Demo video: https://youtu.be/6C6sChkfbv0
Advanced cryptocurrency trading agent powered by Qwen 32B LLM and Mastra framework, designed for high-probability trading opportunities on Hyperliquid DEX.
This intelligent trading agent combines advanced AI reasoning with professional trading strategies to identify and execute high-probability trades. The system focuses on data-driven analysis, emphasizing historical win rates, technical pattern recognition, and dynamic risk management scaled to account size.
- Win Rate Analysis: Primary focus on assets with proven historical success rates (>60% win rate threshold)
- Advanced Pattern Recognition: Real-time identification of technical patterns using actual price data
- Dynamic Risk Management: Automatic position sizing based on account size and win rate metrics
- AI-Powered Decision Making: Qwen 32B LLM for sophisticated trade reasoning and market analysis
- Complete Trading Workflow: From market analysis to trade execution and monitoring
┌─────────────────────────────────────────────────────────────────┐
│ User Interface │
│ (Mastra Playground) │
└─────────────────────┬───────────────────────────────────────────┘
│
┌─────────────────────▼───────────────────────────────────────────┐
│ Trading Agent │
│ (Qwen 32B + Mastra) │
├─────────────────────┬───────────────────────────────────────────┤
│ │ │
│ ┌─────────────────▼─────────────────┐ ┌─────────────────────┐ │
│ │ Market Analysis Tools │ │ Account Tools │ │
│ │ │ │ │ │
│ │ • analyzeMarkets │ │ • getAccountInfo │ │
│ │ • findTopOpportunities │ │ • getPositions │ │
│ │ • recognizePatterns │ │ • getTradeHistory │ │
│ │ • generateRecommendation │ │ │ │
│ └─────────────────┬─────────────────┘ └─────────────────────┘ │
│ │ │
│ ┌─────────────────▼─────────────────┐ ┌─────────────────────┐ │
│ │ Trading Execution Tools │ │ Risk Management │ │
│ │ │ │ │ │
│ │ • calculatePositionSize │ │ • Dynamic Risk % │ │
│ │ • executeTrade │ │ • Win Rate Adj. │ │
│ │ • closePosition │ │ • Leverage Limits │ │
│ │ • updateStopLoss │ │ │ │
│ │ • monitorPositions │ │ │ │
│ └─────────────────┬─────────────────┘ └─────────────────────┘ │
└────────────────────┼─────────────────────────────────────────────┘
│
┌────────────────────▼─────────────────────────────────────────────┐
│ Hyperliquid DEX │
│ │
│ • Real-time market data • Order execution │
│ • Account management • Position monitoring │
│ • Risk metrics • P&L tracking │
└─────────────────────────────────────────────────────────────────┘
Purpose: Comprehensive market analysis with win rate focus and sentiment integration.
Workflow:
Input: symbols[], timeframe, useTopOpportunities
↓
Fetch Market Data → Calculate Win Rates → Generate Technical Indicators
↓
Integrate Fear & Greed Index → Score Opportunities → Rank by Probability
↓
Output: Market overview + Asset analysis + Recommendations
Key Metrics:
- Historical win rates (primary ranking factor)
- Sharpe ratio and risk-adjusted returns
- RSI, momentum, and trend analysis
- Volume and liquidity assessment
- Fear & Greed sentiment integration
Purpose: Identify the highest probability trading setups across all available assets.
Algorithm Flow:
Scan Universe (30+ symbols) → Filter by Volume → Calculate Win Rates
↓
Technical Analysis → Pattern Recognition → Sentiment Analysis
↓
Advanced Scoring → Rank Opportunities → Return Top 5
Scoring Weights:
- Win Rate: 40%
- Sharpe Ratio: 20%
- Technical Momentum: 20%
- Volume/Liquidity: 10%
- Sentiment Divergence: 10%
Purpose: Real-time technical pattern identification using actual price data.
Pattern Categories:
Bearish Reversal Patterns:
Rising Wedge Pattern:
/\ /\ ← Converging trendlines
/ \ / \ ← Higher highs, higher lows
/ \/ \ ← Volume declining
/______________\ ← Breakout expected downward
Bullish Continuation Patterns:
Bull Flag Pattern:
/|
/ | ┌─── ← Consolidation after strong move
/ | └─── ← Volume decreases in flag
/ | ← Breakout continues trend upward
/ | /
/ | /
Confirmation Requirements:
- Multi-timeframe alignment (1h + 4h)
- Volume validation (increase on breakouts)
- Price action confirmation
- Historical pattern success rate
Purpose: AI-powered trade recommendations combining all analysis factors.
