Dataset pages

A dataset is the table a pipeline publishes, or one you uploaded or ingested yourself. Its page shows one version of the data at a time: a header describing where the rows came from, a rail of every version on the left, a paged grid, and a strip at the bottom listing every pipeline, integration and dashboard connected to it.

Datasets are listed on the Pipelines page in two sections under the pipelines: Standalone datasets (uploads, imports and sample data that no pipeline produces) and Connected datasets (the output of a pipeline, collapsed by default). The Show filter narrows the page to Everything, Pipelines, Standalone datasets or Connected datasets. Each dataset card carries an icon for where its data came from: the producing pipeline’s own icon, or a file, sample, ingest, Jason or copy icon.

Open a dataset from there, from a pipeline’s Latest outputs, from a run’s Outputs tab, or from search (Cmd/Ctrl+K).

The header

The header has four parts, left to right and top to bottom.

• Breadcrumbs. Home, Pipelines, the producing pipeline (when there is one), the dataset name, and the version you are viewing (v7). The dataset name is a dropdown that lists every dataset in the workspace, so you can jump between them without going back to the list.
• Title. Click the name to rename the dataset. Enter saves, Escape cancels.
• Badge. A pill next to the title says where the data comes from. See the table below.
• Subtitle. The row count, what each row is, and who wrote the version you are viewing. See below.

Which version you are viewing is shown in the versions rail on the left. See Versions.

What the badges mean

Badge	Meaning
Live	A pipeline produces this dataset. It keeps this badge whatever wrote its latest version, including an upload or an edit.
Pipeline deleted	A pipeline produced the data, but that pipeline no longer exists. The rows are still here; nothing will refresh them.
Sample data	Example rows generated from a description, with Generate data in {worksheet}… or when a workspace is set up. They do not come from any real source.
Uploaded	The data was imported from a file.
Ingested	Rows arrive through the ingest API or an input integration.
Jason	Jason imported, transformed or published the data in a conversation. Hover the badge for which.
Copied	The data was copied from another workspace.

Subtitle and the “collects” label

Under the title the page shows the row count, what each row represents, and where the version you are viewing came from, for example “8,426 products, generated by Price & promotion tracking on Sep 23, 14:02 · used as input by Change detection”.

• The count and noun. “8,426 products”, “340 reviews”. The noun is the dataset’s collects label. Click it to change it (one word, letters only, up to 48 characters; leave it blank to fall back to “rows”). The label is used everywhere the dataset is counted: the pipeline page, the home page, export toasts, the filter box, the rows-per-page menu.
• Who wrote it, and when. “generated by {pipeline}” with the pipeline’s icon and a link to it, “regenerated by” for a rerun, “generated by a deleted pipeline”, “uploaded from products.csv”, “synced”, “generated as sample data”, “imported by Jason”, “transformed by Jason” or “copied”, followed by the date and time.
• Used as input by. When another pipeline reads this dataset as an input, a link to that pipeline.

Actions

The header has up to two buttons and a menu.

• Refresh. Only on datasets a pipeline produces. It opens the producing pipeline’s run dialog, the same one as Run on the pipeline page, so you can check inputs and parameters before you start. You stay on the dataset page while it runs: a banner reads “{pipeline} run abc123 is active” and links to the run. When the run completes, the page reloads on the new version. If it fails, a message says “{pipeline} failed to refresh this dataset” with a View run button. See Running a pipeline.
• Run query. The primary button. It opens the read-only SQL editor described in Query and export.

The menu beside Run query holds everything else, in this order:

Section	Item	What it does
Import	Upload to {worksheet}…	Replace or append rows from a CSV, XLSX, XLS or JSON file. See Upload and append.
Import	Generate data in {worksheet}…	Replace the worksheet with sample rows written from a description. Useful for trying a dashboard before real data exists.
Export	Export Excel workbook	Every worksheet of the current version as one .xlsx.
Export	Export current table as CSV	The worksheet you are viewing as .csv.
Worksheet	New worksheet…	Add an empty worksheet to the dataset.
Worksheet	{worksheet} settings…	Column visibility, order and formats, and the collects label. See Table settings.
Worksheet	Rename {worksheet}…	Rename the worksheet. Old links keep working.
Worksheet	Delete {worksheet}	Remove the worksheet. Disabled when it is the only one.
Dataset	Delete dataset	Delete the dataset and all its versions. Your browser asks “Delete dataset “{name}”? This cannot be undone.” first.

Deleting a dataset that is still used by a pipeline input, a dashboard or an alert rule is refused with the message “This dataset is still referenced by other objects (pipelines, dashboards, alerts). Remove or repoint those first.” Detach it from those first, then delete.

