Release Notes 12.08.2026 - v4.1.0

📅 12.08.2026
🔖 Version 4.1.0

✨ Features

• A new look for Noreja
  The interface has been updated to Noreja's new colors and logos, bringing the application in line with the website. The navigation bar, chat, and all core screens now share one consistent visual identity.
• AI Hub: manage your prompts
  A new AI Hub in the navigation gives you direct control over the AI prompts used across the platform. You can review and edit prompts, see exactly what is sent for each processed request, and refine them over time — no code changes or service restarts required.
• AI Hub: manage your agents
  The AI Hub also includes agent management. Agents can be created, configured, and adjusted from the interface, so setting up and tuning AI behaviour no longer depends on a developer or a deployment.
• The latest AI models
  Opus 5 and GPT 5.6 have been added to the available models, so you get the strongest results currently available for your questions.
• More analysis power for Minerva
  Minerva can now reach considerably more of your data. New tool calls cover the filtered process graph and process list, dashboard-wide key figures, filter options, and batching statistics — so you can ask deeper questions and get answers grounded in more of your process data.
• Share perspectives with your colleagues
  When creating a perspective you can now choose who can access it: keep it private, make it available to everyone, or share it with specific users and departments. Access can be granted as read, edit, or re-share, so colleagues can view or actively work on the perspective as needed. Existing perspectives are migrated automatically.
• Share entity-graphs with your colleagues
  The same sharing options are now available for entity-graphs. Choose between owner-only, shared with everyone, or shared with selected users and departments, and grant read, edit, or re-share rights. Entity-graphs you don't own are marked with an icon so ownership is clear at a glance.
• See who has administrator rights
  Customer administrators are now labelled as such in the user list, making it easy to tell administrators from regular users — and to see at a glance who holds full permissions.
• New connector: OData
  You can now connect to OData sources — including SAP systems — and use them in the Entity-Graph Builder alongside the existing connectors. Both OData v2 and v4 are supported, and the service metadata is read automatically so entity sets, keys, and relationships between them are discovered for you.
• Joined properties, end to end
  Properties joined from another table now work reliably throughout the platform — in relationship mappings, in dimension filters, and in analysis. Properties are matched by identity rather than by name, so tables carrying the same column name in different schemas no longer collide, and every property is shown with its schema and table so you always know which one you are looking at.
• Filter by start, end, and intermediate activities
  Activity types can now be filtered by whether an activity starts a case, ends it, or sits in between. The distinction is available in both the activity filter and the From-To filter, and activity lists can be sorted by title — making it much easier to isolate the part of the process you care about.
• Faster, more predictable imports
  The way causal relationships are derived during an import has been redesigned to use bounded reachability instead of enumerating every possible path. Dimension loads that previously stalled on densely connected data now complete in predictable time, with the same causal result. Import steps also report their time and memory use, so slow imports can be diagnosed precisely.
• Faster property statistics
  Statistics for all properties of an activity or relationship are now calculated in a single database query rather than one query per property. Property panels on large graphs open noticeably faster.
• Less waiting on background checks
  The blocked-state check for dimensions and entity-graphs now includes running indexing directly in its answer, so the interface no longer has to poll for that information every few seconds. The result is a quieter, more responsive application.
• A dedicated service for document ingestion
  Document embedding now runs in its own dedicated service. The pipeline — fetching, partitioning, chunking, and embedding — is separated from query-time AI operations, so both can be scaled independently and document processing no longer competes with your live questions.
• Uploads that don't make you wait
  Document upload is now asynchronous. Once you upload, processing continues in the background and you can leave the page immediately instead of waiting for the ingestion pipeline to finish.
• Actions stay within reach in long flows
  Stepper-based flows now keep their action buttons pinned to the bottom while the step content scrolls, matching the sticky save and cancel bar of the redesigned editors. No more scrolling to the end of a long step to find the next button.
• The newest Jupyter Notebook in the Workbench
  Workbench instances now run the latest Jupyter Notebook release, bringing its newest features and security fixes.

🐛 Bug Fixes

• Reliable permission editing across contexts
  Editing permissions in one context no longer disables the permission controls in every other context. Each context can now be edited independently.
• Language switching that sticks
  Switching the interface language now persists correctly across the session, so the application stays in the language you selected.
• Consistent quantities when importing with batching
  Relationship quantities no longer depend on the configured batch sizes. Imports now produce the same result regardless of how the batching is tuned.
• All tables available for modeling
  Tables that were previously missing from the "Available data objects" list now appear reliably, including when the same table name exists in more than one schema.
• Dependable user information for large teams
  User information is now requested in a way that no longer breaks when many users are looked up at once, so user and permission views load reliably in large organisations.
• Removed variants no longer linger
  When a variant disappears from the data during a reimport, it is now also cleared from your saved selection, so you are no longer pointed at a variant that no longer exists.
• Responsive indirect From-To filtering
  The indirect From-To filter no longer runs indefinitely on large models. The underlying query has been reworked so results come back in a reasonable time.
• Copying API entity-graphs
  Copying an API-based entity-graph no longer fails with an error.