Topics

#ai-trading #algorithmic-trading #binance #bybit #ccxt #crypto #crypto-trading-bot #cryptocurrency #deep-learning #gateio #kucoin #lightgbm #lstm #machine-learning #neural-network #okx #open-source #pump-scanner #python #trading-bot

Language breakdown

Python 96.6%

Jupyter Notebook 2.4%

Batchfile 0.5%

Shell 0.5%

Dockerfile 0.1%

▸ Files click to expand

README

DeepAlpha V11.0

AI-Powered Crypto Trading Bot — 18 Exchanges, 84.6% Accuracy

Website · Dashboard · Telegram Bot · Discord · Channel

🚀 Two Ways to Use DeepAlpha

☁️ Cloud Version (Recommended) — No Setup Required

The easiest way to start. Bot runs 24/7 on our servers.

| | | |---|---| | ✅ No VPS, no Docker, no Python | ✅ 12 exchanges supported | | ✅ Start in Paper Mode (zero risk) | ✅ Real-time dashboard | | ✅ Pump scanner included | ✅ TradingView webhooks | | ✅ Exchange-side stop-loss | ✅ Telegram bot control |

Start Free 7-Day Trial →

No credit card required. Start in Paper Mode — switch to Live when ready.

🖥️ Self-Hosted Version — Full Control

For developers who want to customize the code.

git clone https://github.com/stefanoviana/deepalpha.git
cd deepalpha
pip install -r requirements.txt
cp .env.example .env  # add your API keys
python deepalpha.py

Supports 12 exchanges: Bybit, Binance, OKX, Gate.io, KuCoin, Bitget, HTX, MEXC, BingX, Phemex, BitMart, WhiteBIT

⚠️ About AI Models

The self-hosted version includes training scripts but does NOT include pre-trained models. You have two options:

Option A: Train Your Own Models (Free)

This requires a machine with 8GB+ RAM and takes several hours. Your model accuracy will depend on data quality and training parameters.

Option B: Use Our Pre-Trained Models (Cloud Pro)

Our production models achieve 70.9% walk-forward validated accuracy after months of optimization with:

72 engineered features
12 months of training data
Optuna hyperparameter tuning (200+ trials)
4-window walk-forward validation
Continuous weekly retraining

These models are exclusive to Cloud Pro users and are the same models powering our live trading (verified +05 PnL this week).

Get Pre-Trained Models — 7 Days Free →

No credit card required. Includes models + dashboard + pump scanner + exchange stop-losses.

About AI Models

The self-hosted version includes training scripts but does NOT include pre-trained models. You have two options:

Option A: Train Your Own Models (Free)

python download_data.py   # Download historical data from Bybit
python train.py           # Train XGBoost + LightGBM models (~2-4 hours)

This requires 8GB+ RAM and takes several hours. Your accuracy depends on data quality and parameters.

Option B: Use Our Pre-Trained Models (Cloud Pro)

Our production models achieve 84.6% walk-forward validated accuracy after months of optimization:

72 engineered features
12 months of curated training data
Optuna hyperparameter tuning (200+ trials)
4-window walk-forward validation
Weekly retraining with fresh market data

These models are exclusive to Cloud Pro users — the same models powering our live trading.

Get Pre-Trained Models — 7 Days Free (no credit card)

Cloud vs Self-Hosted

| Feature | Cloud (Pro) | Self-Hosted (Free) | |---|---|---| | Setup time | 30 seconds | 30+ minutes | | AI Bot | ✅ 24/7 managed | ✅ You manage | | Pump Scanner | ✅ Included | ✅ Included | | Dashboard | ✅ Web + Mobile | ❌ CLI only | | Telegram Bot | ✅ Full control | ❌ | | TradingView Webhooks | ✅ | ❌ | | Auto-restart on crash | ✅ | ❌ | | Exchange-side SL | ✅ | ❌ | | Paper Mode | ✅ | ❌ | | Price | $39/mo (7 days free) | Free forever |

Try Cloud Free → | Self-Host Guide →

Cloud-hosted. No installation. No VPS. No coding. The AI trades 24/7 for you.