The grid

The grid shows one worksheet of one version, fifty rows at a time. When a dataset has more than one worksheet, a tab strip above the grid lets you switch between them; switching resets the page, sort and filter.

The bottom strip

Under the grid, one strip holds the connections on the left (see The connections strip), paging in the middle, and the view toggle and filter on the right.

Option	Values	Default	What it does
Previous page / Next page			Step through pages. The strip reads “Page 3 of 24”.
Rows per page	25, 50 or 100, labelled with the collects noun (“25 products”)	50	How many rows each page shows. Changing it returns you to page 1.
Table view / Card view	Table, Cards	Table	How rows are laid out. See Card view.
Filter	Free text	Empty	See Filtering and search.

The page, rows per page, sort and filter are kept in the address, so a link you send opens on the same view.

Card view

Card view lays each row out as a card instead of a table row: the row’s identifier at the top in small type, then its first column in bold and its second column underneath. Click a card to open the row. Paging and the filter work as in the table.

Card view has no column headers, so there is no sorting or column resizing, and cards cannot be selected. Switch back to Table view to select rows.

The choice is saved on the dataset, not in your browser. Everyone who opens the dataset sees the view you picked, and the table is the default for new datasets.

Sorting

Click a column header to sort ascending; click it again to flip. Only one column sorts at a time, empty values always sort last, and sorting returns you to page 1. The default sort is the first column, ascending.

Filtering and search

The filter box at the bottom right (placeholder “Filter products…”) searches every column of the worksheet, including hidden ones, for the text you type. It is a contains-match, not case sensitive, applied after a short pause while you type. The box is open when the page loads. Escape or its cross collapses it to an icon; a purple dot on the collapsed icon means a filter is still applied.

A second filter, source, narrows the grid to rows collected from one website. It appears as a chip with the site’s name when you arrive from a link that carries it (for example from a pipeline’s Sources tab). Click the chip’s cross to clear it.

There are no per-column operators such as “price greater than 100”. For anything beyond a text match, use Run query. See Query and export.

Selecting rows

Hover a row and its number becomes a checkbox. The header checkbox selects the current page. A bar appears with the count and the actions available:

• Rerun {n} rows. Shown only when another pipeline uses this dataset as an input. Sends just those rows through that pipeline as a manual run and opens the run.
• Delete {n} rows. Removes the rows in a new version after a confirmation. You cannot delete every row this way.
• Clear selection.

The selection is also visible to Jason in the sidebar, so you can select a handful of rows and ask “why are these missing a brand?” or “fix the category on these”.

How cells are rendered

Cells are drawn according to the column’s format (auto-detected or set in table settings) and the shape of the value.

Value	Shown as
Empty	An em dash.
A web address	A link pill with the site’s favicon, host and path. Opens in a new tab; hover for the full address.
An image address (column named image, photo, logo, thumbnail and similar, or ending in .png/.jpg)	A thumbnail. Click for a full-size view with a link to the original.
A video address (column named video, clip, recording, or ending in .mp4/.webm/.mov)	A play button. Click to play in place.
A column named source	A favicon pill for the site.
True / false	A green tick or a red cross.
A number	Formatted with thousands separators. Currency formats add the symbol and two decimals; percentage adds one decimal and %.
A column whose name ends in id, ref or fingerprint	Fixed-width, unformatted, so identifiers are easy to compare.
A list or nested record	A purple count badge and a short teaser; hover for the full structure. Lists of images show up to three thumbnails.
Long text	Cut at 40 characters; hover for the whole value.

Column headers carry a small type icon: a dollar sign for currency, a percent sign, a hash for numbers, a calendar for dates, a globe for links, an image icon, braces for nested data, or a “T” for text.

Column width and order

Drag a header’s right edge to resize it; double-click the edge to reset. Widths are remembered in your browser. Drag a header left or right to reorder columns; the new order is saved for everyone.

Click any row to open it in the row dialog. See Row detail and provenance.

Table settings

Open {worksheet} settings… from the header menu.

Option	Values	Default	What it does
Collects	One word, letters only, up to 48 characters	Blank (uses “rows”, or a label detected from the data)	What each row represents. The link Use automatic label clears an override.
Column checkbox	Checked / unchecked	Checked	Whether the column is shown in the grid and on share links. Hidden columns are still searched and exported. Show all and Hide all act on every column.
Column order	Drag the handle	Order from the pipeline or file	The order columns appear in the grid and in exports.
Format	Auto, USD ($), EUR (€), GBP (£), JPY (¥), INR (₹), Percentage (%), URL	Auto	How the column is rendered. “Auto” shows what was detected, for example “Auto (USD)”.

