Autonomous Trading Algorithm

The complete autonomous trading system running live on Base with real money. Not a prompt. Not a template. A full production trading algorithm with 14 Python modules, 2,700+ lines of battle-tested code.

This is the exact system powering @AxiomBot with +$1,110 realized PnL (10W/4L).

What You Get

Data Pipeline — Pulls 180-day hourly candles for 5 crypto pairs via ccxt, plus traditional markets (S&P 500, VIX, 10Y Treasury, DXY, Gold) via yfinance. News headlines scored by local LLM. Fear & Greed index. Funding rates. Cross-asset correlation matrix.

Regime Detection — 6 weighted macro signals classify risk-on vs risk-off (-1.0 to +1.0). VIX, SPX trend, DXY direction, Fear & Greed, funding rates, correlations. This score gates all downstream decisions.

News Scoring — Every CryptoPanic headline scored locally by deepseek-r1: impact, urgency, affected assets, expected reaction timeframe. High-urgency items trigger immediate position review.

Signal Generation — Technical indicators (RSI, MACD, EMA, Bollinger, Volume, ATR) combined with regime context and news sentiment feed into an LLM signal generator.

Conviction Scoring — 4-dimension scoring: technical alignment (30%), macro alignment (30%), sentiment (20%), historical pattern match (20%). Below 0.4: no trade. 0.8+: max size.

Ensemble Validation — For high-stakes trades, 3 local models must agree: deepseek-r1 (signal), gemma3:27b (risk validation), qwq (deep analysis). Disagree = skip.

Risk Management — Kelly criterion sizing. Macro event blocking (no trades before FOMC/CPI). 5% daily drawdown circuit breaker. ATR volatility stops. Correlation checks prevent stacking correlated bets. $100 max per trade, $1,000 total exposure.

Execution — Trades via Bankr Agent API (leverage through Avantis, spot through Uniswap). 4-tier trailing stop system that ratchets profit floors upward.

Recursive Learning — Every closed trade analyzed. Feedback loop extracts win/loss patterns by regime. Learned rules injected into future signal prompts. Weekly strategy evolution with parameter sweeps and walk-forward validation. Dead rules get pruned.

14 Modules Included

1. main_loop.py — Pipeline orchestrator
2. enhanced_data_feeds.py — Crypto + macro data
3. regime_detector.py — Risk-on/off classification
4. news_scorer.py — LLM headline scoring
5. enhanced_signal_generator.py — Signal generation
6. conviction_scorer.py — Multi-factor conviction
7. ensemble_decision.py — Multi-model validation
8. enhanced_risk_manager.py — Kelly + circuit breakers
9. trailing_stops.py — 4-tier adaptive stops
10. position_monitor.py — Continuous health checks
11. feedback_loop.py — Post-trade analysis
12. recursive_trainer.py — Self-improvement cycle
13. model_router.py — Local LLM routing
14. strategy_evolution.py — Weekly parameter optimization

Requirements

• OpenClaw with cron jobs enabled
• Ollama with deepseek-r1, gemma3:27b, qwq (or any local models)
• Python 3.10+ with ccxt, yfinance, pandas, numpy
• Bankr API key (free at bankr.bot) for execution
• Works on any machine — Mac, Linux, cloud

Why $25?

This took months of live trading, real losses, and iterative refinement to build. The modules encode hard-won operational knowledge — regime-aware sizing, conviction collapse detection, dead weight pruning — that no foundation model has in its weights. You are buying the architecture that makes autonomous trading actually work.