🤖 Multi-Agent Workflow Demonstrations in LangGraph

A collection of demonstrations showcasing different patterns for implementing multi-agent workflows using LangGraph. Each example highlights specific orchestration approaches to help developers understand and build collaborative AI systems.

📋 Features

Hierarchical Agent Teams

• 🏗️ Orchestrates complex workflows with multiple levels of supervision
• 🔄 Seamlessly coordinates specialized agent teams (research, document writing)
• 🧠 Efficiently delegates tasks to lower-level agents with appropriate tools
• 📊 Follows proper escalation and reporting paths in the agent hierarchy

Agent Supervision

• 👨‍💼 Models a supervisor-worker relationship for intelligent task delegation
• 🔄 Routes tasks between research and coding agents based on requirements
• 🚦 Manages conversation flow with clear transitions between agents
• 🔍 Makes real-time decisions about which agent should act next

Multi-Agent Collaboration

• 🤝 Enables direct peer-to-peer collaboration between agents
• 🛠️ Shares tools and information across collaborating agents
• 📊 Demonstrates fluid conversation flow between specialized agents
• 📡 Facilitates tool calling between different agent types

🏗️ Architecture

The demonstrations showcase three distinct multi-agent architecture patterns:

Hierarchical Teams

Agent Supervisor

Multi-Agent Collaboration

📦 Implementation Details

hierarchical_agent_teams.py

This demonstration implements a sophisticated hierarchical team structure with multiple levels of supervision:

• Top-level Supervisor: Coordinates between specialized teams
• Research Team: Combines web search and web scraping capabilities
• Document Writing Team: Creates, edits and manages document creation
• Tool Integration: Implements document creation/editing tools, search tools, and web scraping
• State Management: Shows how to manage complex state across the hierarchy

agent_supervisor.py

This demonstration implements a supervisor-worker architecture:

• Supervisor Agent: Makes routing decisions about which agent to activate
• Research Agent: Uses search tools to gather information
• Coding Agent: Executes Python code for calculations and analysis
• Decision Logic: Shows how to implement routing logic for multi-agent systems

multi_agent_collaboration.py

This demonstration implements a peer-to-peer collaborative agent system:

• Researcher Agent: Gathers data from web sources
• Chart Generator Agent: Creates visualizations from research data
• Tool Sharing: Shows how tools can be used across agent boundaries
• Collaborative Workflow: Demonstrates agents working together on a shared task

🛠️ Requirements

• Python 3.9+
• OpenAI API key
• Tavily API key
• LangChain and LangGraph libraries

📦 Installation

# Install dependencies
pip install -r requirements.txt

# Set up environment variables
export OPENAI_API_KEY="your-openai-api-key"
export TAVILY_API_KEY="your-tavily-api-key"

# Optional: Enable LangChain tracing
export LANGCHAIN_TRACING_V2="true"
export LANGCHAIN_PROJECT="Multi-Agent-Workflow"

🚀 Usage

Each demonstration can be run as a standalone script:

# Run the hierarchical agent teams demo
python scripts/hierarchical_agent_teams.py

# Run the agent supervisor demo
python scripts/agent_supervisor.py

# Run the multi-agent collaboration demo
python scripts/multi_agent_collaboration.py

🏆 Key Benefits

• Modularity: Each example demonstrates a reusable pattern for multi-agent systems
• Flexibility: Shows different approaches to agent coordination and collaboration
• Adaptability: Provides templates that can be customized for specific use cases
• Scalability: Demonstrates patterns that can scale to complex multi-agent systems