---
title: Agentic RAG System
emoji: ⚡
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.43.1
app_file: app.py
pinned: false
license: mit
short_description: Agentic RAG application with enhanced web search.
---

# ⚡ Agentic RAG System

An Agentic Retrieval-Augmented Generation (RAG) system that combines **local vector search** with **live web retrieval** to provide grounded, cited answers.

## Features

- 🔍 Semantic retrieval with **FAISS**
- 🌐 Live web search fallback via **Tavily**
- 💬 Multi-turn conversations with contextual memory
- 📚 Automatic source citations
- ⚡ Powered by **Qwen2.5-7B-Instruct**

## Tech Stack

- **Framework:** LangGraph
- **LLM:** Qwen2.5-7B-Instruct
- **Vector Store:** FAISS
- **Web Search:** Tavily

## Usage

1. Ask a question.
2. The system first searches its local knowledge base.
3. If relevant information isn't available, it retrieves information from the web.
4. Responses include references to the retrieved sources.

## Configuration

This Space requires the following secret:

| Secret | Description |
|---------|-------------|
| `TAVILY_API_KEY` | Tavily Search API key |

The Space is configured to run on **ZeroGPU** hardware.

## Source Code

The complete implementation and project documentation are available on [GitHub](https://github.com/SaintJeane/agentic-rag).