🚀 QaMail-LLM

---

QaMail-LLM is a local python program retrieves emails via POP3, understands their context with RAG techniques, and generates smart, accurate replies—making your email conversations faster and more efficient.

🌟 Features

• Multi-LLM Support: Compatible with Ollama, OpenAI, and other OpenAI API-compatible LLMs.

• Advanced Email Processing: Integrates time and keyword filtering, self-RAG, and hybrid search for accurate responses.

• Rich Email Display: Parses and renders HTML email content for a seamless viewing experience.

• Bilingual Embedding Model: Utilizes an embedding model that supports both Chinese and English for better multilingual processing.

---

📚 Table of Contents

• 🧰 Prerequisites
• 🔧 Installation
• ⚡ Quick Start
• 🔮 Future Work
• 📜 Declaration

---

🧰 Prerequisites

Before you begin, ensure you have met the following requirements:

• Software/Version: python 3.8+ (we use python 3.9.21)

• Dependencies: pip, Anaconda

To run this project, you need to install the following dependencies. We recommend using either pip (Python's default package manager) or Anaconda (a popular Python distribution).

• Hardware Requirements:

if you use local LLM, usually at least 3GB of free VRAM are necessary to generates a fluent and accurate response. if you use external LLM, it doesn't matter.

• Accounts: ollama or other LLM with api_keys, your email_account with POP3 server support
  • Downloading Ollama
  • Gmail POP3 instruction
    • Common Question for Gmail POP3 Setting:https://support.google.com/accounts/thread/174976836/pop3-gmail-settings-not-working

---

🔧 Installation

To install and set up the project, follow these steps:

1. Clone the repository:

   git clone https://github.com/georgong/QaMail-LLM.git

2. Install the required package:

   cd QaMail-LLM
   pip install -r requirements.txt

⚡ Quick Start

📥 Pull the Model from Ollama (Optional)

If you are using Ollama as your LLM provider, you need to pull the model before running it.

ollama pull qwen2.5:3b

• You can replace qwen2.5:3b with any other model of your choice.
• Larger models 🏋️‍♂️ perform better on complex tasks.
• Smaller models 🏃‍♂️ require less RAM and run more efficiently.

🛠️ Create user.json in the folder

Create a user.json file and fill in your email configuration:

This is the default configuration used by the demo. Update the values as needed.

{
  "email": "put your email here",
  "password": "put your password here",
  "server": "put your server here",
  "user_id": "create your user_id for login",
  "initial_load_num": 100,
  "llm_url": "http://localhost:11434",
  "llm_provider": "ollama",
  "model": "qwen2.5:3b",
  "enable_summary": false,
  "n_results": 10,
  "port": 5001,
  "api_key":"put your api key here"
}

🔹 Notes:

• Replace "put your email", "put your password", and "put your server" with actual values.
• "enable_summary" is not used yet, it will coming soon with hybrid search together.
• Ensure llm_url points to your running LLM service.

---

🚀 Run the Server

After setting up the configuration, start the server with:

python server.py

The server should now be running and processing emails with the configured LLM. 🎯

🚨 Note: The first time you run the program, it may take a long time to initialize, as it needs to parse and embed a large number of emails.
⏱ The exact time depends on initial_load_num — a larger value means longer processing time.

• If you want faster startup, consider reducing initial_load_num.
• Subsequent runs will be faster, as previously processed emails are already embedded.

---

🌍 View Results

Once the server is running, open your browser and visit:

http://127.0.0.1:PORT

Replace PORT with the actual port number used by your server (e.g., 8000, 5000, etc.). You can also find the address in the output of console.

🎯 Now, your AI email assistant is ready to process emails and generate responses! 🚀

🔮 Future Work

• Daily Email Summary: Generate a concise summary of the day's emails for quick review.
• Scheduled Email Fetching: Implement a scheduler to automatically retrieve and process emails at a specified time each day.
• Support for IMAP: Extend email retrieval to IMAP for broader compatibility.
• Improved Summarization: Enhance email summarization with more context-aware techniques.
• More LLM Integrations: Add support for additional LLM providers and local models.
• User Interface Improvements: Develop a web-based dashboard for easier email management and configuration.

📜 Declaration

• Frontend Attribution: The frontend part of this project is modified from simple_chat.
• Network Usage & Security Notice:
  • This project operates entirely offline, except for POP3 server communication and external LLM API calls.
  • Users should be aware that using external LLMs may result in data exposure due to API requests being sent over the network.
  • To maximize security and privacy, it is recommended to use local LLMs via Ollama or trusted official providers such as OpenAI or Anthropic (Claude 3).

---

🚀 Stay secure and enjoy the AI-powered email assistant!