Release Notes 09.06.2026 - v3.30.0
📝 This release makes Minerva, our AI assistant, noticeably smarter and more dependable — with more accurate document retrieval, fine-grained access control over the company context it can use, non-blocking document ingestion with live progress, and graceful handling of long conversations. It also brings searchable, paginated administration for users and permissions, and resolves several issues across the workbench, conformance analysis, and filters.
📅 09.06.2026
🔖 Version 3.30.0
✨ Features
- Access control for company context
Restrict company-context documents to specific departments and permission roles, or make them available to everyone. Assignments can be managed and revoked at any time, and Minerva only draws on context the current user is allowed to see — so sensitive information stays protected. (Access changes take effect for new messages; earlier conversation history is unaffected.) - More accurate answers from Minerva
Document search now combines meaning-based search with exact keyword matching, every passage is enriched with a note on where it fits in the wider document, and PDFs are parsed with visual understanding — so content in tables, diagrams, and process flows is captured far more faithfully. - Non-blocking document uploads with live progress
Uploaded documents and text now process in the background: you can leave the page immediately while embedding and indexing continue, and each document shows its own live progress. Related dimension and entity-graph imports are briefly locked while a context is still ingesting to keep data consistent, and interrupted ingestions recover automatically. - A clearer view of how Minerva works
The assistant's thinking panel now presents each planned step in readable form with a reason tailored to your question, and while it works you can watch the tools it calls in real time — each with a status indicator, how long it ran, and an inspectable preview of its result. - Dependable long conversations
Minerva now manages its context window automatically, protecting the most important content while trimming older history. Oversized questions, very large graph results, or overloaded conversations return a clear, actionable message — such as suggesting a shorter query or a larger model — instead of failing. - Availability now reflects the embedding service
If the embedding service is unavailable, every model is correctly reported as unavailable and Minerva clearly prevents use, rather than appearing available and then failing mid-request. - Searchable, paginated administration lists
The users, departments, and permissions lists are now paginated and searchable by name, so large organizations stay fast and easy to navigate.
🐛 Bug Fixes
- The workbench list loads correctly again
Workbench Settings once again shows running and stopped workbench instances and resolves user names properly — including showing the correct owner for workbenches whose user has since been deleted. - More accurate conformance results
Process steps that should be treated as hypothetical are no longer shown as actual causal relationships — they now appear correctly as violations, giving a truer conformance picture. - Filters can be deleted again
Improved filter deletion.