Run shared data sets from chat

Team members with the Business user role get a chat-only workspace. The cards under Set up for your team on their chat page are the scenarios your data team shared with them - and each card runs directly. Nobody has to phrase a request or know what a scenario is.

Running a data set

A data set without settings starts the moment you click it.
A data set with settings shows a confirm card first: each setting with a plain-language label, its current value, and room to change it. Press Run to start.
Changes on the confirm card apply to that run only. The shared setup your data team configured stays untouched, and two people running the same data set with different values never interfere with each other.

The run happens in your chat

Starting a run opens a conversation for it. Your request appears as your own message, and the run is the answer to it: a single card that leads with a sentence saying where the run has got to, rather than with a console.

While it runs, the card says the data set is being prepared and how long it has been going. You can leave the page open; the result appears in place.
When it finishes, the files it produced appear in the card with download buttons, next to Run it again and Back to data sets.
The thread is saved like any other conversation, so a reload or a later visit brings the run and its result back. Each run gets its own thread instead of replacing the previous one.

Any settings you confirmed are listed back to you on the card, so it stays clear which values produced which files.

When a run fails

A failed run is explained in two or three sentences: what went wrong, described in terms of the data or the system it was talking to, and whether it is something you can act on or something your data team has to fix. The card also says plainly that nothing you did caused it and nothing in your systems was changed.

Show technical details expands the exact log, unchanged, for anyone who wants to read it.
Copy details for your data team puts the run identifiers and the end of the log on your clipboard as one block to paste into a message. That is the quickest route to a fix: it carries everything the team needs to find the run.
If the run wrote any files before it stopped, they appear as partial files rather than as your data.

The explanation is written from the run’s own log, so it is only as specific as the log is - see Logs & retries for what scenario authors can do to make it better. Where no AI provider is available, the card keeps its plain status copy and the log stays one click away.

What business users can and cannot do

Running a shared data set needs no extra permissions: the Business user role already includes it. The role cannot create or edit scenarios, so the cards are exactly what the team shared - nothing else is reachable from this page. The failure explanation is covered by the same permission as reading the log itself, so it is never shown to anyone who could not open the log directly.

Anyone who can edit a scenario shares it from the scenario’s settings: enable Share with business users, then give it a business label and description written in the language of the people who will run it. The card shows how many settings the run will ask about; variables reserved for infrastructure (DATAMAKER_*, proxy and interpreter variables) are never shown or editable.

Per-run values in agent chat

The same one-run-only semantics apply when the agent runs a scenario in chat. When you confirm the run with different values, the agent passes them as per-run overrides instead of rewriting the scenario’s stored configuration. Changing the shared defaults for everyone remains an explicit, separate action.