# RAG Agent with ChromaDB and Web Search

This repository implements a simple RAG (Retrieval-Augmented Generation) agent that can:

1. Search a local knowledge base stored in **ChromaDB** using semantic embeddings from **Ollama**.
2. Perform real‑time web search via **Tavily**.
3. Decide automatically which source to use and indicate the source in the final answer.

## Prerequisites

- Python 3.10+ (recommended via `pyenv` or `conda`).
- Ollama installed locally and the following models pulled:
  ```bash
  ollama pull llama3
  ollama pull nomic-embed-text
  ```
- A Tavily API key. Create a `.env` file in the project root with:
  ```text
  TAVILY_API_KEY=YOUR_KEY_HERE
  ```

## Installation

```bash
# Optional: create a virtual environment
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt
```

## Preparing the Knowledge Base

Place any `.txt` or `.md` files you want the agent to know about in the `documents/` folder.
Run the following command once to load them into ChromaDB:

```bash
python -c "from vectorstore import create_vectorstore, load_documents; store=create_vectorstore(); load_documents('./documents', store)"
```

The vector store is persisted in the `chroma_db/` directory, so the data will be available for subsequent runs.

## Running the Agent

```bash
python main.py
```

You will see a simple chat loop. Type your questions and the agent will answer.

```
Welcome to the RAG agent. Type 'exit' to quit.

User: What is LangGraph?

Assistant: LangGraph is a framework for building ...
Source: chromadb
```

If the information is not present locally, the agent will automatically perform a web search and label the answer with `Source: tavily`.

## Project Structure

```
├── agent.py          # Core agent logic
├── rag_tools.py      # Tool implementations
├── vectorstore.py    # ChromaDB utilities
├── main.py           # Entry point
├── requirements.txt
├── .gitignore
└── README.md
```

## Extending

- Add more tools by creating new functions decorated with `@tool`.
- Replace the LLM or embeddings with other Ollama models.
- Switch to a different vector store (e.g., Qdrant) by updating `vectorstore.py`.

## License

MIT License.