Dashboards that keep running: quick to build, cheap to use
Bram worked out with an agent where margin is leaking. The answer is a chart per product group. Tomorrow he wants the same chart with tomorrow's figures. A picture with the numbers baked in does not help him.
In Plainwater AI an analysis ends in a dashboard that keeps running on the source. The model is needed to build and change it, not to look at it. That makes reporting quick, simple and cheap.
Updated .
What is a dashboard in Plainwater AI?
A dashboard is a page the agent builds that pulls its data from named datasets on the source. It refreshes by itself, without running an AI model again.
Bram asks for the margin dashboard in a conversation. The agent writes the page and the datasets, shows a preview and works in Bram's comments. The dashboard then lives in a folder, and every morning it is right again.
The agent sees its own rendering errors and fixes them, so Bram never has to retype an error message.
Why are datasets objects of their own?
A dataset is a reusable object with one definition, so 'revenue per hour' exists in exactly one place. Several dashboards point to the same dataset.
There are two kinds. A query dataset is one query on one source. A derived dataset combines other datasets with DuckDB, for example margin from finance divided by hours from planning.
Datasets refresh on a schedule, and results are stored encrypted. When a definition changes, it applies at once to every dashboard that uses it.
| Kind | What it is | Example |
|---|---|---|
| Query dataset | One query on one source, with a refresh schedule | Revenue per product group per day |
| Derived dataset | One SELECT over other datasets | Margin per hour worked |
Why does a dashboard cost almost no AI budget?
Because no AI model runs when it refreshes or when someone views it. The model only costs money when building and changing it, which is a few conversations.
A report a hundred colleagues open every day therefore costs the same in model usage as one opened by a single person. The source is queried on the dataset's schedule, not per viewer.
What may a viewer see?
Each source has a separate mode for people: off, aggregates or rows. It is independent of what an AI model may see, because a colleague viewing their own figures is a different risk.
The browser only knows dataset names. Dashboard code can never read more than what was approved at save time, even if someone edits the code later.
Dashboards live in folders with permissions per folder: view, edit or manage. Purchasing sees its own folder, finance sees its own.
Can you ask questions about a dashboard?
Yes. Right-click a chart to open a read-only conversation. Bram asks why margin dropped in week 41, and the agent investigates without changing the dashboard.
If Bram then wants an extra chart, that becomes a change in the dashboard conversation. Reading and changing stay separate.
Where it stops
Dashboards do not replace a full BI platform with complex semantic models. For Power BI the agent works in your existing models through the connector.
A dataset refreshes on a schedule, not live on every click. Figures are as current as the last refresh.
A dashboard shows what the dataset defines. If the definition of margin is wrong, so is the dashboard; that is what the conversation with the agent is for.
Frequently asked questions
Does an AI model run when someone opens a dashboard?
No. A dashboard reads its datasets; a model is only needed to build or change it.
Can I reuse datasets in several dashboards?
Yes. A dataset is an object of its own with one definition that several dashboards point to. A derived dataset combines other datasets.
How current are the figures?
As current as the dataset's last refresh. You set the refresh schedule per dataset.
Who can see a dashboard?
Whoever has permission on its folder: view, edit or manage. On top of that, the source decides whether people see aggregates or rows.
Does this work alongside Power BI?
Yes. Through the Power BI connector the agent can also work in existing semantic models; dashboards in Plainwater AI are a fast route for new questions.