Local RAG Agent with Ollama & Streamlit

A professional, local-first Retrieval-Augmented Generation (RAG) agent. This application allows users to log in with a unique ID, index web documents via URL, and chat with an AI assistant that has "memory" of both the document and the conversation history.

---

🛠️ Prerequisites

Before running the application, ensure you have the following installed:

1. Python 3.11 or 3.12 (Recommended for stability).
2. Ollama: Download Ollama here.
3. Local Models: Once Ollama is installed, run these commands in your terminal to download the necessary models:

   ollama pull llama3.1
   ollama pull nomic-embed-text

---

🚀 Getting Started

Follow these steps to set up the project on your local machine:

1. Project Setup

Create a project folder and navigate into it:

mkdir rag-agent-app
cd rag-agent-app

2. Virtual Environment

Create and activate a virtual environment to keep your dependencies isolated:

# Create the environment
python -m venv venv

# Activate it (macOS/Linux)
source venv/bin/activate

# Activate it (Windows)
# venv\Scripts\activate

3. Install Dependencies

pip install streamlit beautifulsoup4 langchain langchain-ollama langchain-community langchain-core langchain-text-splitters

4. Run the App

streamlit run app.py

⚙️ Configuration

The project is designed with Separation of Concerns:

• LLM Selection: To change the reasoning model, edit get_llm() in engine.py (e.g., change llama3.1 to gemma2).

• Embedding Selection: By default, it uses HuggingFace models, update get_embeddings() in engine.py to change it if required.

• Chunking: You can adjust the chunk_size and chunk_overlap in engine.py to optimize how the agent "reads" longer or shorter articles.

📝 Usage

1. Login: Enter any Unique ID. This ID acts as your "account" key.

2. Paste URL: In the sidebar, paste a link to a blog post, news article, or documentation.

3. Chat: Ask questions like "What are the main points of this article?" or "Explain the specific methodology used."

4. Persistent History: Even if you close the app, logging back in with the same ID will retrieve your previous chat history from the SQLite database.

5. Multi-Document: If you paste a new URL, the agent will re-index and begin answering based on the new context.

📂 Project Structure

rag-agent-app/
├── app.py              # Main Entry point (Streamlit UI)
├── engine.py           # Core RAG & Agent Logic (LangChain/Ollama)
├── database.py         # SQLite Database handlers
├── requirements.txt    # Project dependencies
└── readme.md           # Project documentation