Skip to main content

February 2026

February added roles to AI Studio, search ranked by relevance and a reworked template editor. You can now chat with support from inside the product.

Highlights​

AI Studio roles​

AI Studio (previously Data Studio) now has its own access model. Each user is a member of an objective, with a role that sets what they can do with its samples and models. Organisation administrators manage these memberships in organisation settings. Creating an objective makes you a member automatically. When you remove someone's role, they are also disconnected from the models they were attached to.

Week of 2 February​

AI & data​

  • AI Studio roles: described above.
  • Advanced sample filters: filter samples on many more criteria, when browsing and when importing.
  • Sample sorting: sort the sample list.

Validation​

  • API lookup values kept: values from an API lookup count as user edits, so a later prediction does not overwrite them.

Interface​

  • In-app support chat: start a support conversation from inside the product.

Fixes​

  • Your edits are no longer lost when the whole task is validated.
  • Pagination in single-step projects works again.
  • Creating a task no longer requires an unrelated permission.

Week of 9 February​

AI & data​

  • Benchmark settings: a model can store its own benchmark configuration.
  • Version switching: switch versions in the interface.

Validation​

  • Faster lookups: lookups are much faster.

Import & document processing​

  • TIFF support: TIFF files are accepted, and a project setting lists the accepted file formats.
  • Text grouping v2: a new way of grouping text in plain-text documents.
  • Re-upload tasks: re-upload a task from the interface.
  • Relevance ranking: list tasks and uploads by search score, with matched terms highlighted.

Interface​

  • Readable audit log: the audit log is formatted for reading.
  • Diagnose button: trace why a model produced a given prediction.

Fixes​

  • Searching by task or upload name returns complete results again.
  • Navigation no longer mixes up current and cached state.

Week of 16 February​

AI & data​

  • Objective management: create, edit and delete objectives, with confirmation before deleting.
  • Metadata filters: filter samples and imports on document metadata.
  • Page-based chunking: documents are chunked by page, so the model keeps each page's context.

Validation​

  • Field validation: configure validation per field.
  • Labels table: a new labels table with more controls.
  • Accuracy filter: filter tasks by prediction accuracy, for single-step projects first.
  • Annotation preferences: set your preferences in the annotation editor.

Fixes​

  • RTF documents no longer fail to import.
  • Deleting a label also removes it from templates.

Week of 23 February​

Validation​

  • Template editor: empty lines stay in the grid, labels are placed automatically, and labels that are not placed are clearly marked.
  • Lookup enums: back a lookup field with a fixed set of values.

Configuration​

  • Pipeline settings: pipeline settings use the same structure as the rest of the product's settings.
  • Segmentation on the pipeline step: segmentation is now set on the pipeline step, no longer at project level.

Notifications​

  • Reply to the uploader: send a notification back to whoever submitted the document.

Fixes​

  • Duplicate names show a clear validation error.
  • More date formats are recognised.
  • Page-based chunking settings now take effect.