Telegram Bot — Trade from Your Phone

Manage everything from Telegram with @DeepAlphaVault_bot:

| Command | What it does | |---------|-------------| | /start | Welcome & account status | | /login | Link your DeepAlpha account | | /start_bot | Start AI / Grid / DCA Bot | | /stop_bot | Stop bot | | /status | Open positions & PnL | | /balance | Exchange balance | | /pnl | PnL summary | | /trades | Recent trades | | /keys | Setup exchange API keys |

Start/stop bots, check positions, get real-time trade notifications — all from Telegram. Open Bot →

Pump Scanner — Catch Explosive Moves

The built-in pump scanner monitors 500+ coins in real-time for volume spikes and automatically trades the pump.

How it works:

Scans all coins every 2 minutes for volume spikes (5x+ above average)
Opens a LONG position automatically when pump detected
Takes profit in 3 tiers: TP1 (+5%), TP2 (+10%), TP3 (+20%)
Trailing stop locks in gains after TP1
Stop loss at -3% for protection

Recent results:

ZEREBRO: +$33 (+7.7%) in 72 minutes
B Token: +$66 (+12.7%) in 2 minutes

The pump scanner runs automatically alongside the AI Bot. No extra configuration needed.

What is DeepAlpha?

DeepAlpha is an open-source ML trading system that predicts crypto price direction on 18 exchanges including Bybit, Binance, OKX, Gate.io, KuCoin, Bitget, HTX, MEXC, BingX, Phemex, BitMart, and WhiteBIT. It uses 72 engineered features from L2 orderbook data, funding rates, and market microstructure signals.

The core model achieves 84.6% directional accuracy on walk-forward validated out-of-sample data.

Plans

| Feature | Free (GitHub) | Pro $39/mo | Lifetime $199 | |---------|:---:|:---:|:---:| | AI Model (LightGBM) | Train yourself | ✅ Pre-trained (84.6%) | ✅ Pre-trained | | XGBoost + RF Ensemble | ❌ | ✅ | ✅ | | Grid Bot (range trading) | ❌ | ✅ 5 strategies | ✅ 5 strategies | | DCA Bot (safety orders) | ❌ | ✅ 5 strategies | ✅ 5 strategies | | Pump Scanner (500+ coins) | ❌ | ✅ Auto | ✅ Auto | | Features | 15 basic | 78 (V11 full) | 78 (V11 full) | | TFT + TransformerGRU | ❌ | ✅ | ✅ | | HMM Regime Detection | ❌ | ✅ | ✅ | | Cloud Dashboard | ❌ | ✅ Real-time | ✅ Real-time | | Telegram Bot Control | ❌ | ✅ @DeepAlphaVault_bot | ✅ | | Auto Retraining (daily) | ❌ | ✅ | ✅ | | Auto Restart & Monitoring | ❌ | ✅ 24/7 | ✅ 24/7 | | 12 Exchange Support | Manual setup | ✅ 1-click | ✅ 1-click | | VPS Required | ✅ ($20-50/mo) | ❌ Cloud hosted | ❌ Cloud hosted | | Source Code | View only | ❌ | ✅ Full download | | Support | GitHub Issues | Discord | Direct Developer |

💡 Why Most Users Choose Cloud

The free version requires you to:

- Set up your own VPS ($20-50/month)

- Train the ML model yourself (hours of compute)

- Monitor and restart the bot manually

- Handle updates, crashes, and maintenance

>

The cloud version gives you everything ready in 2 minutes:

- Pre-trained AI model with 84.6% accuracy

- Grid Bot + DCA Bot + 10 pre-built strategies

- Manage from Telegram (@DeepAlphaVault_bot)

- Auto-restart, daily retraining, 24/7 monitoring

- No VPS, no installation, no maintenance

>

The VPS alone costs $20-50/month — our Pro plan is $39/month with everything included.

