LangGraph Multi-Agent System

A modular multi-agent system built with LangGraph, featuring tool routing, conversational memory, and flexible agent orchestration.

🎯 Features

• Multi-tool routing: Automatically routes queries to web search, calculator, or direct LLM
• Conversational memory: Maintains context across multiple turns
• Modular architecture: Easy to extend with new tools and agents
• Type-safe state management: Pydantic-based state definitions
• Production-ready: Error handling, logging, tests

🏗️ Architecture

Multi-Tool Agent

User Query → Router → [Search | Calculator | Direct] → Synthesizer → Answer

Conversational Agent

Query → Memory Check → [Use History | Fetch New Info] → Answer + Update Memory

🚀 Quick Start

# Setup with uv
uv venv
source .venv/bin/activate
uv pip install -e .

# Configure environment
cp .env.example .env
# Add your TAVILY_API_KEY

# Install Ollama
ollama pull mistral

# Run examples
python examples/basic_usage.py
python examples/with_memory.py

📖 Examples

Basic Multi-Tool Agent

from src.agents.multi_tool import create_multi_tool_agent

agent = create_multi_tool_agent()
result = agent.invoke({"question": "What are the latest AI trends?"})
print(result["final_answer"])

With Conversation Memory

from src.agents.conversational import create_conversational_agent

agent = create_conversational_agent()
messages = []

# First question
result = agent.invoke({
    "messages": messages,
    "current_question": "Who created LangGraph?"
})

# Follow-up (uses memory)
messages.append({"role": "user", "content": "Who created LangGraph?"})
messages.append({"role": "assistant", "content": result["answer"]})

result = agent.invoke({
    "messages": messages,
    "current_question": "What else did they build?"
})

🛠️ Tech Stack

• Framework: LangGraph 0.0.20
• LLM: Mistral 7B (via Ollama)
• Search: Tavily API
• Package Manager: uv
• Python: 3.9+

📁 Project Structure

langgraph-multiagent/
├── src/
│   ├── agents/
│   │   ├── multi_tool.py       # Multi-tool routing agent
│   │   └── conversational.py   # Agent with memory
│   ├── tools/
│   │   ├── search.py           # Web search (Tavily)
│   │   └── calculator.py       # Math calculations
│   ├── utils/
│   │   ├── state.py            # State type definitions
│   │   └── prompts.py          # Prompt templates
│   └── main.py
├── tests/
├── examples/
│   ├── basic_usage.py
│   └── with_memory.py
├── docs/
└── pyproject.toml

🧪 Testing

uv pip install -e ".[dev]"
pytest tests/

📚 Documentation

• Architecture Details
• Usage Examples

🤝 Contributing

This is a personal learning project, but suggestions are welcome!

📝 License

MIT

---

Built with LangGraph | Salma Ouardi | 2024