Tiller Intelligence: the architecture a retail business runs on

The grounding framework

Cash, stock, hours and margin. That's the job before it's anything else.

A retail day comes down to a handful of numbers: what's in the drawer, what's on the shelf, what the day cost in hours, and what's left after both. That is the job at our counter, and it is where Tiller starts. Point of sale is the first framework because it's where those numbers get made.

A cashier hands a customer their receipt at a retail counter

Photo: Tessa Bury, Wikimedia Commons, licensed CC BY 4.0.

Cash. Counted twice, reconciled to the cent, variance on the screen before the drawer locks.
Stock. Days of cover tracked live, reordered before the shelf goes empty.
Hours. What the day cost in labor is the number we don't hold yet. It's the next one we want.
Margin. By brand and by category, live, not at month end.

The frameworks

Point of sale first. More plug in behind it.

Same records underneath, same guards, same login. Every framework you turn on runs on the one architecture, not a separate app bolted beside it.

Ticket #4,182 · Dez R.open

Cold brew, 16oz$4.75

Trail mix, family size$8.50

Member discount 15%−$1.99

Due$11.26

Drawer variance today$0.00

Retail counter

Point of sale

Sell, discount, reconcile the drawer to the cent. Built at a counter that does real volume, by the people standing at it.

Cash counted twice, variance on the screen
Every line the day's reporting needs, written as the sale happens
Margin by brand and by type, live
Works with the internet down

Live · our storeOpen framework

Batch AG-0114 · Week 3in band

Requirement coverage147 / 152

5 open2 unsigned

Pest scouting · Room 3overdue 2d

Water test · wellsigned

Cultivation record

Agricultural compliance (GACP)

Good Agricultural and Collection Practices, the broad agricultural standard for herbal and botanical crops, turned into daily work: who does a task, when it's due, who signed, and what's missing right now.

Every requirement mapped to a record you actually keep
Batch identity from planting to transfer
Unsigned and overdue tasks surfaced, not buried
Audit pack exported with the evidence attached

In buildOpen framework

Run 14 · Marcus T.en route

6 stops · 2 doneETA 3:42p

Stop 3 · 1412 RedondoID ok

Stop 4 · 88 Ocean Aveno answer

Manifest value on van$2,910

Cash carried$1,145 limit 3k

On the road

Delivery

Manifest, route, ID check, cash on the van and the drop signature, the whole run, on the driver's phone, reconciled when he walks back in.

Manifest value and cash limit watched live
ID and age check captured at the door
Route re-cut when a stop goes bad
Driver reconciles in under two minutes

Built, not switched onOpen framework

Your brand at our store

Units out, 7 days318

Sell-through74%

Days of cover9

Needs you

Compliance doc · Trail Mixexpired

House Blend Coffeelow, 2d

Invoice 2281paid

For the brands on your shelf

Brand portal

Your vendors log in and see their own sell-through, their own stock cover and their own expired paperwork, and fix it without calling your buyer.

Each brand sees only its own numbers
Expired paperwork flagged to the brand, not to you
Reorder before the shelf goes empty
Invoices and credits in one place

Not built yetOpen framework

Buy one, get everything

There's nothing you don't get.

You're not buying a bundle of separate apps. You're buying the architecture, and every framework we build plugs into the one you bought. The choosing isn't which ones you paid for. It's which ones you turn on.

No paying again to add a framework later.
No re-onboarding when you turn one on. Same login, same records, same guards.
One price for the architecture. What you run on top of it is up to you.

Locked and open

The parts that can put you out of business are not yours to change.

Every install starts from the same locked layer: records, signatures, retention, tenant boundary. You build on top of it. You never build through it. That is the whole reason it is safe to hand you the keys.

Locked

Requirements behind the guards

The record shape, the signature chain and the retention clock are written into the architecture and enforced by the guard layer. An extension that would break one is designed to be refused, with the reason, before it runs. That layer is being built now.

Read the guard rules

Locked

What we hold the line on

Signature chains, record retention, the tenant boundary, and every tax and compliance report the software owes your state. These don't move because someone finds them inconvenient. They move when the law does.

See what's locked

Open

What's yours to shape

Screens, alert rules, reports, integrations, workflow and wording. Anything that isn't one of the parts above is yours to build, break and rebuild without asking us first.

See what's open

Locked, we hold this line

no edits

Signature chain and record retention	architecture
Tenant boundary and access control	architecture
Tax reporting and calculation	law, your jurisdiction, one regime today
Cash drawer reconciliation	architecture

Open, build what you want

yours

New screens, forms and reports	your Claude
New readings and new sensors	your Claude
Your own alert rules and holds	you
Integrations into anything you run	you or us
Workflow, wording, crew tasks	you

You extend it

Your own AI. Your own data. Your own infrastructure.

Above the locked line, the architecture is yours to build on. New screens, new alerts, new reports, new connections into whatever you already run, described in plain language, checked by the guards, shipped when it clears them.

Yours

Your own Claude account

Your account, your keys, your data boundary. It reads your store, writes your extensions, and answers your crew in your own words. Your prompts and your records never cross into another tenant, and nothing about your business trains anything outside your walls.

How tenancy works

Open

Extend it yourself

New screens, new readings, new alerts, new reports, described in plain language, built against the same API we build on. No vendor ticket, no six-week roadmap slot, no charge per field. If it clears the guards, it ships.

See the extension API

The resident expert

Someone who knows your store is on shift every hour you're not.

Not a chatbot bolted onto a dashboard. It has your equipment history, your procedures, your crew and your rules, so it can walk a new hire through a job, and it knows when a job is past what it should handle alone.

It answers in your words

"The cooler in aisle 3 is running warm," not "unit 14, delta-T anomaly, sigma 2.4." Your crew's own words are the interface. That's a rule in the design, not a preference.

It walks people through it

New hire, first night alone, first time resetting the card reader: step, photo, confirm, next step. The walkthrough is written for your equipment, not a manual PDF.

It knows its own edge

When the evidence runs out, or the job is past what your crew should do unsupervised, it says so plainly and opens a service call. It does not guess to look useful.

Then you call us

A named engineer with your whole timeline already in front of them: every alert, every reading, every change you made. The call starts at the problem, not at "can you describe the issue."

What plugs in

It knows the cooler compressor is dying before the aisle does.

Your building already produces the evidence: temperature, amp draw, run time, valve position, humidity, power. Nobody reads all of it every night. The resident expert in your workspace does, and it tells you what it found in the same words you'd use.

Today, store systems

Example store · 4 systems · 61 connected points

1 act today2 watch58 normal

The walk-in cooler compressor is running toward a failure. Get a tech on it today.

Cabinet temperature has held 2.1°F above setpoint for nine consecutive nights while the compressor drew 18% more current than its own four-week average. That is the same pattern the old unit made for eleven nights before it seized in March. Perishables are on that line today, so a dead unit here costs the inventory, not just a weekend.

First seen 9 nights ago · confidence high, matched against your own failure · maintenance log attached

Open work order
Call Tiller

Backroom humidity started climbing the same night the cooler drifted. These are one problem.

You added the backroom humidity sensor six weeks ago. Since then the overnight reading has tracked the compressor's temperature almost exactly, up to 68% from a 40% baseline. Read together, the door seal is failing and the unit is working harder to hold temperature. Check the seal first; if the draw doesn't come back down, the compressor is the problem and not the seal.

New sensor, folded into reasoning automatically · no rule was written for this · 6 weeks of history

See the two lines

Zone 4 is reading 3°F warmer than the rest of the floor.

Nine days of drift, all in one direction, only in zone 4. The other three zones are flat, so this is the zone's damper, not the rooftop unit. Nobody noticed, because the floor still feels fine near the registers.

From your building system · last calibration 71 days ago · maintenance log

Schedule calibration

Tomorrow's price-change batch is held until zone 4 is back in range.

Your own rule: no batch job pushed from a register in a zone reading out of setpoint, so a bad thermostat doesn't also corrupt the update. It's held, not cancelled, and the crew has been told why on their tablets. Release it yourself once the tech signs off, or it releases on its own when the zone reads normal for 24 hours.

Rule you wrote on 14 Aug · applies to all four zones · held 08:10

Review hold

Everything else read normal overnight. Nothing needs you in the other three zones.

58 points checked between 20:00 and 06:00, all inside their own learned band. The empty version of this screen is the good one.

Quiet night · 06:00 sweep complete

See all points

What you plug in

Building management, HVAC, refrigeration, power meters, environmental sensors, scales, door access, cameras. Read-only until you say otherwise.

What it learns

Your store's normal, not a generic one. Nine nights of a thing is only meaningful against your own four weeks, and against your own last failure, which is a comparison no vendor dashboard can make for you.

What it never does

It does not turn your equipment off, change a price, or sign a record for you. It tells you, it holds work when your own rule says hold, and it waits for a person. Acting on it is a decision you make.

What is actually running

Numbers we can show you the source of.

Ticket history migrated

62

months of ticket lines rebuilt off the point-of-sale system our own store ran before, line by line, with 48,910 customers carried across.

Legacy system, being replaced

1

The old point of sale didn't get an app bolted onto it. We're replacing it at our own store, register by register, with the store open, and we're not finished.

Owner on the hook

1

The person who writes this software runs a store on it. If it's wrong, he loses the money first.

The service call. When the resident expert in your workspace says "this one needs a human," it means this: a named engineer, your site, your problem, finished. No ticket queue. If you've got a problem that no one else can help with, and if you can find them.

How it starts

Four steps, not a rollout plan.

01

Stand up the architecture

Locked requirements loaded, your licence or permit type, your locations and your people set up. Day one the records and the retention are already right. What your permit type needs on top of that, we load with you.

02

Connect what you already run

Point of sale, building systems, sensors, cameras, accounting, whatever you're running today. Read-only first, write once you trust it.

03

Build on top

Ask your Claude for the screen or the alert you want. It writes it, the guards check it, you approve it, it goes live in your workspace.

04

Call us when you're stuck

Something you can't figure out, or shouldn't be doing yourself. One number, a named engineer, and a service call that ends with it working.

Pricing

Three ways to get going. Numbers not set yet.

The shape is decided; the figures aren't. We'd rather show you an honest blank than a number we'd have to take back.

Placeholder: figures pending owner ruling

Self-serve

Run the architecture

—

per site, per month · to be set

Locked requirements and guard layer
Your own Claude account
Every framework, turned on as you need it
Record retention and audit export
Email support, next business day

Talk it through

Where we'd start you

Guided build

We help you build it

—