proposes buy, sell or hold actions. A Coinbase AgentKit integration checks limits and places real spot orders. Positive portfolio results stimulate dopamine cells. Negative results stimulate aversive dopamine cells. A memory rule adjusts neural connections. Profitable trading has not been proven. Default runs use paper trading with public BTC-USDC data. Live trading requires a dedicated Coinbase portfolio and API key. Orders are capped at $10 with a $20 drawdown stop. The code is open-source under an MIT license.
What it unlocks
Running a fly-connectome simulation that can execute guarded crypto trades via Coinbase AgentKit.
- 166,700 neurons
- 25.6 million connections
- 3,335 brightness inputs
- 811 R8 color inputs
What you need to act on it
- Python 3.11
- C++17 compiler
- macOS/Linux
- 16 GB RAM recommended
- Coinbase Advanced portfolio with at most 100 USDC
- portfolio-scoped ECDSA API key with View + Trade permissions
Sources