Import & sources

Datasets are tables. Every one is versioned, carries sources — citations for where the data came from — and describes its columns, so any number in the workspace can be traced back to a document and read correctly by a teammate or an agent.

The import wizard

Datasets → Import data walks through the steps:

1. Upload a CSV, JSON or Parquet file (gzipped works too).
2. Preview how it parses. For CSVs, adjust the delimiter, header row and rows to skip until it looks right. Nothing lands until you confirm.
3. Columns: give each column its semantic type, unit, scale and a description. This is what makes a column called amt legible as "funding amount, EUR millions".
4. Sources: cite where the data came from — a URL, a report, a file — and any judgement applied while preparing it.
5. Import. The file becomes a dataset backed by an immutable, versioned table.

The wizard can also target an existing dataset: append rows, or replace with a new version. History is kept.

The dataset page

• Data shows the current rows, with numbers formatted from each column's unit.
• Columns lists the columns; their names, units and descriptions are editable here.
• Sources lists the citations; add or edit them.
• The About rail shows what reads this dataset — transformations, charts, spreadsheet connections, views — and the Docs tab holds notes agents read before using it.

Data arrives clean: aqmen is not a cleaning tool. Prepare the file so it is query-ready, then import.

Why citations matter

A table of numbers with no origin cannot be defended in a meeting, and neither can an agent reasoning from it. Datasets that agents load arrive with sources attached; datasets you import deserve the same care.

Faster with an agent — ask the agent to "research <topic>, clean the results, and load them as a dataset with sources." It gathers the data, normalizes it, uploads it, documents the columns and cites every origin.