Automation-Restaurant

Automation Restaurant

One connected operating system for a restaurant — customer ordering, kitchen tickets, recipes, inventory, suppliers, finance, staff and AI, all on the same live data.

Automation Restaurant website

Contents: Problem statement · Proposed solution · Screenshots · Technology stack · AI methodology · AI models and APIs · Development tools · Setup and installation · Team member contributions

Problem statement

New restaurants often face high upfront costs and technical complexity when they have custom restaurant management software built for billing, orders, inventory, menus and kitchen operations. Building and maintaining that software takes developers, time and money that a new restaurant rarely has.

Proposed solution

Automation Restaurant is an AI-powered, subscription-based Restaurant Management System (RMS). It lets a restaurant manage billing, orders, menus, inventory, kitchen operations, suppliers and finances on one central platform.

Why Automation Restaurant

  Unique selling point What it means for the owner
🔗 One live data layer An order flows to the kitchen, stock, suppliers, finance and the dashboard at once. Nobody re-types anything.
🧾 Recipe-driven stock Every sale deducts its recipe’s exact grams, millilitres and pieces, so food cost is known per dish as it sells.
🍗 Shared-ingredient availability When chicken runs low, every dish that uses it is recalculated. Guests only see what the kitchen can actually make.
🤖 AI Smart Import Upload an existing menu, recipe, inventory or supplier PDF. AI builds a draft, and nothing is saved until the owner approves it.
💬 AI on every side The owner assistant answers from live data and reads or writes PDFs. Guests get a menu assistant, and visitors get a website AI Guide.
📱 QR ordering, no app Guests scan the table, order, track it live from kitchen to table, order more and leave feedback. They never sign up.
💰 True profit, live Every sale, refund, food cost, expense, supplier bill and cash movement is written to one financial ledger as it happens, so P&L, payables and the day close always agree. Exported as a branded PDF or Excel file.
🔁 Closed purchasing loop Low stock → purchase order → goods received → stock updated → supplier bill due on that supplier’s terms.
🛡️ Portals built from permissions Each staff portal is composed from individual permissions, enforced by the database rather than just hidden buttons.
🏢 A database per restaurant Every restaurant gets its own Supabase project, provisioned automatically at sign-up.

Screenshots

Owner dashboard
Owner dashboard: net sales, tax, gross and net profit, margins, revenue and payment mix		 Kitchen Display
Kitchen Display: live tickets, the ingredients each order consumes, and portions left
AI Smart Import
AI Smart Import: a menu PDF becomes 6 new items in 5 categories, each linked to its recipe, ready to approve		 Menu management
Menu management: price, availability, linked recipe and food cost % per dish
Customer menu
Customer menu: categories, deals that show the saving, notes, promo codes and tax before checkout		 Tables and QR codes
Tables & QR codes: every table gets its own QR code (blurred here)
Order tracking
Order tracking: placed → in the kitchen → ready → served, plus order more		 Guest feedback
Guest feedback: overall rating plus food, service, speed, cleanliness and ambiance
Customer menu assistant
Customer menu assistant: recommends dishes and calculates real deal savings		 Website AI Guide
Website AI Guide: explains the product and walks owners through setup
Profit and loss report
P&L report: a branded PDF of sales, food cost, expenses and net profit		 Sign-up
Sign-up: one form, secure payment, then automatic provisioning

Screenshots use a test restaurant with test data.

