Datasets
A dataset is a named set of tables you can share and publish as one thing.
It is not a model and not a table. A model is where you work; a dataset is what you hand over. The tables in a dataset can come from more than one model, and a table can belong to several datasets at once.
Find them under Datasets in the sidebar.
Creating one
- Open Datasets.
- Click New dataset.
- Give it a name.
Then add tables to it, either from the dataset or from a table's own menu with Add to dataset.
Members
The tables in a dataset are its members. Adding and removing them changes the dataset's current state — what it holds right now.
Some tables cannot be added, and the dialog says which and why:
| Refusal | Reason |
|---|---|
| Tables from an external connection | They are read through a live connector, so they cannot be released on their own |
| Restricted table | You cannot publish a table whose access is walled |
| You don't have access to this table | You cannot add what you cannot read |
When you add tables you own, you are offered "Also give access to the tables I own in this dataset" — sharing the dataset does not by itself share the tables inside it, so this is the shortcut that does both.
Deleted tables stay listed. A table that was removed shows as "This table was permanently deleted. It stays listed so the dataset's history reads true." A dataset's past releases referred to real tables, and quietly dropping them would make old releases describe something that never existed. If a table is merely in its model's trash, the row says so and you can restore it there.
Releases
A release freezes the dataset: it pins each member table at its current version, so anyone reading that release sees exactly what you released, however much the tables change afterwards.
- Open the dataset.
- Click Add a release.
- Optionally give it a label — a release name such as
2026-q3. - Confirm.
Labels must be unique within the dataset.
Once a release exists, every member is shown relative to it:
| State | Meaning |
|---|---|
| Unchanged since this release | The table is exactly as released |
| Ahead of this release | The table has changed since |
| Behind this release | The table is at an older version than the one released |
A new release is only available when something has actually changed — if nothing has, the app tells you "No member has changed since the last release."
Describing a dataset
A dataset carries the metadata that makes it usable by someone who did not build it:
| Field | What it is for |
|---|---|
| Description | What the dataset contains |
| Keywords | Terms people search by |
| Licence | What others may do with the data — see below |
| Attribution | Credit for the source the data came from |
| Creators | The people or organisations behind it |
| Related links | Documentation, a source page, anything relevant (http and https only) |
| Maintenance guide | How the dataset is kept up to date |
| Labels | Your own categories |
Licences
Pick a licence from the list and the link is filled in for you — you never type a licence URL for a known licence. That is what stops two datasets spelling the same licence differently.
If your licence is not on the list, choose the custom option and give a name and a URL. That is the only place a licence URL is entered by hand, and it must be https.
A licence's credit line is not part of the licence. It belongs in Attribution.
Access
Manage access controls who can see the dataset. Someone without access sees a Request access action instead, and once asked, the app tells them "The people who can grant it have been asked."
Datasets you can see but do not own appear under Shared with you.
Publishing
Making a dataset public is a catalog release. See that page for what publishing requires and what it means.