Food cost pack
Actual against theoretical food cost, recipe costs, vendor price creep, deliveries and credits, counts, waste and 86s (7 agents)
Thirty-two small agents that check a restaurant's back office from the files it already exports, food cost, labor, tips, cash, delivery apps, fees, tax and compliance, and wait for a named person before anything is paid, sent or changed.
Actual against theoretical food cost, recipe costs, vendor price creep, deliveries and credits, counts, waste and 86s (7 agents)
Pour cost and liquor variance, menu engineering, discounts and promotions (3 agents)
Voids and comps, discounts, cash drawers, gift cards, no-shows, catering deposits (6 agents)
Tip pool and tip credit, overtime and split shifts, schedule against actual, labor by daypart, breaks and minors, set to your state (5 agents)
DoorDash, Uber Eats and Grubhub payouts against orders, marketing fees, card processing fees (3 agents)
Sales tax by jurisdiction, duplicate invoices and GL coding, utility bills, weekly prime cost (4 agents)
Equipment maintenance, health inspections and permits, food handler cards, allergen matrix (4 agents)
All 31 single-location agents, mapped to your exports and set to your rules, one training session
Everything in Whole restaurant for each location, plus the location comparison against the group
For multi-unit groups, franchise operators and accounting firms serving many restaurants: all 32 agents, one shared setup, every location mapped
Support after setup: $150 a month for one pack, $300 a month for the whole restaurant, $200 a month per location for groups. Includes export changes, rule and rate updates, and a monthly check-in.
Reads the files you export; it does not log in to your POS, scheduling, payroll, delivery-app, reservation, distributor or bank account.
Wage, tip, overtime, break, tax, inspection and allergen rules are settings you confirm with your accountant, payroll provider or health department; the defaults are examples.
Column layouts for Toast, Square, 7shifts, Sysco, DoorDash and the other named exports were written from knowledge of those systems; the first setup confirms them on your files.
Lost-sales and premium figures are estimates from your own numbers, not forecasts.
Not legal, tax, payroll, health-code or food-safety advice.
Report pages come in nine languages, translated by Claude and not yet reviewed by native speakers.
Your files and reports stay on your computer. Staff and guest names are masked in anything sent to the AI model. No agent pays, refunds, charges a card, changes the POS, publishes a schedule, runs payroll, files taxes or messages anyone: drafts wait for one named person.
What the recipes and sales say you used against what the counts and invoices say you did, per ingredient.
Every plate costed at today's prices, with the menu price that would meet your target.
Price per pound or each across invoices, prices over your bid, packs that quietly got smaller.
Shorts and cases refused at the door that were never credited.
Counts against expected, priced, and counts that look like the wrong unit.
What was thrown out and why, and the days nobody logged anything.
Bottles used against drinks sold, per spirit, and pour cost against target.
Stars, plowhorses, puzzles and dogs from your sales and plate costs.
Items that keep running out, for how long, and what that likely cost.
Self-approved voids, voids after payment, missing reasons, rates against sales.
Promos used after they ended, outside their hours, over their cap, or not on your list.
Expected cash against the count and the deposit, and shorts that repeat.
What you still owe on cards, by location, and cards whose numbers do not add up.
No-show rate, repeat no-shows, large parties with no deposit, fees your policy sets.
Deposits missing before the event, balances overdue after it, refunds owed.
Pool shares by hours and role, and cash wage plus tips against the minimum for your state.
Weekly and daily overtime, double time, split shifts, spread of hours and clopens, priced.
Early clock-ins, late clock-outs, no-shows and shifts nobody scheduled.
Labor against sales for breakfast, lunch, dinner and late night.
Missed, short and late meals, and minors working too long or too late.
Commissions over your rate, short payouts, error charges and orders never paid.
Ad and promo charges outside the campaigns you opted into, and what each campaign returned.
Effective rate, markup over your agreement, and fees you never agreed to.
Tax collected against your rates per location and category, marketplace orders kept apart.