>

👉 Start Free 7-Day Trial — No Credit Card

How It Works

Market Data (Bybit / Binance API via ccxt)
    ↓
Feature Engineering (RSI, ATR, EMA, Momentum, Volume)
    ↓
LightGBM Model (trained on 1M+ samples)
    ↓
Signal Generation (LONG/SHORT with confidence %)
    ↓
Risk Management (position sizing, stop-loss, max positions)
    ↓
Trade Execution (Bybit / Binance API via ccxt)

What's New

V11.1 (April 2026) — Liquidation map + pump scanner:

Liquidation level estimator — detects where stop-loss clusters are, optimizes TP/SL placement
Pump scanner — real-time detection of volume spikes and new listings on Bybit + Binance
Binance cross-exchange alerts — detects Binance listings that pump on Bybit

V11.0 (April 2026) — Major accuracy upgrade:

72 features (10 new: Hurst exponent, VPIN, volatility regime, fractal efficiency, multi-timeframe alignment)
84.6% walk-forward validated accuracy (up from 60%)
TFT (Temporal Fusion Transformer) + TransformerGRU neural models
HMM 3-state regime detection (bull/bear/sideways)
Dynamic ATR-based TP/SL with multi-target take profit
Cloud dashboard with backtest and live signals
Daily LSTM auto-retraining

How the AI Works

DeepAlpha uses a LightGBM gradient boosting model trained on historical crypto data with walk-forward validation.

Training Pipeline:

Download 1 year of hourly candles for 20 coins
Compute 15 technical features (RSI, ATR, EMA, momentum, etc.)
Generate labels: does price move significantly in the next hour?
Split data chronologically: 70% train, 15% validation, 15% test
Train with early stopping to prevent overfitting
Evaluate on test set (never seen during training)

Key Design Decisions:

Walk-forward validation (not random split) — prevents overfitting
Conservative features — only battle-tested technical indicators
Early stopping — model stops training when validation loss increases
Symmetric labels — model predicts both up and down moves

The Pro version adds XGBoost ensemble, PPO reinforcement learning, 50 features, and ATR-based prediction targets.

Live Results (Verified on Bybit)

| Metric | Value | |--------|-------| | Weekly ROI | +25.7% | | AI Accuracy | 84.9% (walk-forward validated) | | Profit/Loss Ratio | 2.61:1 | | Sharpe Ratio | 1.27 | | AI-Trained Coins | 25 | | Avg Holding Time | 5.58 hours |

Deep neural network (BiLSTM + Attention) trained on GPU across 25 cryptocurrency pairs.

Quick Start (Pro — 2 minutes)

Windows: double-click setup.bat Mac/Linux: run bash setup.sh

The setup wizard asks for your license key and exchange API keys (Bybit, Binance, or Bitget), then starts trading automatically. The AI model is downloaded from our server — no training needed.

Manual Install

git clone https://github.com/stefanoviana/deepalpha.git
cd deepalpha
pip install -r requirements.txt
cp .env.example .env        # edit with your keys
python deepalpha.py          # start trading

Docker

git clone https://github.com/stefanoviana/deepalpha.git
cd deepalpha
cp .env.example .env        # edit with your keys
docker compose up -d deepalpha

Free Version (train your own model)

python download_data.py      # download historical data
python train.py              # train model (~5 min)
python deepalpha.py           # start trading

Trading Strategies

DeepAlpha combines multiple strategies for maximum edge:

1. AI Directional Prediction (Primary)

XGBoost + LightGBM ensemble predicts price direction with 84.6% accuracy using 72 features from L2 orderbook data, funding rates, and market microstructure.

2. HMM Regime Detection

Hidden Markov Model identifies bull/bear/sideways regimes. The bot adapts: wider TP in bull, tighter SL in bear, reduced activity in sideways.

3. Liquidation Level Analysis

Estimates where stop-loss clusters are based on open interest and leverage distribution. Uses liquidation zones to:

Place TP before liquidation cascades (take profit before bounce)
Place SL beyond danger zones (avoid getting caught in cascades)
Boost confidence when liquidation cascades favor our direction

4. Pump Scanner

Real-time detection of volume spikes and new listings across Bybit and Binance. Auto-enters pumps with tight risk management (5% equity, 2h max hold).

5. Multi-Target Take Profit

T1 (0.8x ATR): Close 33% — lock some profit early
T2 (1.3x ATR): Close 33% — let winner run
Trailing: ATR-based trailing stop on remaining 34%

6. Auto-Unstuck

Graduated exit when trade goes wrong: -2% close 25%, -3% close 25%, -4% close 25%, -5% hard cap close all.

4. Run

python deepalpha.py        # Start AI trading bot
python pump_scanner.py     # Start pump scanner (standalone)

Pump Scanner

Real-time pump detection system that monitors all 500+ Bybit USDT perpetual pairs every 3 seconds.