What it does

Area Features
Guests Table QR codes, no-login ordering (/order/<slug>?table=…), search and categories, deals with the saving shown, special instructions, promo codes, tax shown before checkout, live order tracking (placed → in the kitchen → ready → served), order more from the same table, five-part ratings, “Ask about the menu” AI
Kitchen Kitchen Display (KOT queue → preparing → ready), the ingredients each ticket consumes, portions left per dish, KOT history
Menu & recipes Categories, variants, modifiers, deals, promotions; recipes in base units (g / ml / piece) linked to dishes; per-dish food cost and margin; availability calculated from stock
Inventory Stock on hand, low-stock thresholds and email alerts, automatic deduction by recipe on every order, shared-ingredient impact across dishes
Purchasing Supplier directory with payment terms, purchase requests → purchase orders → goods received → stock update → supplier invoices with a 3-way match (PO ↔ delivery ↔ invoice), typed exceptions, credit notes, and AI reading of invoice PDFs and photos
Finance Finance overview (P&L vs the previous period, food cost % vs target, payables aging, needs-attention list), append-only ledger with corrections, expense approval workflow, supplier payables, cash movements and a locked day close, branded PDF and Excel finance reports — see Finance model
Staff & access Staff without logins (name, job, shift), custom portals built from individual permissions with a live preview, approvals, audit log, attendance and scheduling
AI Owner assistant (live answers from restaurant data, reads and creates PDFs, every chat saved), Smart Import (upload a menu, recipe, inventory, supplier, table, staff or PO document → reviewable draft → approve), customer menu assistant, website AI Guide
Brand Brand Kit (logo, colours, receipt style) applied to the menu, receipts, invoices and reports
Branches Owner panel for a group of locations: create independent branches (each its own database), a login per branch (password shown once), per-branch receipt details, rename, archive / restore, delete permanently, side-by-side comparison, one-click switching between workspaces — see Multi-branch
Platform Marketing site, self-service sign-up and payment, automatic provisioning, plan tiers and upgrades, super-admin console

How a restaurant goes live

Step What happens
1. Sign up Pick a plan on the website, fill in one form, pay. The owner connects a free Supabase account and a dedicated database is provisioned automatically, then a set-password email is sent.
2. AI import Upload existing menu, recipe, inventory and supplier files; review the draft (new / changed / missing prices); approve.
3–6. Check the data Suppliers, inventory (costs and alert levels), recipes (food cost per dish), menu (each dish linked to its recipe).
7. Portals Give each job its own portal from individual permissions.
8. Order Guests scan the table QR code, or the cashier rings the order up.
9. Kitchen The KOT appears on the Kitchen Display; the guest’s tracking page follows it.
10. Stock Every ingredient in the order’s recipes is deducted; availability updates.
11. Purchasing Low stock alerts the owner and drafts a purchase order to the supplier.
12. Results Dashboard, reports and the AI assistant all read the same numbers.

Technology stack

Layer Technologies
Frontend (apps/web) Next.js 15 (App Router), React 19, TypeScript, Tailwind CSS, Lucide icons, Recharts (finance and performance charts), jsPDF + jspdf-autotable (PDF reports), qrcode (table QR codes), Zod (input validation)
Backend (apps/api) Node.js, Express and TypeScript, deployed as Vercel serverless functions; Zod (request validation), ExcelJS (Excel exports), unpdf (reading supplier invoice PDFs), Nodemailer / Resend (transactional e-mail and supplier messages), Stripe (subscriptions and plan billing)
Database & multi-tenancy Supabase PostgreSQL with a dedicated Supabase project for each restaurant, so restaurant data is physically separate
Control plane A central Supabase project for restaurant registrations, subscriptions, plans and connections to owners’ Supabase accounts
Provisioning Supabase OAuth and the Management API create each restaurant’s database in the owner’s own Supabase account, including independent branches
Security About 78 Row-Level Security policies and 247 restricted (security-definer) database functions; permissions are enforced in Postgres, not only in the UI
Auth & realtime Supabase Auth, with separate session slots so the owner panel, staff portals and branches can stay signed in side by side; Supabase Realtime for live kitchen, order and checkout updates
Schema management 107 versioned tenant migrations plus a reusable template schema for every new restaurant
AI Google Gemini (Flash and Flash-Lite) as the primary provider; the Anthropic Claude SDK as a second provider — see AI models and API usage
Infrastructure Vercel for both apps, redeployed automatically from GitHub; npm workspaces monorepo (apps/web, apps/api, packages/shared)
Testing Test scripts run with tsx against real tenant databases inside transactions that are rolled back afterwards

AI methodology and implementation

AI in Automation Restaurant is built to deliver practical outcomes across day-to-day operations, always on the restaurant’s own live data and always under human control.

