Real case · In production since May 2026
554 autonomous actions. 26 agents running 24/7. 98.2% uptime with no human intervention. This is the story of how a premium cafe in Monterrey became the smartest restaurant in the country.
The restaurant
AMALAY is not a chain restaurant with an IT department. It is an independent cafe run by its owner, with all the limitations and pressures that implies.
Every peso counts. Every operational decision directly impacts the margin. And until May 2026, all of those decisions depended on one person reviewing numbers manually at the end of the day.
The challenge wasn't adopting technology. It was finding technology that worked on its own.
The problem
This is what happened every day at AMALAY before connecting the 26 agents. It's not a hypothetical scenario -- it's what thousands of independent restaurants in Mexico live through.
The owner logged into the POS system every night to review numbers. No comparisons, no trends, no historical context.
Every morning, the reporting agent sends an executive summary with KPIs, comparisons vs. last week and 3 priority actions.
Suspicious cancellations, unauthorized discounts and irregular patterns went completely unnoticed.
The anti-fraud agent analyzes cancellations, discounts and per-server patterns every Friday. It detected $8,400 MXN/month in suspicious patterns.
The implementation
It wasn't a 6-month project with consultants. It was a direct, iterative connection, without pausing the restaurant's operation.
We connected the POS system (Wansoft) to Supabase via an automated scraper. Without changing the existing POS, without installing anything at the restaurant.
The first agent in production: the Morning Briefing. At 7AM, the owner received a complete summary of sales, reservations and the day's actions via Telegram.
Pending reservations, staleness alerts, anomaly detector and wansoft query (natural-language questions about data, 24/7).
Staffing optimizer, menu engineering, anti-fraud, tips analyzer, kitchen quality, table time, close predictor, upselling. All agents deployed.
With two weeks of historical data, the Close Predictor reached 97% accuracy compared to the actual close.
554 autonomous actions completed. 98.2% uptime. The owner spends 2 minutes a day reviewing AI reports on Telegram.
The results
Every metric comes directly from AMALAY's production logs. No generous rounding, no asterisks.
The agents in action
Each agent has a specific role. These are the ones that have had the biggest impact on AMALAY's daily operation.
Analyzes cancellation, discount and comp patterns per server every Friday. Cross-references historical data to detect statistical anomalies.
At 2pm, 4pm and 6pm it generates a prediction of how much will be sold by the end of the day.
Classifies every dish as a star (high profitability + high sales), cow, dog or puzzle.
The owner types a natural-language question in Telegram and the orchestrator routes it to the right agent.
"I used to close the restaurant and stay 45 minutes going over numbers. Now everything arrives on Telegram before we even open. What impressed me most was when the anti-fraud agent detected a discount pattern I never would have seen. Fullsite isn't just another tool -- it's like having an operations manager who never sleeps."Owner, AMALAY Coffee & Market · Monterrey, NL · May 2026
Your turn
It doesn't matter which POS you use. We connect your data, deploy the agents and in two days you have your first briefing running. Per-location pricing per proposal, no surprises.