VendingBench Simulation

This repository implements an autonomous vending‑machine business simulation. The agent runs daily loops, uses tool‑calling (email, storage queries, waiting for next day, etc.), and logs everything to a SQLite database.

New Features (Added in the latest update)

1. Back‑room ➜ Vending‑Machine Restocking – After a supplier delivery is processed, the simulation now automatically moves matching items from back‑room storage into empty vending‑machine slots.
2. Sales Simulation – When items are stocked, daily sales are calculated, logged in the sales table, and the balance is updated.
3. Reporting Script (report.py) – A lightweight CLI script that prints:
   • Balance history over time
   • Email activity summary (inbound/outbound counts)
   • Current back‑room storage inventory
   • Daily sales totals
4. Tool‑Calling Support for Cerebras – The model client now forwards tools schemas to the Cerebras endpoint, parses tool_calls correctly, and stores them as JSON.
5. Database Enhancements – Added a machine_slots table to persist the vending‑machine slot state each day.

How to Run the Simulation

# Activate the virtual environment (already set up)
source venv/bin/activate

# Run a short simulation (e.g., 5 actions)
python run_sim.py --max-messages 5

The simulation prints daily reports, executes tools, and updates the SQLite DB at /home/hnai/vending_simulation.db.

Generating a Report

python report.py

The script reads the same SQLite DB and outputs concise tables of balance, emails, storage, and sales.

Repository Structure (relevant files)

• agent.py – Core autonomous agent logic.
• main_simulation.py – Orchestrates daily cycles, deliveries, restocking, and reporting.
• model_client.py – Handles calls to the Cerebras API with function‑calling support.
• tools.py – Implements send_email, check_storage_quantities, wait_for_next_day, etc.
• storage.py – Back‑room inventory and pending deliveries.
• email_system.py – Email sending/receiving with DB logging.
• database.py – SQLite schema (including machine_slots).
• report.py – Quick‑look reporting CLI.

Setup & Dependencies

# Install dependencies (inside the repo)
pip install -r requirements.txt

Make sure the .env file contains a valid CEREBRAS_API_KEY (lower‑case csk-…).

---

Feel free to extend the simulation (add more items, pricing, or richer supplier interactions) and use the report.py script to monitor performance over time.