How it works

What actually happens between your upload and your dashboard

No magic, and deliberately no shortcuts. Here is each stage, what the agent does, and where you decide.

Step 1

The file arrives

Upload from your desktop or let a connected source bring it in. Every sheet becomes a table, and every row remembers which file it came from — which is what makes a bad import removable later.

  • Excel and CSV up to a gigabyte, streamed rather than held in memory.
  • PDFs and Word documents are kept as reference material for the agent.
  • Connected folders are walked on a schedule and new files picked up on their own.

Step 2

The agent reads it

It samples the headings and column types, profiles the values, and works out how your files relate to one another. Then it proposes a dataset — and stops.

  • It names the grain: one row per receipt, per store per day, per invoice line.
  • It states its assumptions, including the ones you will disagree with.
  • It never writes the SQL your dashboard will run. That is the engine's job.

Step 3

You approve it

This is the point of the whole design. A proposal you have not approved loads nothing and builds nothing. Approve it, change it, or send it back with what it got wrong.

  • Approving builds a real table with named measures and quality checks.
  • Approve fifteen proposals at once and get one dashboard across all of them.
  • Datasets are named with their ids, because two can legitimately share a name.

Step 4

The dashboard gets built

Widgets are declarative specs the engine executes, so results are reproducible. Filters reach across linked datasets, and clicking a bar cross-filters everything connected to it.

  • A warning appears when a filter reaches only some of the datasets on a dashboard.
  • Each dataset is aggregated before joining, so one can never multiply another.
  • Ask the dashboard analyst why a figure moved, at any time.

Step 5

You publish it

Publishing copies the draft to a frozen layout at its own address over live data. Until then it is yours alone, and later edits wait as unpublished changes.

  • Grant the whole report list, or one named report.
  • Schedule it daily, weekly or monthly, with its own filters.
  • Unpublish removes the report without deleting your work.

Step 6

Next month happens

Drop the new export on the dataset page. Its sheets are matched to the dataset by their column headings, and every dataset and report downstream rebuilds.

  • Add rows for a monthly export, or replace all rows for a full snapshot.
  • Re-adding the same file name replaces its rows instead of doubling them.
  • Files can be removed again, and everything built on them rebuilds.

Try it with one file

The whole loop, from upload to published report, fits in an afternoon.

Start freeTalk to us