Decision Matrix:
Win Rate Analysis + Technical Patterns + Sentiment → Confidence Score
↓
Risk Assessment + Position Sizing + Entry/Exit Levels
↓
Final Recommendation with Reasoning
Purpose: Retrieve comprehensive account information for margin and risk calculations.
Data Structure:
Account Overview:
├── Account Value: $X,XXX
├── Total Margin Used: $X,XXX
├── Available Margin: $X,XXX
├── Margin Ratio: XX%
└── Free Collateral: $X,XXX
Purpose: Monitor all open positions with real-time P&L and risk metrics.
Position Analysis:
For Each Position:
├── Symbol & Side (LONG/SHORT)
├── Size & Entry Price
├── Current P&L (USD & %)
├── Leverage & Margin Used
├── Liquidation Distance
└── Risk Alerts
Purpose: Analyze trading performance and calculate win rate statistics.
Performance Metrics:
- Total trades and P&L
- Win rate percentage
- Average win vs average loss
- Profit factor (gross profit/gross loss)
- Risk-adjusted returns
Purpose: Dynamic position sizing based on account size and risk parameters.
Risk Scaling Algorithm:
Account Size Assessment:
├── Micro (<$10): 100% risk - Account building
├── Small ($10-50): 50% risk - Aggressive growth
├── Medium ($50-500): 20% risk - Balanced approach
└── Large (>$500): 2% risk - Capital preservation
Win Rate Adjustment:
├── >70% Win Rate: 1.5x position multiplier
├── 60-70% Win Rate: 1.2x position multiplier
├── 50-60% Win Rate: 1.0x position multiplier
└── <50% Win Rate: Avoid trading
Position Calculation:
Risk Amount = Account Balance × Dynamic Risk %
Win Rate Adj = Risk Amount × Win Rate Multiplier
Stop Distance = |Entry Price - Stop Loss|
Base Size = Win Rate Adj ÷ Stop Distance
Final Size = Base Size × Leverage
Purpose: Execute trades with proper risk management and order placement.
Execution Flow:
Pre-Trade Validation → Set Leverage → Place Order → Confirm Execution
↓
Order Types:
├── Market Orders (IOC execution)
└── Limit Orders (GTC with spread protection)
Purpose: Close positions partially or completely with P&L calculation.
Closing Logic:
Position Validation → Size Calculation → Reverse Order → P&L Analysis
↓
Support:
├── Full position closure
├── Partial position closure
└── Automatic P&L calculation
Purpose: Dynamic stop loss management for risk control.
Stop Loss Management:
Cancel Existing Stops → Calculate New Level → Place Stop Order
↓
Features:
├── Trailing stop functionality
├── Pattern-based stop placement
└── Risk percentage protection
Purpose: Real-time position monitoring with alerts and risk assessment.
Monitoring System:
Position Analysis:
├── Real-time P&L tracking
├── Liquidation distance alerts
├── Margin usage warnings
├── Portfolio risk assessment
└── Automated alert generation
Risk Status Levels:
├── LOW: <50% margin usage
├── MEDIUM: 50-70% margin usage
├── HIGH: 70-90% margin usage
└── CRITICAL: >90% margin usage
Purpose: Secure authentication with Hyperliquid API credentials.
Authentication Flow:
Input: auth [private-key] [wallet-address]
↓
Validation → Connection Test → Session Creation → Account Verification
↓
Session Management:
├── 60-minute session timeout
├── Automatic cleanup
└── Account info caching
User Action: Start the trading session
User: "I want to start trading"
Agent Response: Check authentication status and guide setup
Agent: "Let me check your authentication status..."
