Agent - LangGraph Agentic AI Framework

A sophisticated agentic AI system built with LangGraph that demonstrates advanced multi-agent coordination, web search integration, and real-time information retrieval.

Overview

This project implements an intelligent agent framework that can:

• Process complex queries using agentic workflows
• Perform web searches using Tavily Search API for real-time information
• Handle multi-step reasoning and decision-making
• Manage state across distributed agents
• Serve web-based frontend interface

Features

• LangGraph State Management: Robust state handling with TypedDict for type safety
• Web Search Integration: Tavily Search API integration for fetching current information
• HuggingFace LLM: Support for HuggingFace models and endpoints
• Multi-Agent Architecture: Distributed task execution with agent coordination
• Freshness Detection: Built-in logic to detect freshness-related queries
• Full-Stack Application: Python backend + Next.js frontend

Project Structure

agent/
├── basic.py              # Core agent logic and workflows
├── server.py             # Backend server implementation
├── frontend/             # Next.js frontend application
│   ├── src/
│   ├── public/
│   └── package.json
├── blog_posts/           # Research and documentation
├── requirements.txt      # Python dependencies
└── .env                  # Environment variables

Getting Started

Prerequisites

• Python 3.10+
• Node.js 18+
• Tavily API Key
• HuggingFace API Token

Installation

# Install Python dependencies
pip install -r requirements.txt

# Install frontend dependencies
cd frontend
npm install
cd ..

Configuration

Create a .env file in the parent directory with:

TAVILY_API_KEY=your_tavily_key
HUGGINGFACEHUB_API_TOKEN=your_hf_token

Running the Application

# Start backend server
python server.py

# In another terminal, start frontend (from frontend directory)
npm run dev

Key Components

basic.py

• State Management: Defines agent state with message history and routing logic
• Agent Chain: Creates multi-step workflows using LangGraph StateGraph
• Search Integration: Handles web searches for freshness-related queries
• Output Parsing: Uses Pydantic for structured output validation

frontend/

Next.js-based web interface for interacting with the agent system, featuring:

• Real-time chat interface
• Response streaming
• Type-safe configuration with TypeScript

Technologies

• LangGraph: Agent orchestration and state management
• LangChain: LLM integration and prompt management
• HuggingFace: LLM models and embeddings
• Tavily: Web search API
• Next.js: Frontend framework
• FastAPI/Starlette: Backend server

Environment Variables

• TAVILY_API_KEY: API key for Tavily web search
• HUGGINGFACEHUB_API_TOKEN: HuggingFace authentication token
• HF_LLM_MODEL: HuggingFace model to use (default: meta-llama/Llama-2-7b-chat)

Usage Examples

The agent can handle various query types:

from basic import create_agent_graph

# Create agent graph
graph = create_agent_graph()

# Run agent with a query
response = graph.invoke({
    "messages": [HumanMessage("What are the latest trends in AI?")],
    "routing": "research"
})

License

MIT

Contributing

Contributions welcome! Please submit pull requests with clear descriptions of changes.