Release Notes for Mindbreeze InSpire
Version 25.6

Innovations and new features

Faster and more accurate answers through optimizations of the vector index and similarity search

As part of the Mindbreeze InSpire 25.6 Release, the vector index and similarity search have been extensively optimized. In addition to the tailored optimization of the query for the similarity search and additional optimizations of algorithms, the vector index format has also been revised. The understanding of permissions and search restrictions takes place in the vector index, which makes the similarity search more efficient and even more precise in recognizing matches and generating answers from them.

These optimizations enable Mindbreeze users to receive their rights-checked and fact-based answers faster. This increases efficiency, especially for answers generated from large amounts of data.

The optimizations to the vector index and similarity search are now available to users with the new vector index format. No configuration adjustments are necessary due to the automatic application via requests processed by Retrieval Augmented Generation (RAG) pipelines.

Link to the documentation

Mindbreeze InSpire supports OpenAI's open-source models gpt-oss-20b and gpt-oss-120b

Mindbreeze InSpire expands the selection of available large language models to include OpenAI's open-source models gpt-oss-20b and gpt-oss-120b. These are now available to Mindbreeze customers for their GenAI and agentic RAG use cases. By using 20 billion and 120 billion parameters and a context length of up to 128,000 tokens, Insight Apps become significantly more powerful. Due to optimized model efficiency, the upgrade of the infrastructure is not necessary. Customers can start the deployment and use it immediately.

Mindbreeze InSpire app available in Microsoft Teams

For seamless integration of Mindbreeze InSpire into workplace environments and business processes, end users now have access to the Mindbreeze InSpire app in the Microsoft Teams App Store. This allows users to work comfortably in their familiar Microsoft Teams environment with all the conveniences and advantages of Mindbreeze InSpire combined. In addition to significantly more direct access, users can easily maintain their existing workflows and initiate them from the Mindbreeze InSpire app.

Additional integration options such as the Microsoft Outlook Add-in, Microsoft SharePoint Online, and Salesforce are available to Mindbreeze customers.

Link to the documentation

More precise adjustment of Neural Reranking with new boost factors

With the Mindbreeze InSpire 25.6 Release, search experience administrators get two new boost factors to adjust the ranking of Neural Reranking more precisely. By configuring the boost factors “Reranked Answer Doc Boost Exponent” and “Similarity Weight For Reranking Exponent,” the new ranking can be even better adapted to individual needs. This allows administrators to prioritize certain documents, such as company guidelines, and thus make answer generation more granular.

Link to the documentation

Status information for Mindbreeze InSpire and PostgreSQL now available

Administrators can now obtain status information about Mindbreeze InSpire and PostgreSQL via the newly available menu item “Administration”. Information such as the currently installed Mindbreeze InSpire version and the amount of used storage space is now visible at a glance. In addition, the status of PostgreSQL is displayed and the necessary migration to a higher version is also possible via this menu item.

To use the menu item “Administration”, the new Mindbreeze InSpire roles “InSpire Builtin Database Administrator” or “InSpire System Administrator” are required. Afterward, the migration to PostgreSQL version 15 can be carried out.

Link to the documentation

Additional improvements

Optimized security through various updates

With the Mindbreeze InSpire 25.6 Release, security has been optimized through updates to CoreOS, Chromium, and Python packages.

Security relevant changes

• Updated: CoreOS Security to version 42.20250803.3.0 (CVE-2025-0725, CVE-2025-5399, CVE-2025-0665, CVE-2025-4947, CVE-2025-5025, CVE-2025-8058, CVE-2025-0167).
• Updated: Python dependency (CVE-2025-55197, CVE-2025-2828 , CVE-2025-4565).
• Updated: Chromium to version 139.0.7258.154 (CVE-2025-8010, CVE-2025-8011, CVE-2025-8292, CVE-2025-8576, CVE-2025-8577, CVE-2025-8578, CVE-2025-8579, CVE-2025-8580, CVE-2025-8581, CVE-2025-8582, CVE-2025-8583, CVE-2025-8879, CVE-2025-8880, CVE-2025-8901, CVE-2025-8881, CVE-2025-8882, CVE-2025-9132, CVE-2025-9478).
• Updated: Poco to version 1.14.2 (CVE-2025-6375).

Additional changes

• Added: New Management Center Service Administration UI (Service Status, PostgreSQL migration) and new Keycloak roles "InSpire Builtin Database Administrator" and "InSpire System Administrator".
• Added: Support for defining credentials, endpoints and certificates in helm chart.
• Added: New relevance parameters in the Management Center menu item "Search Experience", so Search Experience Administrators can customize the answer scoring when a neural reranker is being used.
• Added: User Input Transformation via Prompt in Generative Pipelines.
• Added: Metrics for RAG Pipeline Evaluations in Insight Services for RAG.
• Fix for: Usage with authenticated proxy configurations.
• Fix for: Built-in credentials were passing to services even if no restart policy was active.
• Fix for: Blank error page in the Management Center Kerberos configuration tool, if the password of the account expired.
• Fix for: Navigation for the chapter “searchform” was not working correctly in help.mindbreeze.com.
• Fix for: Fallback for failed MSAL requests.
• Fix for: Fallback for failed MSAL requests on mobile.
• Fix for: Email-Alerts did not show the correct selected search tab.
• Fix for: Stateless InSpire Kubernetes containers may hang and do not recover.
• Fix for: OAuth issue with mesindexclient.
• Fix for: Fallback handling for missing translations.
• Fix for: Collapsed service name list in RAG page header had bad visibility in the case of multiple services.
• Fix for: Runtime CUDA errors like “Out of memory” can lead to temporary incomplete inverted index construction.
• Fix for: Parallel login requests were blocked by Keycloak due to brute force detection.
• Improved: Performance for Management Center Configuration Rendering.
• Improved: Robustness of displayed answers and sources.
• Improved: Performance for low-latency indexing use-cases especially in the case of smaller sized documents.
• Improved: Vector Index Search with constraints.
• Improved: Filter Service supports located content with HTTPS URLs and "access-token" URL parameters.
• Changed: Count PDF Diff Service Calls are now counted as Insight Service Calls.
• Removed: Support for Aleph Alpha.