Agent Learning Machine

A complete autonomous Solana trading agent system with 4 layers: Observation, Decision, Execution, and Learning. Built with React, Express, tRPC, and MySQL.

🏗️ Architecture

4-Layer Agent System

1. 👁 Observation Layer (Data Engine)

Real-time token data ingestion from Helius API
Market snapshot collection (price, volume, liquidity, RSI, MACD, sentiment)
Wallet activity tracking and whale detection
Buy/sell pressure analysis
Technical indicator calculation

2. 🧠 Decision Layer (Strategy Agent)

Weighted scoring system with configurable parameters
Liquidity score: Evaluates pool depth and market stability
Volume score: Analyzes trading activity relative to market cap
Wallet score: Tracks whale accumulation patterns
Composite score: liquidity_score * 0.3 + volume_score * 0.3 + wallet_score * 0.4
Decision logic: BUY (score > 0.75), SELL (score < 0.25), HOLD (0.25-0.75), SKIP

3. ⚡ Execution Layer (Execution Engine)

Automated trade execution based on agent decisions
Position tracking with entry/exit prices
Risk management: Stop loss and take profit levels
Position sizing based on max position limits
Transaction history and execution validation
Real-time PnL calculation

4. 📈 Learning Layer (Learning Engine)

Performance metrics calculation (win rate, Sharpe ratio, max drawdown)
Decision feedback tracking
Weight optimization based on historical success rates
Continuous improvement through feedback loops
Performance history and trend analysis

Database Schema

12 Tables:

users - User authentication
market_snapshots - Historical token data
wallet_activity - Whale wallet movements
agent_config - Agent configuration and weights
agent_decisions - All agent decisions with scores
agent_signals - Market signals and research data
trade_executions - Executed trades
positions - Open and closed positions
performance_metrics - Trading performance data
decision_feedback - Decision outcomes
weight_history - Weight adjustments over time
telegram_config - Telegram bot integration

🚀 Getting Started

Prerequisites

Node.js 22+
MySQL/TiDB database
Helius API key (optional, for real Solana data)
Telegram bot token (optional, for notifications)

Installation

# Install dependencies
pnpm install

# Set up environment variables
cp .env.example .env

# Run database migrations
pnpm drizzle-kit generate
pnpm drizzle-kit migrate

# Start development server
pnpm dev

Environment Variables

DATABASE_URL=mysql://user:password@localhost/agent_learning_machine
JWT_SECRET=your-secret-key
HELIUS_API_KEY=your-helius-api-key
TELEGRAM_BOT_TOKEN=your-telegram-bot-token

📊 API Endpoints

Agent Configuration

GET /api/trpc/agent.getConfigs - Get all agent configs
POST /api/trpc/agent.createConfig - Create new config
GET /api/trpc/agent.getConfig - Get specific config
POST /api/trpc/agent.updateConfig - Update weights/thresholds

Data Collection

POST /api/trpc/data.collectToken - Collect data for single token
POST /api/trpc/data.collectMultiple - Collect data for multiple tokens
GET /api/trpc/data.getTrend - Get market trend

Strategy Analysis

GET /api/trpc/strategy.analyzeToken - Analyze single token
GET /api/trpc/strategy.analyzeMultiple - Analyze multiple tokens
GET /api/trpc/strategy.getRecommendations - Get portfolio recommendations
GET /api/trpc/strategy.getDecisions - Get recent decisions
GET /api/trpc/strategy.getSignals - Get active signals

Execution

GET /api/trpc/execution.getPortfolio - Get portfolio summary
GET /api/trpc/execution.getPositions - Get open positions
GET /api/trpc/execution.getTradeHistory - Get trade history
POST /api/trpc/execution.updatePrices - Update position prices

Learning

GET /api/trpc/learning.getMetrics - Get performance metrics
GET /api/trpc/learning.getHistory - Get performance history
GET /api/trpc/learning.getLatest - Get latest metrics
POST /api/trpc/learning.runCycle - Run learning cycle

Dashboard

GET /api/trpc/dashboard.getSummary - Get complete dashboard data

🎯 Scoring System

Liquidity Score

Evaluates pool depth and stability
Ideal ratio: 10-30% of market cap
High liquidity = easier entries/exits

Volume Score

Analyzes trading activity
Ideal ratio: 10-50% of market cap
High volume = healthy market activity

Wallet Score

Detects whale accumulation
Tracks large wallet movements
Positive signal: Whales accumulating
Negative signal: Whales dumping

Technical Score

RSI (30 = oversold, 70 = overbought)
MACD (positive = bullish)
Sentiment (0-1 scale)
Buy/sell pressure ratio

🔧 Configuration

Adjust Scoring Weights

// Default weights
liquidityWeight: 0.30
volumeWeight: 0.30
walletWeight: 0.40

// Customize via API
POST /api/trpc/agent.updateConfig
{
  configId: 1,
  liquidityWeight: 0.25,
  volumeWeight: 0.35,
  walletWeight: 0.40
}

Set Decision Thresholds

buyThreshold: 0.75      // Score must exceed this to BUY
sellThreshold: 0.25     // Score must fall below this to SELL
holdThreshold: 0.50     // Between sell and buy = HOLD