AI feature What it does Outcome
Owner & staff assistant Answers questions about sales, profit, food cost, stock, suppliers, kitchen, staff and branches (one branch, or all side by side), and proposes actions (drafting a purchase order, adjusting stock, adding menu items). Knows whether it is in the owner panel or a branch, and who it is talking to Faster decisions; less time spent digging through reports
AI Guide Step-by-step, role-specific help on the website and inside the app, including setting up and managing branches and fixing a failed branch setup Faster staff onboarding, less training effort, fewer operational errors
AI Import Turns prompts or uploaded menu, recipe, inventory, supplier, invoice, purchase-order, staff and table files into structured, validated drafts Much less manual data entry; faster menu and stock setup
Customer AI Helps guests explore the menu, understand ingredients and find relevant deals A better ordering experience, more conversions and a higher average order value
Billing assistant Answers the owner’s questions about plans and the subscription, and explains that branches are billed on the main plan while their Supabase projects are Supabase billing Self-service billing support

How it is implemented

  1. Tools, not guesses. The assistant reads business data only through 71 permission-checked tools (53 that read figures, 17 that propose actions, and one that writes a PDF document). Each tool checks the signed-in user’s permissions before it runs, so the AI sees only what that person may see.
  2. Human confirmation. An action tool only proposes a change. It runs after a person taps Confirm. The AI never approves expenses or invoices and never makes payments.
  3. Drafts, then approval. AI Import (aiDocumentEngine) extracts text from PDFs, images, spreadsheets and text files, asks the model for output that must match a fixed schema, validates and cleans the result, compares it with live data (new / changed / missing), and shows a draft. Nothing is saved until the owner approves it.
  4. Grounded answers. Figures come from the same ledger and reports as the rest of the app; the AI is instructed never to invent numbers.
  5. Knows the platform as it is now. Each assistant’s instructions describe the current product, and are updated with every feature:
    • the owner panel, independent branches and their management;
    • branch logins and receipt details;
    • what a branch manager can and can’t do.

    A workspace line tells the owner assistant where it is (the owner panel with its branches, or one branch) and who it is talking to (the owner or the branch manager). It can’t create, archive or delete branches, set logins or change receipt details. It explains where to do that instead, and it never states or asks for a password.

  6. One platform. Combined with POS, kitchen, inventory, recipes, suppliers and finance, the AI reduces administrative work, improves stock visibility and keeps the customer experience consistent.

AI models and API usage

Automation Restaurant uses Google Gemini as its primary AI provider, with the Anthropic Claude API as a built-in alternative. The server picks the provider from its configuration (GEMINI_API_KEY first, then ANTHROPIC_API_KEY); if neither is set, AI features switch off gracefully. The live deployment uses Gemini.

Provider Models Used for
Google Gemini gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-3.8-flash, gemini-3.7-flash, gemini-3.5-flash, gemini-flash-latest, gemini-flash-lite-latest (default GEMINI_MODEL: gemini-flash-lite-latest) Every AI feature (primary)
Anthropic Claude claude-sonnet-5 (configurable with AI_MODEL) Every AI feature (alternative)
API How it is used
Gemini streamGenerateContent (generativelanguage.googleapis.com/v1beta) Streaming chat replies, sent to the browser as server-sent events so answers appear as they are written
Gemini generateContent Structured one-shot tasks such as document import and the AI Guide
Gemini function calling Lets the model call the app’s permission-checked tools
Anthropic Messages API (@anthropic-ai/sdk) The same features with tool use, when Claude is the configured provider
Endpoint Feature
/api/ai Owner & staff assistant (71 tools; actions confirmed through /api/ai/confirm)
/api/public/ai/chat Customer AI on the guest menu (read-only, rate-limited per visitor)
/api/public/guide-ai/chat, /api/guide-ai/chat AI Guide (public and signed-in)
/api/saas-ai/chat Billing assistant (owner only)
Import routes (menu, recipe, inventory, supplier, supplier price, supplier invoice, PO, staff, table) AI document import into reviewable drafts

