✉️ LangGraph Email Assistant

A smart, production-ready email assistant built using LangGraph and integrated with Gmail and Google Calendar via Auth0. This assistant is designed to simplify email workflows by:

• Classifying incoming emails as: (respond, ignore, notify)
• Drafting responses to emails intelligently
• Scheduling meetings with context from the conversation
• Remembering cross-thread context using semantic long-term memory
• Fully secure with Google OAuth authentication through Auth0

---

Features

• ✅ Real-time email classification
• ✅ Smart response drafting
• ✅ Meeting scheduling via Google Calendar
• ✅ Semantic memory for contextual awareness across threads
• ✅ Can handle user commands like sending emails, retrieving the last sent email, and more
• 🔜 Future: Episodic and procedural memory integration for even more personalized, adaptive behavior

---

Example Execution

Below are example images of the email assistant in action:

---

How It Works

The LangGraph-powered assistant uses a graph-based architecture to manage its workflow. Key components include:

Detect Email Received Node

• This node is responsible for detecting if the input is related to a received email or a general user request.
• If it detects a received email, it forwards it to the Triage Router Node.
• If it detects a simple request (like asking for the last sent email or sending a new email), it forwards it directly to the React Agent.

Triage Router Node

• This node is responsible for deciding what to do with each incoming email.
• Based on the email's content and metadata, it routes the message to one of three paths:
  • Respond
  • Ignore
  • Notify

React Agent Node

• If the Triage Router decides the assistant should respond, it forwards the email to a React Agent.
• This agent composes a reply by analyzing context, thread history, and memory.
• It is also capable of handling various user requests, such as:
  • Sending an email to a specified recipient
  • Telling the user the last email they sent
  • Summarizing recent conversations
  • And much more

📸 Check the image below to visualize the flow inside LangGraph Studio.

---

🛠️ Getting Started

Prerequisites

1. Install uv package manager:

   pip install uv

---

Local Setup

Follow these steps to get the project running locally:

1. Clone the repository:

   git clone https://github.com/your-username/langgraph-email-assistant.git
   cd langgraph-email-assistant

2. Create your environment file:

   cp .env.example .env

   Update it with your Auth0, Gmail, and Calendar credentials.

3. Set up the virtual environment:

   a. Create and activate the environment:

   uv venv
   source .venv/bin/activate

   b. Install dependencies:

   uv run

4. Run the LangGraph development server:

   uv pip install "langgraph-cli[inmem]"
   uv run langgraph dev

---

Google Auth Setup

To enable Gmail and Calendar integration, you’ll need to set up your OAuth credentials. Follow this tutorial to create your credentials.json and token.json:

Google OAuth Setup Guide

Once created, place them in:

src/agent/credentials/

---

What's Next?

We're planning to enhance the assistant's memory using:

Episodic Memory

• Stores experiences and sequences from previous tasks
• Enables the assistant to "recall how" it performed a task
• Supports few-shot learning for behavior adaptation

Procedural Memory

• Encodes rules and routines the assistant follows
• Typically lives in code, prompt templates, or configurations
• Provides consistency across sessions

Just like humans remember both how to ride a bike (procedural) and a memorable bike ride (episodic), our assistant will evolve to remember both the rules and experiences to be more helpful.

---

Contributions

Open to ideas, improvements, and feature suggestions. Feel free to fork and submit a PR!

---

License

MIT License – see LICENSE for details.