Manage datasets

Once a dataset has data in it, the work shifts from adding rows to keeping the table itself in order: finding the right dataset, making a copy, pulling data out, fixing a column, or getting rid of what you no longer need. This guide covers all of that on an existing dataset. For putting data in, see Create a dataset, Add rows, and Add columns.

Find a dataset in the list

• Only Dataset Name and Datapoints are sortable: click either header to reorder the list by it
• Use the search box above the table to filter by name
• If nothing matches, the list shows No datasets found instead of an empty table

The list is paginated; see Limits & Data Types for the page size.

Act on several datasets at once

Select one or more datasets with the row checkboxes and a bulk action bar takes over the toolbar, showing {n} Selected alongside Delete and Cancel. Select exactly one dataset and Duplicate joins the bar; select two or more and Duplicate is replaced by Compare, which opens a Select Base Columns drawer where you pick the one column the selected datasets share. Cancel clears the selection without doing anything.

Duplicate a dataset

Check the dataset’s row checkbox, then click Duplicate in the bulk action bar to open the Duplicate Dataset dialog. Enter a name for the copy in Enter Dataset Name, for example support-agent-eval-copy, then click Create. The dialog validates the name before it lets you proceed, and a success toast confirms once the copy exists. Click Cancel to back out without duplicating anything.

The duplicate is a separate dataset from the moment it’s created: editing it doesn’t touch the original, and deleting one doesn’t touch the other. It isn’t a full copy, though: duplicating only carries over rows and static columns. Dynamic columns, and the computed values in them, don’t come across, so a duplicated dataset can have fewer columns than the one it was made from.

Export a dataset

Click the download icon in the dataset’s toolbar to export it. A Download has been started… toast appears immediately, followed by Dataset downloaded successfully once the file is ready.

Edit data in the grid

Inside a dataset, the grid supports these changes directly, without leaving the Data tab. Renaming a column, changing its data type, and deleting it all live in the column header menu, under Edit Column Name, Edit Column Type, and Delete Column.

Action	What it does
Rename a column (response_quality to quality_score, for example)	The cell values stay the same, but SDK and cURL calls that reference the old name break
Change a column’s data type	Reinterprets how the column’s stored values are treated
Edit a cell in a static column	Overwrites that one cell’s value; audio and persona cells can’t be edited this way
Delete a column	Removes the column and every cell in it

To delete rows, select them with their row checkboxes and confirm the delete action in the grid toolbar.

Cells in a dynamic column can’t be edited at all: the column is managed by whatever produces it.

Delete a dataset

Select one or more datasets with the row checkboxes, then click Delete in the bulk action bar. The dialog title switches between Delete Dataset and Delete Datasets depending on how many you selected. Confirm with Delete, or back out with Cancel. A success toast confirms once it’s done.

Bulk delete is capped at 50 datasets per request. Deleting more than that means running the action in batches.

What deleting actually removes

Deletes here are final: once you delete a row, a column, or a dataset, it’s gone. Deleting rows removes the selected rows and every cell in them. Deleting a column removes the column and every cell in it.

Deleting a dataset also deletes its experiments.

Dive deeper