Concierge

Project Note (Current Focus)

Active development is currently focused on the Chrome extension implementation in concierge-chrome/ (Manifest V3, local-first AI models). Start with concierge-chrome/README.md

---

This is an agentic script which automates product search, cart management, and checkout on Amazon.in. At it's core, it uses Playwright, with LLM-assisted product selection and guardrails for spend limits.

Overview

Concierge is a full-stack workflow for Amazon.in ordering. A FastAPI backend drives Playwright automation while a Vite + React dashboard lets you watch each step in real time. The backend handles product search, selection, cart management, and checkout. The UI turns this into a readable timeline showing what the agent is doing. On the frontend, lucide-react icons and custom styling power the status indicators and card layouts.

• FastAPI endpoints orchestrate Playwright sessions and expose order state for the UI.
• The dashboard surfaces task progress, order metadata, and system health in one place.
• Product selection runs through Llama 3.1 via Groq with a safe fallback model.
• The flow supports retail and Fresh carts, proceeds through checkout, and can use Amazon Pay Later.

Tech Stack

Backend

• Python 3.9+ - Core automation runtime
• Playwright - Browser automation for Amazon.in
• FastAPI - REST API server
• Uvicorn - ASGI server
• Groq API - LLM provider (Llama 3.1)
• Pandas - Order history tracking

Frontend

• React 18 - UI framework
• Vite - Build tool and dev server
• Lucide React - Icon library

Integration

• Amazon.in - Target e-commerce platform
• Amazon Pay Later - Payment method

Requirements

• Python 3.9+
• Playwright (Chromium) and its browser binaries
• Groq API key
• Amazon.in account credentials
• FastAPI + Uvicorn (for the backend API)

Setup

1. Create a virtual environment and install dependencies:

   python -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt
   playwright install

Usage

Web Dashboard (Recommended)

1. Start the backend server:

   uvicorn backend.server:app --reload --host 0.0.0.0 --port 8000

2. Start the frontend (in a separate terminal):

   cd frontend
   npm install
   npm run dev -- --host 0.0.0.0 --port 5173

3. Open your browser to http://localhost:5173 and place orders through the dashboard

CLI Mode

You can also run the script directly from the command line:

python backend/amazon_order.py

You will be prompted to choose a mode:

1. Test Mode (dry run, no order placed)
2. Real Mode (places orders)
3. Login / Refresh Session Only

Then enter product queries like 'Hershey's Cookies and Creme Bar' or 'Amul Toned Milk 1L'

Configuration

• REQUIRE_CONFIRMATION: confirmation guardrails

Logs and session data

• backend/logs/amazon_orders.log: runtime logs
• backend/logs/amazon_session.pkl: saved login session

Future Plans

• Reinstate persistent login and session restore functionality with re-authentication handling during checkout flow
• Strengthen product selection by refining the LLM prompt, adding heuristic filters, and surfacing confidence metrics for each suggestion.
• Reintroduce configurable spend tracking (per-order/daily) with editable limits, spend history visualization, and optional alerts before crossing thresholds.
• Multi-account support for switching between different Amazon accounts
• Expand beyond Amazon.in to other platforms: food delivery (Swiggy, Doordash), travel booking (MakeMyTrip, Expedia), grocery delivery (BigBasket, Blinkit), and general e-commerce (Flipkart)

Notes and safety

• Real Mode will place actual orders, and is only recommended when user is closely following the agent's actions.
• This is intended as a demo for now; and users are advised to use Test Mode
• The agent launches the locally installed Chrome browser via Playwright's channel="chrome" flag; keep Chrome up to date so automation stays compatible.
• Amazon may prompt for OTP during login.
• The checkout flow assumes Amazon Pay Later is available and selected during checkout, and that the default address is the correct one.