🤖 Auto-Procure: Autonomous Supply Chain Agent

A Full-Stack AI Agent that autonomously monitors inventory levels, detects shortages, and drafts Purchase Orders for human approval.

---

📖 Overview

Auto-Procure solves the "Blank Page Problem" in supply chain management. Instead of a human manually checking spreadsheets for low stock, an Autonomous AI Agent wakes up, scans the database, and proactively drafts orders when inventory hits a critical threshold.

The system features a Human-in-the-Loop workflow: the AI proposes the order with reasoning (e.g., "Stock is 5, minimum is 10"), but a human manager must click "Approve" on the dashboard to finalize it.

---

🏗️ Architecture

The system is built as a monorepo with a clear separation of concerns:

1. The Brain (Backend)

• Framework: Python FastAPI
• AI Orchestration: CrewAI & LangChain
• LLM: OpenAI GPT-4
• Database: Supabase (PostgreSQL)

2. The Interface (Frontend)

• Framework: Next.js 15 (App Router)
• Styling: Tailwind CSS & Shadcn UI
• State: Server Actions & React Hooks

---

⚡ Key Features

• Autonomous Monitoring: The Agent connects directly to the SQL database to check real-time stock levels.
• Intelligent Reasoning: It doesn't just match numbers; it generates a natural language explanation for why an order is needed.
• Human-in-the-Loop UI: A dedicated "Approval Queue" in the dashboard allows managers to review AI drafts before execution.
• Real-Time Updates: The frontend refreshes instantly upon approval, providing immediate feedback (Optimistic UI).

---

🚀 Getting Started

Prerequisites

• Docker Desktop (Recommended method)
• OR: Node.js 18+ & Python 3.10+ (Manual method)
• OpenAI API Key & Supabase Credentials

Option A: Run with Docker 🐳 (Recommended)

The fastest way to test the system. No need to install Python or Node locally.

1. Clone the repository:

git clone [https://github.com/yourusername/auto-procure.git](https://github.com/yourusername/auto-procure.git)
cd auto-procure

2. Configure Environment: Create a .env file in the backend/ folder:

OPENAI_API_KEY=sk-...
SUPABASE_URL=http://host.docker.internal:54321  # Use this for local Docker (Windows/Mac)
   # OR use your cloud URL: https://xyz.supabase.co
SUPABASE_KEY=...

3. Launch the System: Run this command in the root folder:

docker-compose up --build

The Dashboard will be available at http://localhost:3000 and the API at http://localhost:8000.

---

Option B: Manual Installation 🛠️

Use this if you want to modify the code or debug locally.

1. Backend Setup

cd backend
python -m venv venv
# Windows: venv\Scripts\activate  |  Mac/Linux: source venv/bin/activate

pip install -r requirements.txt

2. Configure Keys Create a .env file in backend/ as shown in Option A.

3. Run Backend

python run.py
# Server starts at [http://127.0.0.1:8000](http://127.0.0.1:8000)

4. Frontend Setup Open a new terminal:

cd frontend
npm install
npm run dev
# Dashboard runs at http://localhost:3000

---

▶️ Usage / Demo

Once both servers are running, you can manually trigger the AI Agent to perform a supply chain check cycle.

Option 1: Via Terminal (API) Open a new terminal and run:

# Windows (PowerShell) - Note the .exe usage
curl.exe -X POST http://localhost:8000/run-agent

# Mac / Linux / Git Bash
curl -X POST http://localhost:8000/run-agent

Option 2: Via Dashboard

1. Navigate to http://localhost:3000.

2. View the Live Inventory Feed.

3. If an order is needed, the "AI Drafted Orders" section will appear automatically.

4. Click "Approve" to finalize the order.

---

🧠 How the Agent Works

The Supply Chain Analyst agent follows a strict logic loop:

• OBSERVE: Calls the check_low_stock tool to scan the products table in Supabase.

• REASON: Identifies items where current_stock < min_threshold.

• ACT: Groups items by vendor and calls the create_draft_order tool.

• RECORD: Writes a new row to the purchase_orders table with status pending_approval.

Example AI Output: "I have detected low stock for RTX 3080 (Count: 2, Min: 10). I am creating a draft order for 8 units to restore healthy levels."

---

📸 Screenshots

The Dashboard Interface Real-time inventory tracking with the AI Approval Queue active.

---

🔮 Future Improvements

• Email Notifications: Send an alert to the manager when a draft is created.

• Multi-Agent System: Add a "Price Negotiator" agent to check competitor prices.

• Historical Analysis: Use vector search to predict seasonal trends.

---

🔧 Troubleshooting

• Backend Connection Refused?

  Ensure your SUPABASE_URL in .env uses http://host.docker.internal:54321 if running Supabase locally. localhost inside a container refers to the container itself, not your machine.

• Frontend Fetch Failed?

  Ensure your frontend code uses the Docker service name (http://backend:8000) for server-side fetches, and localhost for client-side fetches.

---

🤝 Contributing

Contributions, issues, and feature requests are welcome!

---

📝 License

This project is MIT licensed.