README.md

metadata

title: LogisticsFlow OpenEnv
emoji: 📦
colorFrom: blue
colorTo: green
sdk: docker
pinned: false

📦 LogisticsFlow-OpenEnv

A real-world OpenEnv simulation for testing AI agents on Supply Chain Management, Budgeting, and Resource Allocation.

🎯 The Real-World Problem

Modern E-commerce fulfillment requires balancing strict budgets with customer satisfaction. AI agents must learn to prioritize high-value VIP orders, manage limited budgets, and anticipate stock-outs before they happen.

This environment simulates a live warehouse dispatch system. Agents are not just playing a game; they are optimizing a simulated business.

🧠 Environment Features

• Strict OpenEnv Compliance: Fully typed Action and Observation models using Pydantic.
• Continuous Partial Rewards: Agents receive granular reward signals for every successful shipment, not just a binary score at the end.
• Dynamic State Degradation: Customer satisfaction drops as orders age in the queue, forcing the agent to act efficiently.
• Deterministic Task Graders: Built-in programmatic graders (Easy, Medium, Hard) that return a strict 0.0 to 1.0 score based on budget retention and order completion.

🚀 Tasks & Difficulty

1. Easy (/reset/easy): Basic capability test. Agent must ship existing inventory.
2. Medium (/reset/medium): Constraint optimization. Agent is given a severely limited budget and must choose the cheapest shipping carriers to survive.
3. Hard (/reset/hard): Multi-step reasoning. The agent is faced with a "Stock-out Crisis" (0 inventory). It must realize the shortage, execute a restock command, and then fulfill the orders.

🛠️ Action Space

• ship: Dispatch an order (Requires order_id and carrier).
• restock: Purchase more inventory to fulfill future orders (Requires item).

📊 Observation Space

• inventory: Current stock counts (e.g., {"Electronics": 10}).
• pending_orders: Queue of orders with priority levels and age.
• budget: Available capital.

🏃 How to Run Locally

1. Install dependencies: pip install -r requirements.txt
2. Start the server: python -m uvicorn main:app --port 7860
3. The environment will be available at http://localhost:7860