Em(AI)l Agent

Em(AI)l Agent is a Python-based AI agent that leverages Large Language Models (LLMs) to automatically draft and send emails. It is controlled via a simple Telegram bot interface, allowing users to send complex email requests using natural language.

How It Works

The agent operates through a two-step process:

1. Content Generation: When a user provides a topic or instruction via the Telegram bot, the agent first uses a local LLM (via Ollama and langchain) to generate an appropriate subject and body for the email. This process is guided by a prompt template that ensures the email is well-structured and signed correctly.

2. Action Execution: The generated subject and body are then passed to a second LangChain agent equipped with GmailToolkit. This agent interprets the user's initial command (e.g., "send" or "draft") and the recipient's details to perform the final action: either sending the email directly or saving it as a draft in your Gmail account.

Features

• Natural Language Commands: Draft or send emails by describing your request in plain English.
• Telegram Bot Integration: Easy-to-use interface through a Telegram bot.
• Local LLM Powered: Uses Ollama with models like Llama 3.1 for privacy and cost-effectiveness.
• Gmail Integration: Securely interacts with your Gmail account to send emails and create drafts using the GmailToolkit.
• Structured Output: Employs Pydantic models to ensure reliable generation of email subjects and bodies from the LLM.

Setup and Installation

Prerequisites

• Python 3.8+
• Ollama installed and running.
• The Llama 3.1 model pulled in Ollama: ollama pull llama3.1
• Telegram API credentials (API ID, API Hash) and a Bot Token.
• Google Cloud Project with the Gmail API enabled. You will need to create OAuth 2.0 credentials and download the credentials.json file.

Installation

1. Clone the repository:

   git clone https://github.com/Soham-KT/Em-AI-l-Agent.git
   cd Em-AI-l-Agent

2. Set up Gmail API Credentials:

   • Follow the instructions from the Google API Python Quickstart to enable the Gmail API and download your credentials.json file.
   • Place the credentials.json file in the root directory of this project.
   • The first time you run the application, you will be prompted to authorize access to your Gmail account through a browser window. A token.json file will be created to store your credentials for future runs.

3. Install Python dependencies:

   pip install -r requirements.txt

4. Configure environment variables: Create a .env file in the root directory and add your Telegram credentials:

   TELEGRAM_API=<YOUR_TELEGRAM_API_ID>
   TELEGRAM_HASH=<YOUR_TELEGRAM_API_HASH>
   TELEGRAM_BOT_TOKEN=<YOUR_TELEGRAM_BOT_TOKEN>

Usage

1. Start Ollama: Make sure the Ollama service is running on your machine.

2. Run the application:

   python main.py

   The Telegram bot will start and connect to your account.

3. Interact with the bot: Open Telegram and start a conversation with your bot. You can use the following commands:

   • /start: Initializes the conversation with the bot.
   • /help: Provides instructions and the prompt format.
   • /info: Gives a brief description of the bot.

4. Send or Draft an Email: To instruct the agent, send a message following the format provided by the /help command. The agent will parse your request, generate the content, and perform the requested action.

   Example Prompts:

   • Send an email to John Doe, their email: johndoe@fake.com. The topic is a discussion about explaining black holes.
   • Draft an email to contact@company.com requesting a project update.

The bot will confirm once the task is completed.

File Breakdown

• main.py: The entry point of the application. It runs the Telethon client for the Telegram bot, handles commands, and passes user requests to the email agent.
• llm_mail.py: Contains the core logic for the email agent. It defines functions to generate email content using an LLM and to create and invoke the Gmail agent for sending/drafting.
• requirements.txt: Lists all the necessary Python packages for the project.
• test.py: A simple script for testing the Gmail agent functionality independently of the Telegram bot.
• LICENSE: The Apache 2.0 license for the project.

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.