NanoTrend AI

NanoTrend AI — Deep Learning Trading System

Smarter entries. Cleaner risk. No martingale. No grid.

ONNX Neural Network XAUUSD Optimised Any Symbol Any Timeframe No Martingale No Grid

Overview

NanoTrend AI v1.4 replaces traditional indicator-based entries with a trained deep learning model. A Bidirectional LSTM network with attention mechanism reads 60 bars of market context and classifies each moment as SELL, HOLD, or BUY — firing only when confidence exceeds the configured threshold. The AI decides when to enter. Classic systems handle when to exit.

Every trade has a defined SL and TP. No martingale, no grid, no averaging down.

Core methodology

ONNX BiLSTM + Attention — primary signal engine A bidirectional LSTM trained on 17 months of XAUUSD M5 data. Reads 60 bars simultaneously forward and backward. Attention pooling learns which bars in the sequence matter most. 23 normalised features including H1 trend context, session timing, volume pressure, and volatility regime. Updated every new M5 bar. Only fires when model confidence ≥ threshold (default 0.62 — tunable).

ADX gate — choppiness filter Minimum ADX threshold blocks entries during low-momentum, ranging markets. Works alongside the ONNX model to prevent high-confidence signals in directionless price action.

S/R cascade emergency exit — fast risk management Two or more S/R flips of the same type within 15 minutes closes all opposing positions immediately. Operates between model decisions — reacts within seconds, not bars.

5-timeframe alignment — MTF confirmation M1 · M5 · M15 · H1 · H4 · D1 chain driven by a single EntryTF input. Three or more timeframes must agree before the local indicator fallback fires.

TEMA/DEMA lag reduction TEMA on fast EMA (~75% less lag), DEMA on medium/slow (~50% less lag) applied across all timeframes for cleaner trend detection in the local signal path.

Probability engine — entry quality scoring Every local-path trade receives a probability score before execution. Low-probability setups are filtered. OpenRouter AI optionally adjusts confidence using live market context.

23-feature ONNX input vector

Key features

Protection systems

• News blackout — US data (12:30 UTC), FOMC, Fridays, end-of-month windows blocked
• Cascade S/R exit — emergency close on consecutive structure breaks within 15 minutes
• Big loss pause — cooldown after any loss exceeding threshold
• MaxFloatLoss — closes at a set fraction of SL distance, not a fixed dollar amount
• Session quality filter — reduces activity during dead and low-liquidity periods
• ONNX sanity check — validates model output probabilities before every trade

What's different in v1.4

NanoTrend AI v1.x used OpenRouter AI and indicator scoring as its primary entry mechanism. v1.4 introduces a trained ONNX model as the decision engine:

• Before: indicators score a setup → AI confirms → trade fires
• After: ONNX model evaluates 60 bars of 23 features → confidence above threshold → trade fires. OpenRouter AI remains available as a secondary layer.

Pure ATR SL/TP replaces the previous S/R-targeted TP, ensuring every backtest and live trade uses the same risk parameters the model was trained on.

Specifications

Quick start

1. Copy NanoTrendAI_ONNX.onnx and NanoTrendAI_Scaler.csv to MQL5/Files/
2. Attach EA to XAUUSD M5 chart and enable AutoTrading
3. (Optional) Get a free API key at openrouter.ai and paste into APIKey input for AI layer
4. Enable WebRequest for https://openrouter.ai in MT5 Tools → Options if using AI
5. Adjust ONNXThreshold (default 0.62) — higher = fewer but more selective entries

Past performance does not guarantee future results. Trading involves substantial risk. Only trade with capital you can afford to lose.