Invoices entered twice, line math, and liquor or paper goods coded to food.
Usage spikes, rate jumps, estimated reads, duplicate bills and late fees.
Hoods, grease traps, fire suppression and refrigeration overdue or coming up.
Re-inspections due, critical violations not followed up, permits about to expire.
Missing, expired and expiring cards, and locations without a certified manager.
Allergens in a dish that the matrix does not declare, and free-from claims a recipe breaks.
Weekly food, beverage and labor against sales for each location.
Every location against the group median, on the same measures.
Each one does one job, on your computer, from a file you export. Nothing is paid, refunded, sent or changed in your POS until a named person on your team says so. Pairs with G6daddy's restaurant-and-bar AI agent: that one talks with your guests, these check the back office.
Measured in the final pass, 19 Nov 2026
On fictional test data, 19 Nov 2026. The clean set covers 31 of the 32 agents: menu engineering sorts every dish into a group, so a menu with no labels does not exist. This pass also found and fixed one wrong finding on the demo data and five wrong or noisy results on public data. The test files and the checks were written by the same team, so these numbers show the rules do what they say. How often they are right on your own files is measured at your setup.
git clone https://github.com/fpavia111-creator/fairtable-agents
cd fairtable-agents
npm install
node packages/fairtable-food-cost/bin/fairtable-food-cost.js demo
Open source, MIT licence, on GitHub: fpavia111-creator/fairtable-agents. Needs Node.js 20 or later. The demo runs on made-up files and needs no network and no setup.
| Source | What ran and what it found | Licence |
|---|---|---|
| NYC Department of Health, Restaurant Inspection Results (one ZIP code, 2024 to Nov 2026) 2026-11-19 | fairtable-health-inspections on 1,758 inspections at 648 restaurants (5,479 published rows, one per violation, grouped into one row per inspection). It showed four problems, now fixed: NYC scores count violation points (lower is better), NYC names routine visits Cycle Inspection, same-day records changed the result, and critical violations after a passing visit were reported as open (now a follow-up score setting). After the fixes, 0 disagreements in 499 checks against an independent re-count from the same rows. Tested from one year earlier, routine-due had flagged 36 of the 48 restaurants inspected in the next 45 days, and 57 of its 199 flags were inspected within 90 days. | NYC Open Data, public; NYC Open Data terms of use |
| U.S. Bureau of Labor Statistics average retail prices, 8 food items, via FRED 2026-11-19 | fairtable-price-creep on the last 13 monthly prices of each item, written as invoice lines. Price rises over 5% were reported for ground beef (+6.8%) and tomatoes (+9.1%), as the series show. The steady-creep rule was too noisy: it fired for 7 of 8 items, including items that ended cheaper. It now needs a current run of rises that adds up to 5%, and fires once (eggs, up three months running). 0 disagreements with an independent re-check. | BLS data, public domain; served by FRED, Federal Reserve Bank of St. Louis |
Public, lawful files only; no private data was used. The files are not copied into the repository: a script there downloads them again.