Save applies the settings to this worksheet only. Formats you set here override detection everywhere the dataset is shown, including share links.

Choosing URL as a format is not saved today; links are detected automatically from the value, so leave those columns on Auto.

Upload and append

Open Upload to {worksheet}… from the header menu.

1. Drop a file or click to browse. CSV, XLSX, XLS or JSON, up to 100 MB. The file’s first row (or the JSON keys) becomes the columns; there is no mapping step, so match the existing column names when you append.
2. Choose a mode. Replace worksheet (default) makes a new version whose worksheet contains only the file’s rows; other worksheets are carried forward unchanged. Append to worksheet adds the file’s rows to the existing ones.
3. Upload. The new version becomes active immediately. The toast reads “Uploaded 340 rows (v4)” or “Appended 20 rows (v4)”.

Option	Values	Default	What it does
Mode	Replace worksheet, Append to worksheet	Replace worksheet	Whether the file replaces or extends the worksheet.

Key columns and append. If the dataset has key columns (a pipeline sets these, for example product_id), appending a file whose rows carry those columns updates matching rows in place instead of duplicating them. Without key columns, append is a plain add.

Multi-sheet workbooks. Replacing a dataset with a workbook that has several sheets creates one worksheet per sheet, named after the sheets. Replacing or appending a single worksheet needs a single-sheet file.

Because every upload is a new version, the previous rows are never lost. Uploading also triggers everything that listens for new versions: alert rules are evaluated (including rules that send rows to an output), and dashboards read the new rows the next time they load.

A pipeline-produced dataset can be uploaded to as well, but the next run publishes a new version from the pipeline. Use uploads for datasets you own, and ask Jason to change the pipeline for datasets it produces.

Editing cells

Cells are not editable in the grid itself. Click the row to open it, press Edit at the bottom right of the dialog’s Data tab, and click the value you want to change. On the row’s full page, values are editable straight away: click one. In both places Enter or clicking away saves and Escape cancels. Numbers, currency and percentages are checked as you save: “$1,299.00” becomes 1299, “abc” in a price column is rejected.

Each save writes a new version containing the whole worksheet with your one change, and makes it active. See Row detail and provenance and Versions.

Worksheets

A dataset can hold more than one table. Pipelines that publish several related tables (products and their offers, venues and their menu items) use worksheets; so does a workbook you upload.

• New worksheet… asks for a name (up to 31 characters; no \ / * ? : [ ]) and creates it empty in a new version. Upload data into it afterwards from the header menu.
• Rename keeps the old name as an alias so links, queries and dashboards keep resolving.
• Delete creates a version without that worksheet. A dataset always keeps at least one.

Worksheet tabs appear above the grid only when there is more than one. In SQL, the worksheet you are viewing is data and the others get sheet_ aliases; see Query and export.

Deleting rows

Select rows and choose Delete {n} rows. The confirmation reads “Delete 12 rows? A new dataset version will be created.” The rows disappear from the active version; the previous version still contains them. A pipeline that publishes to this dataset will include them again on its next run if they are still on the source, so for a permanent exclusion ask Jason to change the pipeline instead.

The connections strip

The bottom-left corner lists every place this dataset receives data from or sends data to.

In order:

• from {pipeline} (violet). The pipeline that produces this dataset.
• to {pipeline}. A pipeline that reads this dataset as an input.
• Sources (green when live, amber when not). Integrations that push rows in, such as a Google Sheet or an ingest webhook. Live push sources show a copy button for their ingest address.
• Outputs (blue). Output integrations tied to this dataset. Pull-style outputs show a copy button for their serve address.
• in {dashboard} (indigo). Dashboards that read this dataset.

Only the first two connections show inline. View {n} more opens a dialog with a card per connection and an Add connection card that leads to Integrations. With two or fewer connections there is an inline Add link instead.

A Sync button appears when the dataset has at least one live pull-style source (a Google Sheet, for example). It fetches fresh rows from every such source now, the same way a scheduled run does before it starts, and reports “Synced 120 rows from 1 source” or “Already up to date”.

Alert rules that watch the dataset, and outputs that rules send its rows to, are not shown here; find them under Alert rules.

Empty states

• No data yet. The dataset exists but has no versions. If a pipeline builds it, run the pipeline.
• No data in this version. The version you opened has no columns in this worksheet. Choose another version or worksheet.
• No products match your search. Clear the filter or the source chip.

Which product had the biggest price drop in this version compared with the last one?

Jason compares the active version with the previous one on product_id, lists the five largest drops with their old and new prices, and links each row so you can open it. No pipeline change is made.

What’s next