Skip to main content

Framework Architecture

TradingAgents employs a sophisticated multi-agent architecture that mirrors the organizational structure of professional trading firms. The framework is built using LangGraph to ensure flexibility and modularity.

System Overview​

The TradingAgents framework consists of four main components that work together in a coordinated manner:

┌─────────────────────────────────────────────────────────────┐
│                    TradingAgents Framework                   │
├─────────────────┬─────────────────┬─────────────────────────┤
│   Analyst Team  │ Researcher Team │    Trading & Risk       │
│                 │                 │      Management         │
├─────────────────┼─────────────────┼─────────────────────────┤
│ • Fundamentals  │ • Bullish       │ • Trader Agent          │
│ • Sentiment     │   Researcher    │ • Risk Management       │
│ • News          │ • Bearish       │ • Portfolio Manager     │
│ • Technical     │   Researcher    │                         │
└─────────────────┴─────────────────┴─────────────────────────┘

Component Details​

1. Analyst Team​

The analyst team consists of four specialized agents:

  • Fundamentals Analyst: Evaluates company financials and performance metrics
  • Sentiment Analyst: Analyzes social media and public sentiment using sentiment scoring
  • News Analyst: Monitors global news and macroeconomic indicators
  • Technical Analyst: Utilizes technical indicators (MACD, RSI) to detect patterns

2. Researcher Team​

The researcher team provides critical analysis through structured debate:

  • Bullish Researcher: Advocates for positive market positions
  • Bearish Researcher: Provides contrarian analysis and risk assessment
  • Debate Mechanism: Facilitates structured discussions between researchers

3. Trader Agent​

The trader agent synthesizes information from all sources:

  • Composes comprehensive reports from analysts and researchers
  • Makes informed trading decisions based on collective insights
  • Determines optimal timing and magnitude of trades

4. Risk Management & Portfolio Manager​

The final decision-making layer:

  • Risk Management: Continuously evaluates portfolio risk and market volatility
  • Portfolio Manager: Makes final approval/rejection decisions on trade proposals

Technical Implementation​

LangGraph Architecture​

TradingAgents is built on LangGraph, providing:

  • Modularity: Each agent is an independent, replaceable module
  • Flexibility: Easy to modify agent behaviors and interactions
  • Scalability: Support for adding new agents or modifying existing ones

LLM Integration​

The framework supports multiple LLM configurations:

  • Deep Thinking Models: o1-preview for complex analysis
  • Fast Thinking Models: gpt-4o for quick responses
  • Cost-Effective Options: gpt-4o-mini and o1-mini for testing

Data Flow Architecture​

Market Data → Analysts → Researchers → Trader → Risk Management → Decision
     ↑           ↓          ↓          ↓            ↓              ↓
API Sources   Analysis   Debate    Strategy    Risk Assessment   Execution

Configuration System​

The framework uses a comprehensive configuration system that allows customization of:

  • LLM Models: Choose different models for different agent types
  • Debate Rounds: Control the depth of researcher discussions
  • Data Sources: Toggle between online tools and cached data
  • Risk Parameters: Adjust risk tolerance and portfolio constraints

Scalability Considerations​

The modular design enables:

  • Horizontal Scaling: Add more specialized analysts
  • Vertical Scaling: Enhance individual agent capabilities
  • Custom Implementations: Replace agents with domain-specific versions

Next Steps​