📨 Sendora - The Invisible AI Email Agent

Sendora is a fully autonomous, AI-driven background agent built on the Gmail API and Google's Gemini models. Unlike traditional AI writing assistants that require you to learn a new email client, Sendora operates invisibly. It reads, categorizes, and answers incoming emails using a custom Retrieval-Augmented Generation (RAG) pipeline—pulling exact facts from your uploaded business PDFs to generate highly accurate replies while you sleep.

---

✨ Key Features

• Invisible Autonomy: Once connected via Google OAuth, Sendora uses backend cron jobs to monitor your inbox. You never have to abandon your native Gmail app.
• Custom Knowledge Base (RAG): Upload your pricing, policies, and FAQs as PDFs. Sendora embeds these into a MongoDB Vector Database and searches them before replying, eliminating AI hallucinations.
• Dynamic Intent Routing: Define custom labels (e.g., "Support", "Urgent"). Sendora categorizes the emotional intent of the sender and acts accordingly.
• Draft vs. Auto-Reply: Tell the AI to either silently save perfect responses into your Drafts folder for human review or send them instantly for zero-touch automation.
• Persona Cloning: Define your brand's voice (Friendly, Professional, Witty). Sendora guarantees the AI writes exactly how you would.
• Analytics Dashboard: A beautiful React UI tracking total emails processed, hours of typing saved, and detailed historical logs of every AI action.

---

🏗️ System Architecture

Sendora operates through asynchronous background polling.

flowchart TD
    Fetch[node-cron Fetches UNREAD emails] --> Extract[Extract Sender & Body]
    Extract --> Embed[Convert email to Vector]
    Embed --> VectorSearch[(Atlas Vector DB)]
    VectorSearch -->|Retrieved PDF Facts| Prompt[Assemble AI Prompt]
    Prompt --> Gemini{Gemini 2.5 Flash}
    Gemini -->|Outputs JSON| Parse[Parse Label & Action]
    Parse --> Draft[Save to Gmail Drafts]

Loading

---

🚀 Getting Started

Prerequisites

• Node.js (v20+ recommended)
• MongoDB Atlas account (required for Vector Search)
• Google Cloud Console account (for OAuth and Gmail API credentials)
• Google AI Studio API Key (for Gemini 2.5 Flash and embeddings)

1. Clone the Repository

git clone https://github.com/s4hil-dev/sendora-email-agent.git
cd sendora-email-agent

2. Environment Variables

Create a .env file in the /server directory and configure the following:

MONGODB_URI=......

GOOGLE_CLIENT_ID=......
GOOGLE_CLIENT_SECRET=......
GOOGLE_CALLBACK_URL=http://localhost:5000/auth/google/callback

GEMINI_API_KEY=......

3. Setup MongoDB Atlas Vector Search

You must create a Vector Search Index in your MongoDB Atlas Dashboard for the RAG feature to work.

1. Go to your Atlas Dashboard -> knowledges collection -> Atlas Search.
2. Create a new Vector Index named vector_index.
3. Use the following JSON configuration:

{
  "fields": [
    { "type": "vector", "path": "embedding", "numDimensions": 3072, "similarity": "cosine" },
    { "type": "filter", "path": "userId" }
  ]
}

4. Install Dependencies

# Install backend dependencies
cd server
npm install

# Install frontend dependencies
cd ../frontend
npm install

5. Run the Application

You need two terminal windows open:

# Terminal 1: Start Backend (Runs on port 5000)
cd server
npm run dev

# Terminal 2: Start Frontend (Runs on port 5173)
cd frontend
npm run dev

---

🔒 Security & Privacy Notes

Sendora requires "Restricted Scopes" (https://www.googleapis.com/auth/gmail.modify) to read and draft emails.

• Data Retention: Sendora does not store the body text of your incoming emails permanently. It only stores metadata IDs for analytics to protect client privacy.
• OAuth: User refreshTokens should be encrypted at rest in a production environment.

🔮 Future Roadmap

• Migration from node-cron polling to Google Cloud Pub/Sub Push Notifications for instantaneous processing.
• Implementation of BullMQ for robust background job queuing and API retry logic.
• "Human-in-the-Loop" Tinder-style swipe dashboard for rapid draft approvals.
• Microsoft Outlook / Office 365 integration.

---

License

This project is licensed under the MIT License.