GPU-accelerated AI that trains, optimizes, and trades automatically — so you stop guessing and start compounding.
Six Transformer blocks followed by a bidirectional LSTM — trained end-to-end on your crypto price data to capture both local patterns and long-range dependencies.
Gradient-boosted trees run in parallel with the neural net. Both models vote; their confidence is weighted and fused. Disagreement means no trade.
40-trial TPE search across learning rate, ATR multiplier, regime weight, horizon, and sequence length — simultaneously, not one variable at a time.
Automatic pivot detection labels every bar as BUY, HOLD UP, HOLD DOWN, or SELL using ATR-scaled thresholds. The model trains on structure, not noise.
One button. Eight stages. Loads 45k bars, optimizes model hyperparameters, trains, validates, backtests, and saves the best configuration — unattended overnight.
Holdout bars are withheld from training entirely. The final stage tests on data the model has never seen — not a backtest artifact, actual out-of-sample performance.
Paper trading through Alpaca's API. Fernet-encrypted, machine-bound API keys. We strongly recommend paper mode to evaluate performance before considering any live deployment.
Full CUDA support for training and inference. Your RTX GPU cuts training time by an order of magnitude. CPU fallback is available if needed.
Enter your Alpaca API keys. They're encrypted with Fernet and bound to your machine — only your computer can use them.
The bot pulls up to 45,000 bars of historical data and automatically labels every bar with the ZigZag engine.
One click. The 8-stage pipeline optimizes hyperparameters, trains, validates, and saves the best configuration — overnight.
The bot runs in paper mode, monitors the market, generates signals, and logs every decision. Review performance and refine your configuration over time.
All pairs trade against USD via Alpaca Markets. The active pair is configurable at any time — switch symbols, retrain, and redeploy in minutes.
Any CUDA-capable NVIDIA GPU works. The bot automatically detects your hardware and uses it for training and inference. CPU fallback is available but training will be significantly slower.
Minimum recommended: 6 GB VRAM. 8 GB+ recommended for full 45k-bar training at default sequence lengths. The RTX 5070 Ti (15.9 GB) and above handle all configurations comfortably.
This isn't a moving-average crossover. The model was built from scratch for crypto price action — continuous sequences, regime shifts, and multi-scale patterns that standard indicators miss entirely.
Scored across five dimensions — AI sophistication, optimization quality, backtesting rigor, customization depth, and value — against the ten most popular retail trading bots.
Secure checkout via Stripe. License key delivered by email within minutes.