Manage access

You need: curator to list members and create invites; owner to grant or revoke membership and revoke invites.

Read Roles and permissions first — the role you hand someone is the whole decision.

Steps

1. Open the dataset's access page.
2. Create an invite and choose the role it grants.
3. Send the invite link to the person.
4. They redeem it and appear as a member.

Access page → create an invite → assign a role → the same surface as annotator vs. curator.

Choosing a role

Give them	If they should
viewer	look, and nothing else
annotator	annotate, and edit only their own work
reviewer	edit anyone's annotations, run review, export measurements
curator	also upload images, manage labels, calibrate, run inference and training
owner	also delete the dataset, and manage membership

Calibration and metadata are curator-level for a reason

Both silently change what existing measurements mean. If someone only needs to annotate, annotator is the correct answer even when they ask for more.

Exceptions without new roles

A membership can carry extra or denied permissions individually, so one collaborator can be given — or refused — a single capability without inventing a role for them.

Invites never downgrade

Redeeming an invite keeps the higher of the two roles, so re-inviting an existing member cannot accidentally demote them.

Platform-level accounts

Creating accounts, setting global roles and deciding whether self-registration is allowed are platform concerns, not dataset ones. See Configuration.