| Package | What a person reviews and decides |
|---|---|
| Food cost pack | A manager checks each flagged ingredient against the count sheet, recipe and invoices, then decides whether to recount, retrain, or change a recipe or vendor. |
| Bar and menu pack | The bar manager decides whether a pour gap is over-pouring, unrung drinks or a count error; menu and price changes are the owner's call. |
| Front of house and cash pack | A manager talks through each void, comp and drawer short with the staff member, and a person decides on any no-show fee, refund or gift card correction. |
| Labor pack | The owner or payroll provider checks wage, tip, overtime and break findings against your state's current rules and payroll records; the agents never run payroll. |
| Delivery apps and fees pack | Someone disputes commissions, error charges and card fees in each platform's portal or with the processor; a draft letter is sent only by a person. |
| Back office pack | The bookkeeper confirms tax rates and returns and decides which possible duplicate invoices to hold; nothing is paid or filed by the agents. |
| Compliance calendar pack | A manager books service visits, renews permits and certificates and confirms allergen labels with the kitchen; intervals and rules come from your own jurisdiction. |
| Whole restaurant | Every finding in every pack goes to a named person, who looks and acts; the agents only read files and write reports and drafts. |
| Small group | Each location manager reviews their own findings; the owner or operations lead reviews how each location compares with the group median. |
| Partner rollout | Each location's manager acts on its findings, and the firm or head office confirms each location's settings (wage, tax and inspection rules) with the client. |
| System | Report or export | Status |
|---|---|---|
| Toast | ItemSelectionDetails export | Not yet tested on a real export |
| Toast | TimeEntries export | Not yet tested on a real export |
| Toast | PaymentDetails export | Not yet tested on a real export |
| Toast | Void and discount details export | Not yet tested on a real export |
| Square | Item Sales export (CSV) | Not yet tested on a real export |
| Square | Timecards export (CSV) | Not yet tested on a real export |
| Clover | Item sales export | Not yet tested on a real export |
| Lightspeed Restaurant | Product sales report export | Not yet tested on a real export |
| Revel Systems | Product Mix report export | Not yet tested on a real export |
| 7shifts | Worked hours and wages export | Not yet tested on a real export |
| 7shifts | Schedule export | Not yet tested on a real export |
| Homebase | Timesheets export | Not yet tested on a real export |
| MarginEdge | Invoice line items export | Not yet tested on a real export |
| Restaurant365 | AP invoice detail export | Not yet tested on a real export |
| Restaurant365 | Inventory count export | Not yet tested on a real export |
| xtraCHEF by Toast | Invoice details export | Not yet tested on a real export |
| Sysco | Invoice download (CSV) | Not yet tested on a real export |
| US Foods | Invoice download (CSV) | Not yet tested on a real export |
| DoorDash Merchant Portal | Financials, Transactions export | Not yet tested on a real export |
| Uber Eats Manager | Payment details report | Not yet tested on a real export |
| Grubhub for Restaurants | Transactions export | Not yet tested on a real export |
| OpenTable | Reservations export | Not yet tested on a real export |
| Resy | Reservations export | Not yet tested on a real export |
| NYC Department of Health | Restaurant Inspection Results (open data CSV) | Tested on a public sample |
| U.S. Bureau of Labor Statistics via FRED | Average price series (CSV) | Tested on a public sample |
No layout has yet been tested on a customer's own export. Every setup starts by confirming the layout on your files, and any column can be mapped by hand.
When an assistant calls a tool, its model sees the result: totals, findings, amounts, quantities, dates, rates and finding codes in clear. It also sees text such as item, recipe and vendor names, locations, and free-text reasons and notes exactly as they were typed. Staff, guest and client names, phone numbers, emails, employee IDs, gift card numbers and certificate numbers are replaced by tokens such as A.R.#3f9a2c or card#7ecbb1 (checked with 118 fictional values in 14 agents, 0 leaks). The report and draft files written on your computer keep the real values. Results go to whatever model your assistant uses; for staff pay and schedule data, use a model covered by your own agreements or a local model.
Every rule that changes with the law or a contract is a setting, not code: the minimum and tipped cash wage, overtime and break rules, minors' hours, sales tax rates, marketplace channels, the allergen list, food handler grace periods, inspection and equipment intervals, card fee caps and delivery commission rates. Defaults live in each agent's src/logic.js, with the source named in the review. Your own values go in a settings JSON file that is merged over the defaults, and every result lists the settings it used. You change the file yourself, or support updates it when a rate or rule changes.
We send the list of reports by name. You export them from the systems you already use; nothing is connected.
Over a shared screen: the checks are installed, your columns are matched, and your own rules and dates are set.
The checks run on your files. We go through the worklist with you, and every flagged row is a person's call.
You run them whenever you like. Support keeps the dated rules and the export layouts current.
A fixed scope and a fixed price, agreed before any work starts. Your files stay on your computer the whole time.
Written by the check itself on made-up files. Every person, business and number is invented.
Tip pool and tip credit All samples