→ Calls: checkTradeAuth()
→ If not authenticated: Provides setup instructions
→ If authenticated: Proceeds to market analysis
Authentication Process:
- Visit https://app.hyperliquid.xyz/API
- Generate API wallet
- Authorize for trading
- Execute authentication command:
auth 0x[your-private-key] 0x[your-wallet-address]
User Action: Request trading recommendations
User: "Show me the best trading opportunities"
Agent Processing:
Step 1: Market Overview
→ Calls: analyzeMarkets(useTopOpportunities: true)
→ Fetches Fear & Greed Index
→ Analyzes market sentiment
Step 2: Opportunity Scanning
→ Calls: findTopOpportunities(maxSymbols: 30, topCount: 5)
→ Scans universe for high win rate assets
→ Calculates opportunity scores
Step 3: Pattern Analysis (for top opportunities)
→ Calls: recognizePatterns() for each top asset
→ Identifies technical patterns
→ Confirms pattern validity
Step 4: Recommendation Generation
→ Calls: generateRecommendation() for best opportunity
→ Combines all analysis factors
→ Provides entry/exit strategy
Agent Response Example:
Top 5 Trading Opportunities (Ranked by Win Rate):
1. BTC-PERP: 74.2% Win Rate | Score: 85/100
Setup: Rising Wedge (78% confidence) - BEARISH
Entry: $43,250 | Target: $39,900 | Stop: $44,080
Reasoning: Excellent win rate + bearish reversal pattern + extreme greed
2. ETH-PERP: 68.7% Win Rate | Score: 78/100
Setup: Double Top (72% confidence) - BEARISH
Entry: $2,845 | Target: $2,563 | Stop: $2,930
Reasoning: Strong win rate + rejection at resistance + volume confirmation
[... additional opportunities ...]
Market Sentiment: Extreme Greed (82) - Contrarian opportunities favored
Recommendation: Focus on high win rate bearish setups
User Action: Decide to trade the recommended opportunity
User: "I want to trade BTC short as recommended"
Agent Processing:
Step 1: Account Information
→ Calls: getAccountInfo()
→ Retrieves account balance and margin details
→ Assesses available capital
Step 2: Position Size Calculation
→ Calls: calculatePositionSize(
entryPrice: 43250,
stopLoss: 44080,
leverage: 3,
symbol: "BTC-PERP",
winRate: 74.2
)
→ Applies dynamic risk management
→ Adjusts for win rate confidence
Step 3: Risk Assessment
→ Validates position against account size
→ Checks margin requirements
→ Provides risk warnings if necessary
Agent Response Example:
Account Analysis:
Account Value: $2,847
Available Margin: $2,203
Dynamic Risk Used: 20% (Medium account tier)
Position Sizing for BTC-PERP Short:
Entry Price: $43,250
Stop Loss: $44,080 (1.92% distance)
Leverage: 3x
Win Rate Adjustment: 1.24x (for 74.2% win rate)
Calculated Position:
Size: 0.0423 BTC
Notional Value: $1,828
Margin Required: $609
Risk Amount: $569 (20% of account)
Max Loss: $569
Risk Assessment: MEDIUM
Warnings: None - position within safe parameters
User Action: Confirm trade execution
User: "Execute the BTC short trade"
Agent Processing:
Step 1: Final Validation
→ Calls: getAccountInfo() (confirm sufficient margin)
→ Validates current market price
→ Confirms position parameters
Step 2: Trade Execution
→ Calls: executeTrade(
symbol: "BTC-PERP",
side: "sell",
size: 0.0423,
orderType: "limit",
price: 43250,
leverage: 3,
stopLoss: 44080
)
→ Places order on Hyperliquid
→ Confirms execution
Step 3: Position Setup
→ Calls: updateStopLoss() if needed
→ Sets up monitoring parameters
Agent Response Example:
Trade Execution Successful!
Order Details:
Symbol: BTC-PERP
Side: SHORT
Size: 0.0423 BTC
Executed Price: $43,247
Leverage: 3x
Status: FILLED
Position Summary:
Entry: $43,247
Stop Loss: $44,080
Target 1: $41,100
Target 2: $39,900
Risk: $569 (20% of account)
Trade ID: HL_987654321
Execution Time: 2025-07-14 14:32:15 UTC
User Action: Check position status
User: "How is my BTC position doing?"
Agent Processing:
Step 1: Position Analysis
→ Calls: getPositions()
→ Retrieves current position status
→ Calculates real-time P&L
Step 2: Risk Monitoring
→ Calls: monitorPositions()
→ Assesses current risk levels
→ Checks for alerts
Step 3: Market Update
→ Fetches current price
→ Analyzes position performance
→ Provides management suggestions
Agent Response Example:
BTC-PERP Position Update:
Current Status:
Position: SHORT 0.0423 BTC
Entry Price: $43,247
Current Price: $42,890
Unrealized P&L: +$151 (+26.5%)
Time in Trade: 2h 15m
Performance Metrics:
Profit Target 1: 38% complete
Distance to Stop: $1,190 (2.8%)
Position Health: GOOD
Risk Assessment:
Portfolio Risk: LOW
Margin Usage: 21.4%
No alerts
Recommendation: Position moving favorably. Consider taking partial profits at $42,500 (Target 1) and letting remainder run to Target 2.
