Roles and permissions¶

Access control has two independent levels.

Global role — stored on the account, answers "what may this account do to the platform?" Create datasets at all, manage other users, touch the model registry.
Dataset role — stored on the membership row, answers "what may this account do to this dataset?"

Datasets are IQUANA's unit of governance, so nearly every permission lives at the dataset level. An account can hold a different role in every dataset it belongs to.

Global roles¶

Global role	May
admin	everything, platform-wide — user management, global roles, the model registry
member	create datasets of their own
guest	nothing globally; can only work inside datasets they were invited to

Dataset roles¶

Ranked from least to most privileged. Each role includes everything the role above it in this table can do.

viewer¶

Read-only. Sees the dataset, its images, its label space and its annotations.

annotator¶

Everything a viewer can do, plus:

create annotations, and edit or delete their own
submit masks
use interactive AI segmentation

reviewer¶

Everything an annotator can do, plus:

edit any annotation, not only their own
reopen and delete masks
approve, reject and revoke reviews; purge unreviewed items
export annotations and quantification results

curator¶

Everything a reviewer can do, plus:

update dataset settings; upload and delete images
write image metadata
set the pixel scale and other calibrations
manage the label space
run batch inference and training
export images
list members and create invites

owner¶

Everything a curator can do, plus:

delete the dataset, and transfer its ownership
grant and revoke membership
revoke invites

Why calibration and metadata sit at curator, not annotator¶

Two permissions are deliberately held back from people doing annotation work:

Image metadata

Metadata defines the subgroups a dataset is later compared across. An annotator retagging images would silently redraw the groups someone else's analysis is built on.

Calibration and pixel scale

Changing the scale silently rescales every quantification number in the dataset. Calibration changes what the stored measurements mean, not what is annotated — which is curation, not annotation.

Per-member exceptions¶

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

Invites and privilege¶

Redeeming an invite never downgrades an existing member: roles are ranked, and the higher of the two is kept.

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

The first account

The first account created on an installation becomes the platform administrator, whether or not self-registration is enabled. See Configuration.