# 🚀 Local ChatGPT — Powered by Ollama + Python (Streaming + PDF RAG + Embeddings + FAISS)

![Python](https://img.shields.io/badge/Python-3.10+-blue)
![Flask](https://img.shields.io/badge/Flask-Web%20Framework-green)
![Ollama](https://img.shields.io/badge/Ollama-Local%20LLM-orange)
![FAISS](https://img.shields.io/badge/FAISS-Vector%20Search-purple)
![License](https://img.shields.io/badge/License-MIT-yellow)
![Stars](https://img.shields.io/github/stars/manojrammurthy/ollama-local-chatgpt?style=social)

A **fully local ChatGPT alternative** with:

✔ ChatGPT-style UI  
✔ Local LLMs via Ollama (Phi-3, LLaMA-3, Mistral, etc.)  
✔ PDF Upload + RAG  
✔ Real-time streaming responses  
✔ Embeddings using `nomic-embed-text`  
✔ FAISS vector search  
✔ No cloud, no API keys — **100% offline**

Built using **Flask + TailwindCSS + Ollama + FAISS + PyMuPDF**.

---

---

# ✨ Features

| Feature           | Description                                              |
| ----------------- | -------------------------------------------------------- |
| 🧠 Local LLM Chat | Uses any Ollama model (Phi-3, LLaMA-3, Mistral, Gemma… ) |
| 📄 PDF Upload     | Load documents and ask questions from them               |
| 🔍 RAG            | FAISS-powered retrieval from PDF chunks                  |
| 🧬 Embeddings     | Uses `nomic-embed-text` for dense vector embeddings      |
| 🚀 Live Streaming | Real-time token-by-token streaming like ChatGPT          |
| 🎨 Modern UI      | TailwindCSS dark-mode chat interface                     |
| 🔁 Model Selector | Switch Ollama models dynamically                         |
| 💾 Chat Memory    | Auto-saved in browser localStorage                       |
| 🔐 100% Offline   | No external API calls — everything runs on your laptop   |

---

# 🧠 Architecture Overview

Frontend (HTML + TailwindCSS) ↓ Flask API (Python) ↓ Ollama Chat Models (phi3, llama3, mistral…) ↓ Ollama Embedding Model (nomic-embed-text) ↓ FAISS Vector Search ↓ PDF Question Answering (RAG)

---

# 📁 Project Structure

ollama_web_chat/ │ ├── app.py ├── requirements.txt ├── README.md ├── .gitignore │ ├── templates/ │ └── index.html │ └── static/ (optional for CSS/JS assets)

---

# ⚙️ Installation

## 1️⃣ Install Ollama

Linux/macOS:

```bash
curl -fsSL https://ollama.com/install.sh | sh

Windows: Download from https://ollama.com/download

---

2️⃣ Pull required Ollama models

ollama pull phi3
ollama pull llama3
ollama pull mistral
ollama pull nomic-embed-text

---

3️⃣ Install Python dependencies

pip install -r requirements.txt

---

4️⃣ Run the Flask app

python app.py

Then open:

http://localhost:5000

---

🧬 Embeddings Explained

This project uses:

nomic-embed-text

Each chunk of your PDF is converted into a vector (embedding). These vectors are stored in a FAISS index.

During a question:

1. User question → embedded
2. Vector compared against FAISS index
3. Top-k relevant chunks returned
4. Passed to LLM as RAG prompt
5. LLM answers using context

This gives extremely accurate answers for:

• Annual reports
• Research papers
• Legal documents
• Technical PDFs
• Policies

---

📄 PDF RAG Flow

PDF → Text Extraction → Chunking → Embedding → FAISS Index
                                 ↑
                           User Question
                                 ↓
                            Vector Search
                                 ↓
                          Relevant Chunks
                                 ↓
                         LLM generates answer

---

🔁 Switching Models

The UI pulls all locally installed models:

/models → from Ollama tags API

You can switch between:

• phi3
• mistral
• llama3
• codellama
• gemma
• your custom models

Every change clears chat history automatically.

---

🚀 Roadmap

✔ v1.0 (Current)

• Local chat
• PDF embedding + FAISS
• Model selection
• Streaming
• Modern UI

🔜 v1.1

• Sidebar multi-threaded chat
• Voice input (Whisper)
• Export conversation
• Multi-PDF knowledgebase
• Dark/light toggle

🔮 v2.0

• Custom embedding model fine-tuning
• Workspace mode (knowledge graphs)
• Browser extension version
• Desktop app (Electron or PyInstaller)

---

🤝 Contributing

Pull requests welcome! If you want to add a feature, open an issue first.

---

⭐ Support the Project

If you find this useful, please give this repo a ⭐ on GitHub. It helps more people discover offline LLM tools.

---

📄 License

This project is licensed under the MIT License.