Release Notes 15.09.2026 - v5.0.0

📅 15.09.2026
🔖 Version 5.0.0

✨ Features

• Minerva's new assistant engine
  Minerva now answers through a step-by-step agent engine rather than a single fixed pipeline. It chooses the right tools for your question and streams its progress so you can follow the reasoning as it happens. Your existing conversations remain available and unchanged.
  • Specialist sub-agents for complex questions
    Minerva can now create and delegate work to specialist sub-agents when handling complex questions. These sub-agents can analyze different perspectives and reasoning paths in parallel, allowing larger and more involved analyses to complete reliably. Sub-agents can be configured with dedicated skills and capabilities, helping Minerva select the right expertise for each task. You can follow each sub-agent’s progress as the analysis runs.
  • Minerva Workspaces
    Minerva now has a private working area where it can create and edit real files. Ask it for a spreadsheet, a chart, a converted document or a written report and it produces the actual file for you to download. Every workspace is private to its user, and generated files are marked as AI-produced.
  • Upload files directly to Minerva
    Files can now be attached directly in the chat and uploaded to Minerva’s workspace. Minerva can use these files as part of the conversation, allowing you to analyze, transform or work with your own documents and data without adding them to the Context Hub first.
  • Minerva on on-premise models
    Minerva now works properly on locally hosted models. Context limits, summarisation thresholds and tool responses adapt to the smaller window automatically, so long, multi-step conversations complete without errors. When a question genuinely needs a larger model, Minerva says so instead of failing.
  • Long-running analyses no longer time out
    Deep analyses that run for many minutes now complete and save correctly. Previously a long-running answer could be lost at the final step once the session had been open too long.
• Context Hub
  Uploaded knowledge now has a proper home. The new Context Hub gives every document a card view with file-type colouring, filtering and search, counts per tab, and a detail view showing chapters, images, text chunks and keywords. Files can benreplaced, ingestion re-run, and documents downloaded again.
  • Context for perspectives
    The Context Hub covers perspective context as well, with grouping by perspective or entity so you can see and manage what knowledge is attached where.
  • Context is searchable immediately and survives re-imports
    Knowledge you upload for a company, a perspective or an activity is searchable straight away, without waiting for an import to run. It also stays attached when a perspective is rebuilt, so re-importing no longer costs you the context you added.
  • Documents are summarised on upload
    Every uploaded document now gets an AI-written summary, stored and searchable in its own right. Minerva can tell at a glance what a document is about before reading into it, and chapter-level summaries give the same benefit within long documents.
  • Smarter knowledge retrieval
    Context search now works in stages: it first identifies the most relevant documents, then searches within them, rather than scanning every fragment at once. Minerva can also navigate a document directly — listing its chapters, opening one, or searching inside a single document — so a near-miss can be followed up instead of re-phrased. Answers stay accurate as your knowledge base grows.
  • Images are interpreted and searchable too
    Images contained in uploaded documents are now interpreted as part of the ingestion process. Their content is analyzed and transformed into searchable chunks, allowing Minerva to retrieve and use information found in diagrams, screenshots, charts and other document visuals alongside regular text.
• Faster setup in the Builder
  Creating a perspective or entity-graph now offers "Create & configure", taking you straight into its configuration instead of back to the list. You can also follow the data lineage by clicking through from an entity to its data source, with the right item already selected. Add and search controls have moved to the top right of each table.
• Pick a database from a list
  When setting up a data source, the available databases and schemas are now offered as a list once the connection has been tested successfully, instead of being typed by hand. A confirmation appears when the credentials work, and the database selection unlocks at that point.
• Assign a role when creating a user
  User creation now offers an optional role. Choosing Customer Admin or User places the new user in the matching group automatically; leaving it empty assigns the standard user role as before.
• Clear AI transparency
  Minerva now states plainly that it is an AI system and that its answers should be verified before decisions are based on them, in line with the EU AI Act. Files Minerva generates carry a machine-readable marker and a visible notice identifying them as AI-produced.

🐛 Bug Fixes

• Long conversations no longer come back empty
  A chat running longer than a few minutes on a slower model could finish its analysis and then lose the answer at the last moment. The completed answer is now saved correctly.
• Interrupted chats recover instead of spinning
  If a conversation was interrupted — by a restart or an unexpected failure — it could stay in a running state indefinitely, leaving an endless spinner. Such conversations are now detected, shown with a clear message, and can be continued from where they stopped. Deleting a conversation now removes all of its parts, leaving nothing behind.
• Improved chat interface
  A broad set of chat issues has been resolved: flickering while an answer streams, incorrect icons, the token counter overflowing its area, the model shown not matching the one used for the run, scrolling not jumping to the latest answer, the chat panel disappearing when narrowed, and missing confirmation when deleting a conversation. Created files now show their location, and copying an answer confirms it.
• More reliable tool use
  Task planning and tool execution have been made more robust, so the assistant recovers cleanly when a tool call does not go as expected rather than stopping or repeating itself.
• Newly added tools no longer appear unannounced
  Tools registered on the platform were automatically becoming available to the standard assistant. Tool availability now follows the agent's configuration, and model compatibility set on an agent is properly respected.
• Refreshing a slow view no longer overloads the system
  Switching or refreshing an analyzer or dashboard tab while a slow query was still running sent a fresh request every time, which could overwhelm the database. Identical requests now wait for the one already in flight.
• Filters on activity timestamps load correctly
  Saved filters defined on an activity timestamp could fail to load in the analyzer. They now resolve correctly.
• Date-range filters compare dates as dates
  Date-range filters were comparing text against real timestamps, which produced wrong results on some data sources. Declared date and date-time properties are now stored and compared as proper timestamps, including date-only values and Oracle sources.
• Activities with filters can be deleted
  Deleting an activity in a perspective failed with an error when a filter referenced it. The filter relationship is now resolved and the activity deletes cleanly.

🔮 Coming Soon

• Governance for Agents/Skills/Prompts
  Agents, skills and prompts will be shareable with specific users and departments, giving organizations control over who can access and use them.
• Permissions and Approvals for Minerva Actions
  More granular permissions and approval flows are coming for actions requested by Minerva, giving users greater control over what AI-initiated actions are allowed to execute.
• Folders in the Context Hub
  The Context Hub will gain folder support, making it easier to organize, structure and navigate larger collections of company knowledge.
• Chat UI/UX Improvements
  Minerva’s chat experience will receive further UI and UX improvements, making conversations, ongoing analyses and interactions with agents easier to follow and use.