The Hub

Your numbers, in one place, as of four minutes ago.

Amazon, your website, dealers, distributors, the ERP, QuickBooks, the freight forwarder's emails — combined into one live view. Not a report someone runs on Monday. A system that's already looking, and that tells you when something needs you.

The Problem

Nobody in your company can see the whole picture

Not because your people aren't good. Because the picture is scattered across five systems that were never designed to talk, and assembling it by hand takes most of a morning — by which point it's already wrong.

Inventory truth and money truth are different systems

Your ERP knows units. QuickBooks knows dollars. Neither knows both, so margin per SKU is a spreadsheet someone rebuilds quarterly and stops trusting by February.

Units on the water don't exist anywhere

Three containers are in transit. Your reorder report doesn't subtract them, so you either order the same SKU twice or sit on a stockout waiting for something already 19 days from the port.

The critical view lives on one laptop

Exclusive projects, custom runs, dealer commitments — tracked in one spreadsheet by one person. It works until they're on vacation, and it's invisible to everyone else always.

You find out after it costs you

The Buy Box was lost Monday. You noticed Thursday. That's three days at several hundred dollars a day, and nothing in your stack was watching.

Reports describe, they don't decide

A sell-through report tells you what happened. It doesn't tell you to order 600 units of one SKU from Vendor A this week and cancel a line on an inbound PO. That translation is a person's job, weekly, forever.

Cost sheets drift and nobody notices

Freight moved. A vendor raised prices 7%. Duty was allocated by unit instead of value. You're quoting dealers off numbers that were true last year.

Role-Based Seats

One dataset, four seats

Your CEO, your purchasing manager, and your bookkeeper need different numbers off the same data. Most dashboards pick one and disappoint the other two — which is why most dashboards get opened twice and abandoned. Each seat gets its own KPIs, its own alerts, and its own definition of urgent.

SEAT 01

Owner / CEO

Revenue, real margin, cash tied up in inventory, and a short list of things that need a decision this week. Deliberately sparse — if it's on this screen it's because you have to act on it.

TYPICAL ALERT: Stockout in 8 days on your #1 SKU · $18.4K
SEAT 02

Purchasing

What to order today, how many, from which vendor — net of units already in transit and each vendor's real lead time. Plus what to stop ordering, which nobody's report tells them.

TYPICAL ALERT: Combine two POs · saves $2,100 freight
SEAT 03

Accounting

AR by age matched to order history, AP due, and true landed cost per SKU recomputed on every receipt — so the cost sheet stops quietly lying about your margin.

TYPICAL ALERT: Dealer at 61 days just placed order #3
SEAT 04

Custom & Limited

Exclusive and limited projects modeled as the two-stage chain they really are: resin PO, production PO, sales order, promised ship month. Most ERPs cannot represent this at all.

TYPICAL ALERT: 6 projects can't hit their promised month
Seats are configuration, not separate products. Adding a fifth seat for your warehouse manager or your sales lead is an afternoon, not a project. The data is already there — a seat is a decision about what matters to whoever is looking.
The AI Layer

Ask it the question you'd ask a good analyst

Not a search box, and not a chatbot that invents numbers. It reads the same live data the dashboard reads, does the arithmetic, and shows its work. These are real questions from real operators — you can run all of them on the prototype.

"What should I be worried about this week?"
Comes back with a ranked list carrying dollar figures, not a summary of what happened. Stockout in 8 days on the SKU that carries your best margin, and the inbound PO lands 7 days too late — roughly $18,400. Then the Buy Box you lost, then the dead stock. Ranked by what it costs you, not by what's most recent.
"Can I afford the Q4 buy?"
Models the full buy against open POs, dealer AR including what's aged past 45 days, and 90-day velocity. Doesn't answer yes or no — proposes a split: place $180,000 now on everything under 40 days of stock, hold the rest until the inbound container clears. That keeps you above your cash floor without delaying a single fast mover.
"Which channel is actually making me money?"
Revenue order and margin order are almost never the same, and that gap is the answer. Amazon is usually the biggest line and the thinnest one. The hub nets out channel fees, freight, and true landed cost, so you can see that a smaller channel is earning twice as much per dollar — and put the next marketing dollar there.
"What am I over-ordering?"
The question nobody asks because no report answers it. Surfaces SKUs carrying 180+ days of coverage that are still on the reorder schedule, including inbound lines you can still cancel. Typically frees $20K–$60K in cash on the first run without touching service levels.
"Did any vendor prices change?"
Compares each receipt against your last six from that vendor. A 6.8% increase that nobody booked means you've been quoting dealers off a stale landed cost for months — and it tells you how many margin points that's actually costing, per SKU, in dollars.
Alerts

The part that earns its keep while you're asleep

A dashboard you have to remember to open is a dashboard you'll stop opening. The alerts are the actual product — the hub is where you go when an alert makes you curious.

Thresholds are yours

Your definition of "low," your lead times, your definition of dead stock. Alerts are configuration, not code — so they change the same day you change your mind, not next sprint.

Routed to the right person

Purchasing alerts go to purchasing. Cash alerts go to you. Nobody gets a firehose, because an alert everyone ignores is worse than no alert.

Delivered where you already are

Email, SMS, WhatsApp, or Slack. Most owners take the Monday 7am summary on their phone and nothing else until something's actually wrong.

Every alert carries a number

Not "inventory is low." "8 days of stock, 17/day, next PO lands 7 days late, roughly $18,400." An alert without a dollar figure isn't actionable, it's anxiety.

And a recommended action

Alerts propose the fix: order this quantity from this vendor, cancel this inbound line, hold this dealer's next shipment. You approve or you don't.

They shut up when resolved

Paid invoice stops the chaser mid-sequence. Restocked SKU drops off the list. Nothing nags about something you already handled.

Connections

What plugs in

One connector per source, each normalizing into the same internal shape. That's why a metric written once works whether the order came from Amazon or a dealer's WhatsApp message.

SourceWhat we pullRefresh
ShopifyOrders, products, inventory, customers, fulfillmentNear real time
Amazon SP-APIOrders, FBA inventory, fees, Buy Box status, listingsHourly
QuickBooksInvoices, AR aging, bills, AP, distributor revenueHourly
OrderTime / your ERPInventory, POs, receipts, vendors, sales ordersContinuous
Google SheetsAnything your team maintains by hand — and there's always somethingOn change
CSV / manual uploadLegacy exports and systems with no API worth usingOn upload
Freight & customs emailForwarder and carrier updates parsed into structured ETAsOn arrival
Twilio / WhatsAppDealer messages in, alerts and confirmations outReal time
Something not on this list? If it has an API, a CSV export, or arrives as a predictable email, it can be connected. The manual connector exists precisely because every importer has one critical thing that only lives in somebody's inbox.
Under the Hood

Why the second client is faster than the first

The hub isn't rebuilt per company. There's one canonical data model in the middle and a set of engines around it — adapters, metrics, rules, UI, AI — that don't know or care who the client is. Everything specific to you lives in a single configuration file. That's an architecture decision that pays you back as a shorter timeline and a lower price.

01

Adapters

One connector per source. Each one's only job is turning that system's peculiar output into the canonical shape. Add a source later and nothing downstream changes.

02

Canonical schema

Every source normalizes into one model of orders, products, inventory, POs, and money. Write a metric once; it works across every channel automatically.

03

Metrics engine

Velocity, days of stock, coverage, true landed margin, sell-through — computed the same way for everyone, so the definitions don't drift between screens.

04

Rules engine

Your thresholds and lead times as configuration. Alerts change same-day, and you can change them yourself once you've seen how it works.

05

Seat layer

Which KPIs, which panels, which alerts, for which role. A new seat is a config entry, not a build.

06

AI layer

Reads the canonical data and the computed metrics — never raw guesses. It can only answer from numbers that actually exist, which is why it doesn't hallucinate your revenue.

Claude APIReactGoogle Apps ScriptMake.comNetlifyTwilioChart.js
Questions

Before you ask

How long does a hub take?
Six to ten weeks depending on how many sources and how strange they are. You see your own real data flowing by the end of week one — not a mockup, your numbers. Everything after that is adding seats, metrics, and rules on top of a working foundation.
Does this replace my ERP?
No. It reads from your ERP and everything else, and writes back where that's useful. If we thought your ERP needed replacing we'd tell you and refer you, because that's a different project and we'd do it badly.
Where does my data actually live?
In your accounts, on your infrastructure. We build it there and hand over the keys. If you stop working with us the system keeps running and any competent developer can maintain it — the stack is deliberately boring for exactly this reason.
Can the AI make things up?
It answers only from the canonical data and precomputed metrics, and it shows the numbers it used. It can't invent a revenue figure because it isn't generating figures — it's reading them and explaining what they mean. Where it genuinely doesn't know, it says so rather than guessing.
What if my data is a mess?
It is, and that's normal. The audit surfaces it before you commit. Some cleanup is included; the rest we tell you about honestly rather than discovering it in week four. Occasionally the right answer is "fix this one thing first, then call us," and we'll say that.
Should I start with the hub or an agent?
An agent, usually. It's live in two to three weeks, it pays for itself, and it means the hub decision gets made after you've watched us work rather than before. Clients who start with the hub are typically ones who've already tried to build something like it internally.
What does it cost to run?
Third-party costs are small and yours: AI usage typically $20–$150/month by volume, hosting free to $20/month, messaging around a cent per message if you use it. Set up in your accounts under your billing.
Start Here

See it on your own data

The fastest way to know whether this is worth it is a 45-minute audit where we map your actual systems. You leave with a written plan and a fixed number, whether or not you build with us.

Book a 15-minute call hello@charidan.com · Los Angeles / Tel Aviv