Reliability

Security

Development tools used

Tool Used for
Git & GitHub Version control; every push to main triggers a deploy
Vercel Hosting, preview and production deployments, cron jobs
npm workspaces Monorepo for the web app, the API and the shared package
TypeScript compiler (tsc) Type checking across packages/shared, apps/api and apps/web
tsx Running the API in development and the test and migration scripts
Next.js build / dev server Local development and production builds of the web app
Supabase (dashboard, CLI, Management API) Databases, auth, migrations and automated provisioning
Stripe dashboard Plans, prices and webhooks
Google AI Studio Gemini API keys and prompt / function-declaration testing (docs/ai-studio)
Claude Code AI-assisted development, code review and testing
PostCSS & Autoprefixer Tailwind CSS build pipeline

Finance model

Every finance screen, report and AI answer reads from one place: the financial ledger (financial_events). Database triggers write an entry whenever money moves, so nobody re-types figures and the numbers cannot drift apart.

 orders · payments · refunds ─┐
 expenses (approved / paid) ──┤
 supplier invoices · payments ┼──>  financial_events  ──>  Finance overview · Ledger · P&L
 credit notes · stock · waste ┤     (append-only)          Day close · PDF / Excel · AI
 cash movements · day close ──┘

What the ledger records

Category Written when Effect
Sales An order is served or paid; reversed on refund or void Net sales (subtotal − discount − refunds)
Food cost The same orders, using each line’s recipe cost COGS, the same figure the profit report uses
Customer payments Payment taken, refunded or voided Money held from customers
Expenses An expense is approved (never while still a draft or submitted) Operating cost on its expense date
Supplier payables An invoice is approved (+); supplier payment or credit note (−) What the restaurant owes
Inventory and waste Goods received, stock adjustments, spoilage, at cost Stock value
Cash, day close, purchasing Till movements, counts, closes, purchase orders Information only; no balance change
Adjustments A person with Adjust ledger posts a correction with a reason Fixes a figure; nothing existing is changed

Ledger rows can’t be edited or deleted, even by the database owner. A mistake is fixed by posting a reasoned correction, which is also written to the audit log.

The workflows

Separation of duties

Each step needs its own permission, so the person who records, approves and pays can be three different people.

Step Permission
Record or edit an expense finance.create_expense / finance.update_expense
Approve or reject an expense finance.approve_expense
Pay an expense finance.pay_expense
Create or match a supplier invoice invoices.create / invoices.match
Approve or reject a supplier invoice invoices.approve (not payables.manage)
Pay a supplier payables.record_payment
Cash movements cash.manage
Close / reopen a day finance.close_day / finance.reopen_day
Post a ledger correction finance.adjust_ledger
See food cost and profit finance.view_cogs / finance.view_profit

The owner holds all of them. Other roles and portals get exactly the ones they are granted, and each one works the same way in a custom portal. Every check runs in the database: hiding a button is never the only protection. The AI never approves or pays anything.

Finance screens and reports

Operating profit = net sales − food cost − approved expenses dated in the period.

Finance overview
Finance overview: net sales, food cost, gross and operating profit, payments received, money owed, food cost % against target, what needs attention, month by month		 Ledger
Ledger: every money event with its business day and a link to its source; totals per category; corrections only, never edits

Multi-branch

On the Enterprise plan (branches.multi) an owner runs several locations as independent restaurants under one owner panel. A restaurant on a single-location plan keeps working exactly as before.

Owner panel and branches

Managing a branch (Branches → Manage, owner only)

Option What it does
Branch login Sets the branch manager’s email (and name). The password is typed or generated (e.g. Ab3k-9QxP-7mTd) and shown once to copy. Generate new password makes a new one. Passwords are never stored readably, so they can’t be shown again. Sign-in address: /r/<branch-slug>/login.
Receipt details That branch’s own address, phone, email, website, tax/NTN number and receipt footer, printed on its receipts and invoices only. Set while creating the branch, editable here any time.
Edit Renames the branch. Its code and address stay the same.
Archive / Restore Archive switches the branch off: no sign-ins, no orders, and it leaves the figures. Every record is kept, and it can be restored.
Delete permanently Deletes the branch’s Supabase project and all its data. The owner types the branch code to confirm. Cannot be undone.

What a branch manager can do

Branch manager Access
Orders, checkout, kitchen, tables & QR ✅
Menu, deals, recipes, inventory ✅
Suppliers, purchasing, expenses, finance, day close ✅
Staff, shifts, kiosk portals, AI assistant, AI Guide ✅
Billing & subscription ❌
Restaurant settings (currency, tax, branding, receipt details) ❌
Change password ❌ — only the owner changes passwords inside a branch
Branch creation, branch comparison, other branches ❌

These rules are enforced by the database as well as by the screens:

Safeguards

Where it lives

Piece Where
Control-plane columns (parent_tenant_id, branch_code, copy status) supabase/control-plane/0025_branch_tenants.sql
Receipt details held until a new branch is ready (branch_setup) supabase/control-plane/0026_branch_setup.sql
Create, provision, copy, metrics, logins, receipt details, archive, delete, owner sign-in apps/api/src/lib/branchTenants.ts
API (/api/owner-branches) apps/api/src/routes/ownerBranches.ts
Owner panel overview, switcher, Branches manager, comparison apps/web/src/components/OwnerBranches.tsx
Compare branches page apps/web/src/app/r/[slug]/(portal)/branches/compare
One-time sign-in page apps/web/src/app/r/[slug]/auth/sso
API endpoint Purpose
GET /api/owner-branches The branches with their figures for a period
POST /api/owner-branches Create a branch (idempotent per code; optional receipt details)
POST /:id/advance, POST /:id/retry Move setup forward / retry a failed setup or copy
POST /:id/open One-time owner sign-in into the branch
GET/POST /:id/login The branch login (email only) / create or change it, with the password returned once
GET/POST /:id/receipt The branch’s receipt details
PATCH /:id Rename
POST /:id/archive, POST /:id/restore Switch off and back on, keeping all data
POST /:id/delete Delete the branch’s project and data (confirm = its code)

The earlier in-database branch model (tenant migrations 0098–0106) is retired: its screens were removed and migration 0107 clears the old per-login branch limits. Its data stays in the database.

Architecture

                 Control-plane Supabase project
                 (tenants, plans, subscriptions, provisioning, registry)
                               │
  website sign-up ──> apps/api ──> Supabase OAuth / Management API
                               │      create project → apply tenant schema → seed roles + settings
                               │
   restaurant A project    restaurant B project    restaurant C project …
   (own auth users, menu,  (fully isolated —       (RLS on every table,
    orders, stock, chats)   own database)           permission-checked RPCs)

Repository layout

Path What it is
apps/web Next.js 15 (App Router): marketing site, sign-up and onboarding, restaurant portal, custom portals, guest ordering, super-admin
apps/api Express API (deployed to Vercel as a serverless function): provisioning, billing, staff, AI assistant and imports, customer AI, AI Guide, cron
packages/shared Plan tiers and feature entitlements shared by API and web
supabase/control-plane Migrations for the central registry project
supabase/tenant-template/schema.sql Full schema applied to every new restaurant
supabase/tenant-migrations Numbered migrations rolled out to existing restaurants
docs/ai-studio System prompt and function declarations exported for Google AI Studio

Main routes

Route For
/, /pricing, /guide, /get-started Public website, plans, setup guide, sign-up
/onboarding/… Payment confirmation, database connection, provisioning status, set password
/r/<slug>/login Staff sign-in for one restaurant
/r/<slug> Owner dashboard and every portal page (orders, KDS, menu, recipes, inventory, purchasing, suppliers, finance, staff, AI, settings …)
/r/<slug>/finance, /r/<slug>/finance/ledger Finance overview and the ledger
/r/<slug> (multi-branch) Owner panel overview: a card per branch and this month’s comparison
/r/<slug>/branches Owner panel: create, manage (login, receipt details, rename, archive, delete) and open branches (Enterprise)
/r/<slug>/branches/compare Compare branches side by side for any period
/r/<branch-slug>/login A branch’s own sign-in (e.g. /r/bbq-tonight-dha/login)
/r/<slug>/auth/sso One-time sign-in into a branch workspace
/r/<slug>/portal/<portalKey> A custom portal, composed from its permissions
/order/<slug>?table=… Guest ordering, tracking and feedback
/admin Platform super-admin

Plans

Plan Monthly Annual (per month)
Starter $49 $39
Professional $129 $103
Enterprise $299 $239

Features are enforced per plan in both the API and the portal (packages/shared).

Setup and installation

Prerequisites: Node.js 20+ and npm, Git, a Supabase account (for the control-plane project and restaurant projects), and optionally a Google Gemini or Anthropic API key (AI features), a Stripe account (payments) and Resend or SMTP (e-mail).

  1. Clone and install

    git clone https://github.com/shiraz-eng/Automation-Restaurant.git
cd Automation-Restaurant
npm run setup      # npm install + build packages/shared

  2. Create the control plane. Create a Supabase project and run the files in supabase/control-plane/ in order (SQL editor or Supabase CLI). Restaurant databases are created automatically later, from supabase/tenant-template/schema.sql.

  3. Set environment variables (names only — never commit values):

    • apps/api/.env — SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY (control plane), SUPABASE_ACCESS_TOKEN, SUPABASE_ORG_ID, SUPABASE_OAUTH_CLIENT_ID, SUPABASE_OAUTH_CLIENT_SECRET, SUPABASE_OAUTH_REDIRECT_URI, APP_URL, ALLOWED_ORIGINS, GEMINI_API_KEY (optional GEMINI_MODEL) or ANTHROPIC_API_KEY (optional AI_MODEL), RESEND_API_KEY or SMTP_*, EMAIL_FROM, CRON_SECRET, STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, STRIPE_PRICE_MAP, PAYMENTS_MODE. See apps/api/src/env.ts for the full list and defaults.
    • apps/web/.env.local — NEXT_PUBLIC_CONTROL_PLANE_URL, NEXT_PUBLIC_CONTROL_PLANE_ANON_KEY, NEXT_PUBLIC_API_URL, NEXT_PUBLIC_SITE_URL.

  4. Run locally

    npm run dev:api    # API on http://localhost:4000
npm run dev:web    # web on http://localhost:3000

  5. Check and build

    npm run typecheck                            # shared + api + web
npm run build -w @automation-restaurant/web  # production build of the web app

  6. Create a restaurant. Open http://localhost:3000, pick a plan and sign up. The restaurant’s own database is provisioned automatically and a set-password e-mail is sent to the owner.

Database changes

Every tenant schema change is done three ways so new and existing restaurants match:

  1. Add supabase/tenant-migrations/NNNN_name.sql.
  2. Mirror it into supabase/tenant-template/schema.sql.
  3. Bump SCHEMA_VERSION in apps/api/src/provisioning.ts (currently 107).

Roll it out to every live restaurant:

cd apps/api
npx tsx scripts/apply-tenant-migration.ts ../../supabase/tenant-migrations/NNNN_name.sql

Never weaken RLS; new tables get policies in the same migration.

Deployment

Both apps deploy to Vercel from main:

Scheduled jobs (low-stock alerts and similar) call the API’s cron routes with CRON_SECRET. See VERCEL_DEPLOY.md for the web environment variables.

Test scripts

apps/api/scripts/ holds end-to-end checks against real tenant projects, for example rbac-test.cjs, portals-test.cjs, realtime-test.cjs, test-recipe-linking.ts, test-availability-engine.ts and test-performance-access.ts.

The finance suites run against every reachable restaurant in one go. Each database test is a single transaction that is rolled back, so no real data changes:

cd apps/api
bash scripts/run-finance-suite.sh

It covers the security hardening, ledger, expense workflow, invoice matching, day close and finance overview suites, a full fictional restaurant day (test-finance-scenario.ts) and cross-restaurant isolation against the live API (test-tenant-isolation.ts).

Independent branches:

Team member contributions

Member Role Contributions
Name Role What this member built or did
Name Role What this member built or did

Known limitations