See it work on a real business,
before you connect your own.
Four scenarios ship with the product — a distribution operation, a delivery fleet, a factory floor, a collision repair chain. Each one is a real database with real messiness in it. Load one in a click, paste a question, and watch what the companion actually does with it.
Three steps
Open Tutorials
In the client, open Tutorials and pick a scenario. It creates a project, loads the dataset, and connects it as a data source for you.
Ask it something
Copy a question below, or write your own in plain English. The companion reads the schema first, then writes and runs the query itself.
Push on the answer
Ask why. Ask for a chart. Ask it to draft the email. The interesting part is the second question, not the first.
Demo data is read-only and lives in its own project. Deleting the project removes it. Nothing here touches any data source of your own — you do not need one to try this.
One question, start to finish.
From the Operations scenario, against the 15 support cases and 25 orders that ship with the dataset. The figures are the dataset's own — check them yourself once it's loaded.
Which support cases are breaching SLA right now? Show the customer, how long it has been open, and what they have spent with us.
Note what is not in that sentence: no table names, no join keys, no SQL. The answer needs three tables — support cases, customers, orders — and the companion works that out from the schema.
8 cases are still open. 3 of them are past their SLA.
| Customer | Open | Priority | Category |
|---|---|---|---|
| Yuki Tanaka | 217 days | medium | Access & Billing |
| Liam Walker | 215 days | high | Billing & Payments |
| Elena Vasquez | 213 days | medium | Access & Billing |
Each joined to what that customer has spent, so you can see which neglected case belongs to your largest account. Note the pattern the list makes on its own: every breach is a billing or access problem.
Then ask "draft an apology to the top three, offer a 10% credit" and you get three drafts, in your own voice, ready to send.
Four businesses to borrow.
Each ships with its own operating documents too — dispatch procedures, repair guides, quality manuals — so you can ask questions that need both the numbers and the policy.
Operations
Start hereA small distribution business: 13 customers, 25 orders, 15 support cases, live inventory. The broadest scenario and the easiest to recognise.
What were our best and worst revenue days this quarter, and what made the difference?
Which products are low on stock but still selling? Rank by how soon we run out.
Logistics & fleet
11 tables25 deliveries, 20 drivers, 18 vehicles, plus routes, fuel, GPS events, maintenance logs and customer complaints. Good for "why did this go wrong".
Rank every route by on-time delivery rate, worst first. Flag anything under 80% as AT RISK.
Find the day with the most delayed deliveries. What happened — driver, vehicle, route, or weather?
Manufacturing
10 tables10 machines with telemetry, downtime events, work orders, quality control, defect reports and shift notes. The one where the numbers and the operator's comment disagree.
Show plant-wide machine health: status, downtime over the last 30 days, open work orders and alerts. Flag anything down or over two hours lost.
Which machine is producing the most defects, and does the maintenance history explain it?
Collision repair
12 tables20 repair orders across multiple shops — claims, carriers, parts, paint, calibration, rework and technicians. The most interconnected of the four.
Compare cycle time by shop. Where are jobs stuck — parts hold, supplement, or paint?
Which jobs came back for rework, and is any technician or carrier over-represented?
The questions that separate a tool from a coworker.
Anything can return a row count. Try these once a scenario is loaded.
Build me a dashboard of this, one I can come back to on Monday.
You get a working page, saved to the project — not a screenshot.
Remember that anything over 48 hours is an escalation for us.
Say it once. Open a new thread tomorrow and it still knows.
What's our customer churn rate and net promoter score?
Neither is in the data. It should say so rather than invent a number — worth checking on any tool you're evaluating.
Borrow a business for ten minutes.
Then connect your own and ask it the question you actually came with.
Start freeFree plan, no card. Bring your own OpenAI or Anthropic key and nothing is metered.