Risk Management

maxPositionSize: 1000       // Max $ per position
stopLossPercent: 5          // Stop loss at 5% below entry
takeProfitPercent: 15       // Take profit at 15% above entry

📈 Performance Metrics

Win Rate: Percentage of profitable trades
Profit Factor: Gross profit / Gross loss
Sharpe Ratio: Risk-adjusted returns
Max Drawdown: Largest peak-to-trough decline
Decision Accuracy: Percentage of correct decisions
Average Confidence: Mean confidence of all decisions

🔔 Telegram Integration

Send market snapshots and trading decisions to Telegram:

// Send market snapshot
await sendMarketSnapshot(configId, tokenSymbol, snapshot);

// Send trading decision
await sendTradingDecision(configId, tokenSymbol, decision);

// Send position update
await sendPositionUpdate(configId, tokenSymbol, update);

// Send performance report
await sendPerformanceReport(configId, metrics);

🧠 Learning Cycle

The agent continuously learns from outcomes:

Record Feedback: Track actual prices vs predicted scores
Analyze Signals: Identify which signals were effective
Calculate Metrics: Compute performance statistics
Optimize Weights: Adjust scoring weights based on success rates
Store History: Maintain audit trail of all changes

📊 Dashboard Features

Real-time Metrics: Live PnL, win rate, Sharpe ratio
Position Tracking: Open positions with unrealized PnL
Decision History: Recent agent decisions with confidence scores
Active Signals: Current market signals and their strength
Performance Charts: Historical performance visualization
Configuration UI: Adjust weights and thresholds in real-time

🔐 Security

JWT-based authentication
Protected procedures for sensitive operations
Database encryption for sensitive data
Rate limiting on API endpoints
Input validation on all endpoints

📝 Development

Project Structure

/agent-learning-machine
├── client/                  # React frontend
│   ├── src/
│   │   ├── pages/          # Page components
│   │   ├── components/     # Reusable components
│   │   └── lib/            # Utilities
├── server/                  # Express backend
│   ├── engines/            # Agent engines
│   │   ├── dataEngine.ts
│   │   ├── strategyAgent.ts
│   │   ├── executionEngine.ts
│   │   ├── learningEngine.ts
│   │   └── telegramEngine.ts
│   ├── routers.ts          # tRPC routes
│   └── db.ts               # Database helpers
├── drizzle/                # Database schema
│   └── schema.ts
└── shared/                 # Shared types

Running Tests

pnpm test

Building for Production

pnpm build
pnpm start

🚀 Deployment

The project is ready for deployment on Manus hosting:

Create a checkpoint: webdev_save_checkpoint
Click the Publish button in the Management UI
Configure custom domain if desired

📚 Documentation

Data Engine: Helius API integration, technical indicators
Strategy Agent: Scoring system, decision logic
Execution Engine: Trade execution, position management
Learning Engine: Performance tracking, weight optimization
Telegram Integration: Notifications and alerts

🤝 Contributing

This is an autonomous agent system. Contributions should focus on:

Improving scoring algorithms
Adding new signal types
Enhancing risk management
Optimizing performance

📄 License

MIT License - See LICENSE file for details

🎓 Learning Resources

Solana Development: https://docs.solana.com
Helius API: https://docs.helius.xyz
tRPC: https://trpc.io
Drizzle ORM: https://orm.drizzle.team

🆘 Support

For issues and questions:

Check the dashboard for real-time status
Review performance metrics and decision history
Adjust configuration based on market conditions
Run learning cycles to optimize weights

Built with ❤️ for autonomous trading on Solana

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
.github/workflows		.github/workflows
.manus/db		.manus/db
client		client
drizzle		drizzle
patches		patches
server		server
shared		shared
.gitignore		.gitignore
.gitkeep		.gitkeep
.prettierignore		.prettierignore
.prettierrc		.prettierrc
README.md		README.md
components.json		components.json
drizzle.config.ts		drizzle.config.ts
package.json		package.json
pnpm-lock.yaml		pnpm-lock.yaml
todo.md		todo.md
tsconfig.json		tsconfig.json
vite.config.ts		vite.config.ts
vitest.config.ts		vitest.config.ts

Folders and files

Latest commit

History

Repository files navigation

Agent Learning Machine

🏗️ Architecture

4-Layer Agent System

Database Schema

🚀 Getting Started

Prerequisites

Installation

Environment Variables

📊 API Endpoints

Agent Configuration

Data Collection

Strategy Analysis

Execution

Learning

Dashboard

🎯 Scoring System

Liquidity Score

Volume Score

Wallet Score

Technical Score

🔧 Configuration

Adjust Scoring Weights

Set Decision Thresholds

Risk Management

📈 Performance Metrics

🔔 Telegram Integration

🧠 Learning Cycle

📊 Dashboard Features

🔐 Security

📝 Development

Project Structure

Running Tests

Building for Production

🚀 Deployment

📚 Documentation

🤝 Contributing

📄 License

🎓 Learning Resources

🆘 Support

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages