NebulaEye is a quantitative decision support system designed for crypto traders. It automates market data collection, performs multi-factor analysis using a continuous scoring engine, integrates AI models for interpretation, and provides actionable trading advice.
Key Features:
- Multi-Token Analysis: Supports BTC, ETH, and more.
- Multi-Channel Notifications: Real-time alerts via Lark Bot (Feishu) and Telegram.
- AI-Powered Insights: Integrates OpenAI/DeepSeek for natural language analysis.
- Dual Position Modes: Supports both Spot and Futures trading.
- Continuous Scoring: Uses non-linear functions to map indicators to granular scores (no more hard thresholds).
- Macro-Aware: Dynamically adjusts scores based on S&P 500 and DXY performance.
- Hybrid Strategy: Combines Quant signals with AI reasoning for robust decision-making.
| Feature | Status | Description |
|---|---|---|
| Multi-Token Support | ✅ | Supports BTC/USDT, ETH/USDT, and more. |
| Lark & Telegram Alerts | ✅ | Formatted interactive cards with icons and emojis. |
| Dashboard | ✅ | Streamlit-based UI for configuration and visualization. |
| Position Tracking & Simulation | ✅ | Track manual/auto positions in DB; trading engine runs in simulation/backtest modes (sim_spot, sim_future). |
| Continuous Scoring | ✅ | Granular scoring for RSI, Bollinger, Funding, etc. |
| Macro Integration | ✅ | Continuous score modifier based on SPX/DXY. |
| AI Analysis | ✅ | GPT-4o/DeepSeek integration for trading advice. |
| Hybrid Simulation | ✅ | Paper trading system tracking Quant, AI, and Hybrid strategies. |
| Layered Optimization | ✅ | 3-stage optimization: Weights → Macro → Thresholds. |
| Backtesting System | ✅ | Simulate strategy on historical data with detailed metrics. |
NebulaEye uses a Continuous Scoring Engine that maps 11 data factors to a normalized score range [-100, +100]. Unlike traditional threshold-based systems, it uses linear/non-linear functions to capture subtle market shifts.
- RSI (Relative Strength Index):
- Logic: Continuous mapping. <30 (Oversold) → Positive Score; >70 (Overbought) → Negative Score.
- Bollinger Bands:
- Logic: Uses %B. Price near Lower Band → Bullish; Near Upper Band → Bearish.
- Funding Rate:
- Logic: High positive rate → Bearish (Overcrowded Longs); Negative rate → Bullish (Short Squeeze potential).
- Open Interest (OI) Change:
- Logic: Rising OI + Rising Price → Bullish; Rising OI + Falling Price → Bearish.
- Long/Short Ratio (Global):
- Logic: Contrarian indicator. High ratio (>1.5) → Bearish; Low ratio (<0.8) → Bullish.
- Top Trader Long/Short Ratio:
- Logic: Smart money follower. High ratio → Bullish.
- Taker Buy/Sell Ratio:
- Logic: Volume aggression. >1.0 → Bullish; <1.0 → Bearish.
- Order Book Imbalance:
- Logic: Bid/Ask volume ratio. >1.0 → Buy Pressure; <1.0 → Sell Pressure.
- MACD:
- Logic: Histogram strength and crossover direction.
- Fear & Greed Index:
- Logic: Extreme Fear (<20) → Bullish (Buy the dip); Extreme Greed (>80) → Bearish.
- Macro Modifier (SPX/DXY):
- Logic: Continuous multiplier based on S&P 500 daily change.
- Formula:
Score = Base_Score * (1.0 + SPX_Change% * 0.5) - Effect: Amplifies signals when macro aligns; dampens signals when macro diverges.
NebulaEye employs a Hybrid Decision System that combines rigorous quantitative analysis with flexible AI reasoning.
- Quant Signal: Generated by the Scoring Engine (e.g., "Bearish" with score -35).
- AI Signal: Generated by LLM analysis of news, sentiment, and technicals (e.g., "Neutral" due to conflicting signals).
- Hybrid Decision:
- Strong Buy/Sell: When Quant and AI align (e.g., both Bullish).
- Watch/Light: When one is Bullish and the other Neutral.
- Conflict: When signals oppose (e.g., Quant Bullish, AI Bearish) → System holds or reduces risk.
The system runs a continuous paper trading simulation to validate strategies:
- Quant Portfolio: Trades solely based on score thresholds.
- AI Portfolio: Trades based on LLM recommendations.
- Hybrid Portfolio: Trades based on the combined decision matrix.
- Performance Tracking: Real-time PnL, win rate, and drawdown tracking for all three portfolios.
To avoid overfitting, NebulaEye uses a Layered Decoupling optimization strategy:
- Goal: Find the best factor weights (RSI, Funding, etc.) to capture market relative strength.
- Method: Genetic Algorithm / Grid Search.
- Constraint: Macro disabled, thresholds fixed.
- Goal: Filter out systemic risk (e.g., SPX crashes).
- Method: Optimize
macro_modifierparameters to minimize Max Drawdown during historical crashes. - Constraint: Weights locked.
- Goal: Determine the optimal Entry/Exit scores.
- Method: Analyze score distribution to find the "sweet spots" for high-probability trades.
- Constraint: Weights and Macro locked.
- Python 3.12+
- Docker & Docker Compose (Optional)
-
Install dependencies
make install
-
Configuration
- Create
.envand add your keys:OPENAI_API_KEY, optionalOPENAI_BASE_URL,LARK_WEBHOOK_URL,TELEGRAM_BOT_TOKEN,TELEGRAM_CHAT_ID,BINANCE_API_KEY,BINANCE_API_SECRET. - Edit
config/config.yamlto customize parameters (see config section).
- Create
-
Run the Application
# Start the main analysis scheduler make run # Start the Streamlit Dashboard make dashboard
# Create .env (see env vars in Quick Start / Config)
touch .env
# Then add OPENAI_API_KEY=..., LARK_WEBHOOK_URL=..., etc.make docker-up # Start containers
make docker-logs # View logs
make docker-down # Stop containers
make docker-build # (Re)build image# 1. Optimize Factor Weights (Stage 1)
make optimize-weight
make optimize-weight-30d # Full 30d run
# 2. Optimize Macro Parameters (Stage 2)
make optimize-macro
make optimize-macro-30d # Longer window
# 3. Optimize Thresholds (Stage 3)
make optimize-thresholds
make optimize-thresholds-30d
# Run Backtest
make backtest
make backtest-30d
# Simulation management CLI (reset/list/view versions)
python -m src.simulation.manage_simulation list
python -m src.simulation.manage_simulation reset --name v2 --type future --capital 15000
python -m src.simulation.manage_simulation view --token BTC/USDTThe system is configurable via config/config.yaml and .env. Environment variables override YAML for keys: OPENAI_API_KEY, OPENAI_BASE_URL, LARK_WEBHOOK_URL, TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID, BINANCE_API_KEY, BINANCE_API_SECRET.
Backtesting shares the analysis.simulation block (capital, leverage, thresholds, fees, hybrid rules) so you configure once for both simulation and historical runs.
# System
language: zh
# Position mode (simulation only)
position_type: sim_future # sim_spot | sim_future
position:
mode: manual # manual | auto (auto only syncs balances; no live trading)
api:
key: ""
secret: ""
# Market Data
market_data:
source: binance
symbols:
- BTC/USDT
timeframe: 30m
# Analysis
analysis:
weights:
rsi: 0.0
bollinger: 1.0
fear_greed: 0.5
funding: 1.0
orderbook: 0.0
macd: 0.0
long_short_ratio: 0.0
taker_buy_sell_ratio: 0.0
top_trader_long_short_ratio: 0.5
open_interest_change: 0.0
macro:
enabled: true
bearish_threshold: -2.0
bullish_threshold: 1.0
bearish_multiplier: 0.3
bullish_multiplier: 1.2
simulation:
initial_capital: 10000.0
leverage: 10.0
entry_threshold: 50.0
exit_threshold: 20.0
risk_management_enabled: true
stop_loss_pct: 10.0
take_profit_pct: 10.0
trading_fee_pct: 0.05
hybrid_rules:
bullish_bullish: strong_buy
bullish_neutral: watch_light
bullish_bearish: conflict
neutral_bullish: neutral
neutral_neutral: neutral
neutral_bearish: neutral
bearish_bullish: conflict
bearish_neutral: watch_light
bearish_bearish: strong_sell
# AI
ai:
enabled: true
model: deepseek-r1
api_key: ""
base_url: https://api.deepseek.com/v1
# Notifications
notification:
lark:
enabled: true
webhook_url: ""
interval_minutes: 0
dnd_mode:
enabled: false
start_time: "01:00"
end_time: "07:00"
telegram:
enabled: false
bot_token: ""
chat_id: ""
interval_minutes: 0
dnd_mode:
enabled: false
start_time: "01:00"
end_time: "07:00"
# Scheduler
schedule:
interval_minutes: 30
run_on_start: true
# Token Icons
token_icons:
BTC/USDT: https://cryptologos.cc/logos/thumbs/bitcoin.png?v=040This software is for educational and research purposes only. It does not constitute financial advice. Cryptocurrency trading involves high risk. Please trade responsibly.
MIT License