What it detects

Volume spikes: 5x+ normal volume in a single candle
Price momentum: +3% move with 3+ consecutive green candles
New listings: Monitors Bybit announcements API for new perpetual contracts
Dump exhaustion: RSI > 80 + volume decline for short entries

How it trades

Detects pump signal (volume + price + RSI + buy ratio confirmation)
Opens long with 5x leverage, ATR-based stop loss
TP cascade: TP1 (+5%) close 40% | TP2 (+10%) close 30% + trailing | TP3 (+20%) close rest
Circuit breaker: stops after -$50 daily loss

Quick Start

# 1. Add to your .env file:
BYBITAPIKEY=yourapikey
BYBITAPISECRET=yourapisecret
2. Optional: Telegram alerts
TELEGRAMTOKEN=yourbot_token
TELEGRAMCHATID=yourchatid
3. Run standalone:
pip install ccxt numpy requests python-dotenv
python pump_scanner.py

Configuration

See pumpconfig_example.env for all tunable parameters:

Detection thresholds (volume multiplier, price spike %, RSI range)
Position sizing (leverage, risk budget, max positions)
TP/SL levels (ATR-based SL, cascade TP at 5%/10%/20%)
Fakeout filters (consecutive candles, buy ratio, minimum volume)

Run with Docker

docker compose up -d pump-scanner

Run alongside the AI bot

from pumpscanner import createpumpscannerfrom_config
scanner = createpumpscannerfromconfig() if scanner:     scanner.start()  # runs in background thread

TradingView Webhook

Execute your TradingView alerts automatically on any supported exchange. No coding needed.

Setup (2 minutes)

Get your webhook key from the Dashboard
In TradingView, create an alert and set the webhook URL:

https://deepalphabot.com/api/webhook/tradingview

Set the alert message (JSON):

{
     "action": "buy",
     "symbol": "BTC",
     "qty": 0.01,
     "key": "yourwebhookkey"
   }

Supported Actions

| Action | Description | |--------|-------------| | buy | Open long / buy market order | | sell | Open short / sell market order | | close | Close entire position |

Auto-sizing

Set qty to 0 and DeepAlpha will automatically size the position (5% of equity at 5x leverage).

Works with all 18 exchanges

Bybit, Binance, OKX, Gate.io, KuCoin, Bitget, HTX, MEXC, BingX, Phemex, BitMart, WhiteBIT — same webhook URL, trades on whichever exchange you connected.

Architecture

deepalpha/
├── deepalpha.py          # Main AI trading bot
├── pump_scanner.py       # Real-time pump detection (standalone)
├── exchange_adapter.py   # Multi-exchange adapter layer
├── train.py              # AI training pipeline
├── download_data.py      # Data downloader
├── features.py           # Feature engineering (72 features)
├── risk_manager.py       # Position sizing & risk
├── config.py             # Configuration
├── pumpconfigexample.env  # Pump scanner config template
└── requirements.txt      # Dependencies

Cloud Platform — Premium Features

The DeepAlpha Cloud Platform includes additional features not available in the open-source version:

| Feature | Open Source | Cloud Platform | |---------|-----------|----------------| | AI Trading Bot | ✅ | ✅ | | Backtesting | ✅ | ✅ Enhanced | | Grid Trading Bot | ❌ | ✅ 5 strategies | | DCA Bot + Safety Orders | ❌ | ✅ 5 strategies | | 12 Exchange Support | Bybit only | ✅ All 12 | | Real-time Dashboard | ❌ | ✅ | | TradingView Webhooks | ❌ | ✅ | | Auto-restart | ❌ | ✅ | | Telegram Alerts | Basic | ✅ Advanced |

Try free for 7 days — no credit card required: deepalphabot.com

What you get:

Live AI signals — see every prediction in real-time
Auto-trading — connect your Bybit or Binance API, the bot trades for you
Grid Bot — 5 pre-built grid strategies for range-bound markets
DCA Bot — dollar-cost averaging with safety orders for volatile markets
Backtest viewer — test the AI on historical data with custom parameters
Equity curve — track your portfolio performance over time
Trade history — every trade logged with PnL and exit reason
Email alerts — welcome sequence + trial expiry notifications

How it works:

Register at deepalphabot.com/cloud (7-day free trial)
Connect your exchange API keys (Bybit, Binance, Bitget — read-only supported)
Configure risk settings (leverage, max positions, confidence threshold)
The AI trades automatically — check your dashboard anytime

Security:

API keys encrypted with AES-256 Fernet encryption
Read-only API supported — the bot can trade but never withdraw
Keys stored only on our server, never shared
SSL/TLS 1.3 + HSTS on all connections

Cloud Stack:

Backend: FastAPI + PostgreSQL + JWT auth
AI Engine: Same V11 models as self-hosted (LightGBM + XGBoost + TFT)
Hosting: Dedicated VPS with PM2 process management
Email: Resend transactional email service

Pricing:

| Plan | Price | Features | |------|-------|----------| | Free Trial | $0 for 7 days | Full access, no credit card | | Pro | $39/month | Auto-trading + dashboard + signals | | Lifetime | $199 one-time | Everything forever + source code |

FreqAI Plugin

Already using Freqtrade? Use DeepAlpha's ML pipeline as a FreqAI plugin.

cp freqai-plugin/deepalphamodel.py yourfreqtrade/freqai/prediction_models/

Features:

Triple Barrier Labeling (84.6% accuracy vs ~55-60% standard FreqAI)
SHAP Feature Selection (auto-removes noise features)
Meta-Labeling (filters bad trades)
Walk-Forward CV with 24h embargo

See freqai-plugin/README.md for setup guide.

Supported Exchanges

DeepAlpha supports multiple exchanges via the ccxt library. Switch exchanges with a single env var.

| Exchange | Type | Status | Markets | |----------|------|--------|---------| | Bybit | USDT Perpetual | Supported | 3,200+ | | Binance | USDT-M Futures | Supported | 4,300+ | | OKX | USDT Perpetual | Supported | 3,000+ | | Gate.io | USDT Perpetual | Supported | 6,300+ | | KuCoin | USDT Perpetual | Supported | 1,600+ | | Bitget | USDT-M Futures | Supported | 1,300+ | | HTX | USDT Perpetual | Supported | 900+ | | MEXC | USDT Perpetual | Supported | 2,400+ | | BingX | USDT Perpetual | Supported | 800+ | | Phemex | USDT Perpetual | Supported | 300+ | | BitMart | USDT Perpetual | Supported | 400+ | | WhiteBIT | USDT Perpetual | Supported | 200+ |

Configuration — Bybit (default)

EXCHANGE=bybit
BYBITAPIKEY=yourapikey
BYBITAPISECRET=yourapisecret
BYBIT_TESTNET=false
LEVERAGE=5
MAX_POSITIONS=5
TELEGRAMTOKEN=yourbot_token     # optional
TELEGRAMCHATID=yourchatid     # optional

Configuration — Binance Futures

EXCHANGE=binance
BINANCEAPIKEY=yourapikey
BINANCEAPISECRET=yourapisecret
BINANCE_TESTNET=false
LEVERAGE=5
MAX_POSITIONS=5

Configuration — OKX

EXCHANGE=okx
OKXAPIKEY=yourapikey
OKXAPISECRET=yourapisecret
OKXPASSPHRASE=yourpassphrase
LEVERAGE=5
MAX_POSITIONS=5

Configuration — Gate.io

EXCHANGE=gateio
GATEIOAPIKEY=yourapikey
GATEIOAPISECRET=yourapisecret
LEVERAGE=5
MAX_POSITIONS=5

Configuration — KuCoin

EXCHANGE=kucoin
KUCOINAPIKEY=yourapikey
KUCOINAPISECRET=yourapisecret
KUCOINPASSPHRASE=yourpassphrase
LEVERAGE=5
MAX_POSITIONS=5

Configuration — Bitget

EXCHANGE=bitget
BITGETAPIKEY=yourapikey
BITGETSECRET=yourapi_secret
BITGETPASSPHRASE=yourpassphrase
LEVERAGE=5
MAX_POSITIONS=5

Risk Management

Leverage: 5x (configurable, max 10x)
Position size: 5% of equity per trade
Stop-loss: Fixed 1.5% from entry
Take-profit: 3% from entry (2:1 R:R minimum)
Max positions: 3 simultaneous
Daily loss limit: 5% of equity
Circuit breaker: Pauses after 3 consecutive losses

Performance

Walk-forward validated results (V11 model, out-of-sample):

| Metric | Value | |--------|-------| | Directional Accuracy | 84.6% | | Profit Factor | 2.91 | | Max Drawdown | 20.7% | | Sharpe Ratio | 0.97 | | Avg Win / Avg Loss | 4.6:1 |

Methodology: Walk-forward validation with expanding window and 24h embargo between train/test. All metrics are on strictly out-of-sample data, never seen during training.

