PaviAgent

PaviAgent is an open-source Agentic AI Assistant built with Python, FastAPI, LangGraph, LangChain, Google Gemini, Tavily, ChromaDB, and SQLite.

It supports real-time streaming conversations, document uploads, Retrieval-Augmented Generation (RAG), web search, conversation memory, and a clean web interface.

---

✨ Features

• 🤖 Agentic AI Assistant powered by Google Gemini
• 💬 Real-time streaming responses
• 📄 Upload and chat with documents (PDF, DOCX, TXT, MD, PY, CSV)
• 📚 Retrieval-Augmented Generation (RAG) with ChromaDB
• 🌐 Web Search using Tavily
• 🧠 Long-Term Memory
• 💾 Conversation History Management
• ⚡ FastAPI-powered backend
• 🎨 Simple and responsive web interface
• 🔍 LangSmith integration for tracing and debugging

---

🏗️ Project Overview

PaviAgent combines modern AI technologies to build an intelligent conversational assistant.

Core Technologies

• FastAPI – Backend API and web server
• Jinja2 – Frontend template rendering
• LangGraph – Agent orchestration and workflow management
• LangChain – Tool calling, prompt management, and RAG pipeline
• Google Gemini – Large Language Model (LLM)
• Tavily Search – Real-time web search
• ChromaDB – Vector database for document retrieval
• SQLite – Conversation history and long-term memory storage

---

Prerequisites

Make sure you have the following installed:

• Python 3.11
• pip or Conda
• Git
• Google Gemini API Key
• Tavily API Key

---

🚀 Getting Started

1️⃣ Clone the Repository

git clone https://github.com/Pavithrareddy2702/PaviAgent.git

---

2️⃣ Navigate to the Project Directory

cd PaviAgent

---

3️⃣ Create a Virtual Environment

Using Conda:

conda create -n paviagent python=3.11 -y

---

4️⃣ Activate the Environment

conda activate paviagent

---

5️⃣ Install Dependencies

pip install -r requirements.txt

---

⚙️ Environment Variables

Create a .env file in the project root directory.

GOOGLE_API_KEY=your_google_api_key
GOOGLE_MODEL=gemini-2.5-flash

TAVILY_API_KEY=your_tavily_api_key

LANGSMITH_TRACING=true
LANGSMITH_ENDPOINT=https://api.smith.langchain.com
LANGSMITH_API_KEY=your_langsmith_api_key
LANGSMITH_PROJECT=PaviAgent

If you don't want LangSmith tracing, use:

LANGSMITH_TRACING=false

---

Run the Application

python app.py

The application will be available at:

http://127.0.0.1:8080

---

📁 Project Structure

PaviAgent/
│
├── app.py                  # FastAPI application and streaming endpoints
├── agent.py                # LangGraph agent workflow
├── database.py             # SQLite database and conversation persistence
├── rag.py                  # Document ingestion and Retrieval-Augmented Generation
├── tools.py                # Agent tools (Web Search, Calculator, Memory, RAG)
├── requirements.txt        # Project dependencies
├── .gitignore
├── README.md
│
├── templates/
│   └── index.html          # Chat interface
│
├── uploads/                # Uploaded documents
├── chroma_db/              # Chroma Vector Database
└── data/                   # SQLite database

---

Tech Stack

• Python
• FastAPI
• LangGraph
• LangChain
• Google Gemini
• Tavily Search API
• ChromaDB
• SQLite
• Jinja2
• LangSmith

---

Use Cases

• AI-powered conversational assistant
• Question answering over uploaded documents
• Retrieval-Augmented Generation (RAG)
• Real-time web search
• Long-term memory for conversations
• Multi-session chat management

---