MyMail

An AI-powered email reply assistant for Gmail. MyMail fetches your inbox, drafts intelligent replies using your choice of LLM, and lets you review, edit, and send — all with a human-in-the-loop workflow.

Features

• Gmail Integration — Fetches emails from your primary inbox with full thread/conversation view
• Multi-Provider AI Drafts — Generate reply drafts using OpenAI (GPT-4o-mini), Anthropic (Claude Sonnet), or Google Gemini
• Knowledge Base (RAG) — Upload documents to a vector knowledge base; relevant context is automatically retrieved and injected into LLM prompts for more personalized replies
• Human-in-the-Loop — AI drafts replies but never sends without explicit user approval. Every draft can be reviewed and edited before sending
• Feedback Loop — Star rating (1-5) and optional text feedback on every sent reply
• Dark Mode — Full dark mode support across all components
• LLM Context Panel — See exactly what context was sent to the LLM for each draft

Architecture

frontend/          Next.js (App Router, Tailwind CSS)
backend/           Python (FastAPI)
supabase/          Database migrations

• Auth: Google OAuth via NextAuth → tokens encrypted and stored in Supabase
• Database: Supabase (Postgres + pgvector)
• Tables: users, emails, drafts, feedback, knowledge_base

Prerequisites

• Node.js 18+
• Python 3.11+
• A Supabase project (with pgvector extension enabled)
• Google Cloud project with OAuth credentials and Gmail API enabled
• At least one LLM API key (OpenAI, Anthropic, or Gemini)

Setup

Backend

cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Create backend/.env:

SUPABASE_URL=your_supabase_url
SUPABASE_SERVICE_KEY=your_supabase_service_key
GOOGLE_CLIENT_ID=your_google_client_id
GOOGLE_CLIENT_SECRET=your_google_client_secret
ENCRYPTION_KEY=your_fernet_key
JWT_SECRET=your_jwt_secret

Run the Supabase migrations in order from supabase/migrations/.

Start the server:

uvicorn app.main:app --reload

Frontend

cd frontend
npm install

Create frontend/.env:

NEXT_PUBLIC_BACKEND_URL=http://localhost:8000
GOOGLE_CLIENT_ID=your_google_client_id
GOOGLE_CLIENT_SECRET=your_google_client_secret
NEXTAUTH_SECRET=your_nextauth_secret
NEXTAUTH_URL=http://localhost:3000

Start the dev server:

npm run dev

Usage

1. Open http://localhost:3000 and sign in with Google
2. Click API Key in the navbar to add your LLM API key(s)
3. Browse your inbox and click an email to view the full thread
4. Click Generate (choose provider from dropdown) to draft a reply
5. Review and edit the draft, then click Send
6. Rate the reply quality with star rating and optional feedback

API Key Notes

• All API keys are stored in your browser's localStorage only — never sent to the backend for storage
• Keys are sent per-request to the respective LLM API for draft generation
• OpenAI key is required for Knowledge Base (RAG) features, even when using Anthropic or Gemini for generation (OpenAI embeddings are used for vector search)

Deployment

Frontend → Vercel

• Set root directory to frontend/
• Configure env vars: NEXT_PUBLIC_BACKEND_URL, GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, NEXTAUTH_URL, NEXTAUTH_SECRET

Backend → Railway (or Render for free tier)

• Configure env vars from backend/.env
• Update Google OAuth redirect URIs in Google Cloud Console
• Update CORS origins in backend/app/main.py to allow the Vercel frontend URL