Model Architecture

| Model | Accuracy | Role | |-------|----------|------| | LightGBM | 84.6% | Primary | | XGBoost | 70.8% | Ensemble | | Random Forest | 69.3% | Ensemble | | TransformerGRU | 65.8% | Neural | | TFT | 63.7% | Neural |

72 features including: L2 orderbook proxies, funding rate momentum, cross-asset correlation, volatility regime detection, Hurst exponent, VPIN, multi-timeframe alignment.

Troubleshooting

Having issues? Here are the most common problems and how to fix them.

Missing dependencies

[ERROR] Missing dependencies: lightgbm, ccxt

Fix: Install all required packages:

pip install -r requirements.txt

If you get permission errors, use pip install --user -r requirements.txt or a virtual environment:

python -m venv venv source venv/bin/activate   # Linux/Mac venv\Scripts\activate      # Windows pip install -r requirements.txt

Missing .env file

[ERROR] No .env file found in current directory.

Fix: Copy the example and fill in your API keys:

cp .env.example .env nano .env   # or any text editor

You need at minimum:

EXCHANGE=bybit (or binance, okx, bitget, etc.)
The matching API key + secret for your exchange

Missing model files

[ERROR] Model not found at model_1h.pkl

Fix: You need a trained ML model. Two options:

Train your own (free):

python download_data.py   # download historical data    python train.py           # train model (~5 min)

Use the cloud (no model needed): deepalphabot.com - 7-day free trial

API key errors

AuthenticationError: Invalid API key

Fix:

Double-check your API key and secret in .env (no extra spaces or quotes)
Make sure the API key has trading permissions enabled
For Bybit: enable "Contract - Trade" permission
For Binance: enable "Futures" permission and whitelist your IP
For Bitget/OKX/KuCoin: make sure you also set the passphrase

Exchange not supported

ValueError: Unsupported exchange: xxx

Fix: Set EXCHANGE in your .env to one of the supported values: bybit, binance, okx, bitget, gateio, kucoin, htx, mexc, bingx, phemex, bitmart, whitebit, hyperliquid

Bot crashes in a loop

The bot waits 30 seconds before exiting after a crash to prevent restart-loop spam (PM2, Docker, systemd). Check the error message above the crash for the root cause.

Still stuck?

Setup guide: deepalphabot.com/setup-guide
Telegram support: @DeepAlphaVault_bot
Discord: discord.gg/P4yX686m
GitHub Issues: github.com/stefanoviana/deepalpha/issues

Or just use the cloud version — no installation, no setup, no crashes: deepalphabot.com

Disclaimer

Trading involves significant risk of loss. This software is provided as-is, with no guarantees of profitability. Past backtest performance does not indicate future results. The performance metrics above are from historical backtests and may not reflect live trading conditions. Only trade with money you can afford to lose.

Contributing

We welcome contributions! Whether it's new features, bug fixes, documentation, or exchange adapters — every PR helps.

Read the Contributing Guide
Check open issues for ideas
Fork, branch, code, and submit a PR

High-impact areas: new technical indicators, exchange adapters (dYdX, OKX), tests, and documentation.

Community

Discord — General discussion, help, feature ideas
Telegram — Announcements and trade signals
Blog — Technical deep dives on the architecture
Docs — Full documentation (coming soon)

License

MIT License — see LICENSE

Integrations

Crypto Pump Scanner — Real-time pump detection for Bybit. Monitors 500+ pairs, auto-trades volume spikes. Standalone or integrated.
FreqAI — Use DeepAlpha's ML pipeline as a drop-in FreqAI prediction model. See freqai-plugin/ for setup.
Binance Futures — Full support for Binance USDT-M perpetual futures via ccxt.
Telegram — Real-time trade alerts and portfolio updates via bot.