User Action: Take partial profits
User: "Take partial profits on 50% of BTC position"
Agent Processing:
Step 1: Position Calculation
→ Calls: getPositions() to get current size
→ Calculates partial close amount (50% = 0.02115 BTC)
→ Validates partial close parameters
Step 2: Partial Close Execution
→ Calls: closePosition(
symbol: "BTC-PERP",
size: 0.02115,
orderType: "market"
)
→ Executes partial close
→ Calculates realized P&L
Step 3: Remaining Position Update
→ Updates stop loss for remaining position
→ Recalculates risk metrics
Agent Response Example:
Partial Profit Taking Executed:
Closed Position:
Size Closed: 0.02115 BTC (50%)
Exit Price: $42,885
Realized P&L: +$75.50
Remaining Position:
Size: 0.02115 BTC
Entry: $43,247 (average)
Current P&L: +$76.50
Updated Risk: $284 (10% of account)
Position Status: REDUCED
Risk Level: LOW
Next Target: $39,900 (Target 2)
Recommendation: Stop loss moved to breakeven for remaining position. Let profits run to final target.
User Action: Monitor until exit
User: "Monitor position and close when appropriate"
Agent Processing:
Continuous Monitoring:
→ Regular calls to monitorPositions()
→ Price level monitoring
→ Pattern development tracking
→ Risk assessment updates
Exit Triggers:
→ Target level reached
→ Pattern invalidation
→ Risk parameter breach
→ Time-based exit
- Node.js 20+
- pnpm (recommended package manager)
- Docker (for deployment)
- Hyperliquid account with API access
- Clone and Install Dependencies
git clone https://github.com/chigozzdevv/hyperliquid-trading-agent
cd hyperliquid-trading-agent
pnpm install- Environment Configuration
cp .env.example .env
# Configure your environment variables- Start Development Server
pnpm run dev- Access Playground Open http://localhost:8080 to test your agent
# LLM Configuration
API_BASE_URL=http://127.0.0.1:11434/api or go to the Nosana deployment section to deploy to Nosana
MODEL_NAME_AT_ENDPOINT=qwen2.5:32b
HYPERLIQUID_TESTNET=false
# Production Deployment
NODE_ENV=production
PORT=8080- Build Container
docker build -t yourusername/hyperliquid-trading-agent:latest .- Test Locally
docker run -p 8080:8080 --env-file .env yourusername/hyperliquid-trading-agent:latest- Push to Registry
docker push yourusername/hyperliquid-trading-agent:latest# Install Nosana CLI
npm install -g @nosana/cli
# Fund wallet with NOS and SOL
nosana address
# Deploy model
nosana job post --file ./nos_job_def/qwen_job.json --market nvidia-3090 --timeout 200# Deploy agent
nosana job post --file ./nos_job_def/trading_agent.json --market nvidia-3060 --timeout 200| Component | GPU | VRAM | RAM | Storage |
|---|---|---|---|---|
| Qwen 32B | Required | 64GB | 32GB | 100GB |
| Trading Agent | Optional | 0GB | 4GB | 20GB |
Core Principle: Identify assets with proven winning track records and execute contrarian trades during sentiment extremes.
Selection Criteria:
- Minimum 60% historical win rate
- Sufficient liquidity (>$2M daily volume)
- Clear technical pattern confirmation
- Multi-timeframe alignment
Execution Framework:
- Extreme Greed (75-100) + High Win Rate Bearish Patterns: Premium short opportunities
- Extreme Fear (0-25) + High Win Rate Bullish Patterns: Premium long opportunities
- Pattern Strength Inversely Correlated with Sentiment: Maximum opportunity when crowd is wrong
Dynamic Risk Scaling: Position sizes automatically adjust based on account tier to ensure tradeable amounts while maintaining appropriate risk levels.
Account Tiers:
- Micro Accounts (<$10): 100% risk for account building
- Small Accounts ($10-50): 50% risk for aggressive growth
- Medium Accounts ($50-500): 20% risk for balanced approach
- Large Accounts (>$500): 2% risk for capital preservation
Win Rate Adjustments: Position sizes increase for assets with higher win rates, with multipliers up to 1.5x for >70% win rate assets.
- Win Rate: Primary metric for opportunity assessment
- Sharpe Ratio: Risk-adjusted return measurement
- Maximum Drawdown: Capital preservation metric
- Profit Factor: Gross profit divided by gross loss
- Average Risk-Reward: Target minimum 2:1 ratios
The system emphasizes forward-looking analysis based on:
- Historical win rate patterns
- Technical pattern success rates
- Sentiment divergence profitability
- Risk-adjusted performance metrics