A fly-connectome simulation that can operate a crypto trading account. Actual neural output, actual Coinbase integration. Profitable learning has not been demonstrated.
How it works: Public Coinbase prices become an RGB chart. It stimulates 3,335 brightness inputs and 811 R8 color inputs in the retained MaleCNS v1.0 graph: 166,700 neurons, 25.6 million connections. A fixed neural readout proposes buy, sell or hold. A custom Coinbase AgentKit ActionProvider checks limits and places spot orders through Coinbase Advanced.
Positive portfolio P&L stimulates 15 identified PAM11 dopamine cells; negative P&L stimulates two PPL101 aversive dopamine cells. A candidate memory rule changes existing KC-to-MBON connections. These are engineered reinforcement signals, not modeled pain receptors. Synaptic changes do not establish that it learns to trade profitably. Model and evidence.
Python 3.11, a C++17 compiler, macOS/Linux. Allow several GB for the dataset and dependencies; 16 GB RAM recommended.
python3.11 -m venv .venv
source .venv/bin/activate
pip install -e '.[test]'
python -m stonkfly prepare
python -m stonkfly runDefault: paper trades, real public BTC-USDC data, $100 simulated balance. No key needed. Local logs, sensory images and resumable brain state go in runs/paper/. Ctrl-C stops it; the same command resumes.
For real orders, first create a dedicated Coinbase Advanced portfolio with at most 100 USDC and a portfolio-scoped ECDSA API key with View + Trade, no Transfer. Copy .env.example to .env, fill it in locally, then run these commands yourself:
python -m stonkfly run --live --preflight-only
python -m stonkfly run --liveDefaults: $10 maximum order including reserved fees, 24 attempts/day, no shorts or leverage. A $20 drawdown stops new orders; it does not liquidate holdings or cap further losses. Operation and recovery.
python -m stonkfly status
python -m pytest -qThe repo does not come funded or connected to anyone’s account. Live execution needs your local credentials and explicit opt-in.