Release Notes for Mindbreeze InSpire
Version 25.2

Innovations and new features

Native and seamless transition between Insight Services Administration and Insight App Visualization

Mindbreeze enables with the Mindbreeze InSpire 25.2 release the native and seamless transition between the comprehensive Insight Services Administration and detailed Visualization through Insight Apps. Administrators can now select the RAG pipeline to be used for their AI Answers. This enables users to create a chat-like user interface in their Insight Apps to find relevant and rights-checked information by chatting with documents or teamrooms.

Administrators now select the desired RAG pipeline in the Insight App Designer and define the pipeline in the setting “Pipeline selection” with the Pipeline Key or Pipeline ID.

Link to the documentation

Continuous chat message abstraction available in Insight Services for RAG

With the Mindbreeze InSpire 25.2 release, Mindbreeze customers get a fundamentally new possibility for editing their RAG pipelines. Instead of interacting with the Large Language Model (LLM) in text form, the generation part of the RAG pipeline can now be edited natively with messages. This also makes it easier to edit the LLM. With this, Mindbreeze InSpire creates the basis for implementing Tool Calling and multimodal RAG pipelines (see graphic) in future releases and making them available to customers.

Administrators can now focus on the content of the prompt, while the necessary prompt structure is automatically provided.

Link to the documentation

Extended support for the optimization of enriched datasets from evaluations

Mindbreeze provides its customers with comprehensive support in the creation and optimization of datasets. Administrators evaluate and compare various Large Language Models (LLMs) with the retrieval of contexts. Datasets that originally only consist of questions can be enriched with contexts and subsequently with generated answers and reused as expected answers. The Mindbreeze InSpire 25.2 release makes it possible to download datasets from evaluations so that these datasets can be edited outside Mindbreeze InSpire and shared.

Administrators choose between CSV or JSON file format when downloading the dataset. After sharing or external editing, the dataset can be imported into Mindbreeze InSpire to compare the effective evaluation results based on answer quality, retrieval process or model settings.

Link to the documentation

Duplication of folders in the File Manager possible

With the Mindbreeze InSpire 25.2 release, Mindbreeze customers can now duplicate their folders in the File Manager. This gives administrators an additional option for backing up and sharing data such as indices or log directories.

Link to the documentation

New and advanced connectors

Atlassian Confluence Plugin compatible with Confluence 9.2

With the current Long-Term Support version 9.2, Atlassian Confluence offers various improvements in different areas, such as security and performance. With the Mindbreeze InSpire 25.2 release, Mindbreeze customers also receive these improvements through the compatibility of the Atlassian Confluence plugin with Confluence 9.2.

Link to the documentation

Additional improvements

Increased security and updating of cryptographic libraries

To further increase the security of Mindbreeze InSpire, Chromium, Core OS and the Dell firmware have been updated with the Mindbreeze InSpire 25.2 release. In addition, cryptographic libraries for the Microsoft File and Microsoft CRM connector have been updated.

Link to the documentation

Security relevant changes

25.2.2.396

• Updated: Fedora CoreOS to version 41.20250331.3.0 (CVE-2024-12133, CVE-2024-12243, CVE-2024-56171, CVE-2025-24928, CVE-2024-21978, CVE-2023-31355, CVE-2024-8176).

25.2.0.346

• Updated: Keycloak to version 26.1.3 and OpenJDK to version 8.0.442 (CVE-2025-24970 CVE-2024-47535 CVE-2024-10451 CVE-2024-9666 CVE-2024-10270 CVE-2024-11734 CVE-2024-10492 CVE-2024-21210 CVE-2024-21235 CVE-2025-21502).
• Updated: Chromium to version 134.0.6998.88 (CVE-2025-0444, CVE-2025-0445, CVE-2025-0995, CVE-2025-0996, CVE-2025-0997, CVE-2025-0998, CVE-2025-0999, CVE-2025-1426, CVE-2025-1006, CVE-2025-1914, CVE-2025-1916, CVE-2025-1917, CVE-2025-1918, CVE-2025-1919, CVE-2025-1921, CVE-2025-1922, CVE-2025-1923, CVE-2025-1920, CVE-2025-2135, CVE-2025-2137).
• Updated: Dell Firmware (CVE-2024-39279, CVE-2024-28047, CVE-2024-36293).
• Updated: Fedora CoreOS to version 41.20250215 (CVE-2025-26466).
• Updated: bouncycastle library to version 1.80 in the Microsoft File Connector and in the Microsoft CRM Connector (CVE-2024-30171, CVE-2024-29857).
• Updated: Dependency for cross-spawn, jose, braces, Undici, tar and nanoid to resolve CVEs (CVE-2024-21538, CVE-2024-27088, CVE-2023-24807, CVE-2024-28176, CVE-2024-4068, CVE-2024-28863).
• Changed: Java product build system migration to gradle CVE-2022-40152 CVE-2024-47072 CVE-2024-52317 CVE-2024-50379 CVE-2024-56337 CVE-2022-36033 CVE-2024-52318 CVE-2024-52316).

Additional changes

25.2.2.396

• Added: Alternative synchronization schedule with service restriction for InSpire helm chart.
• Added: The size of /dev/shm is configurable.
• Fix for: Using proxy server with InSpireLLM models.
• Fix for: The proxy read timeout was too small in the GSA FeedAdapter.
• Fix for: Sentence Transformer models with Byte-Pair Encoding (BPE) were not functional.
• Fix for: AI Answer conversationInputPresetsJson init action.
• Fix for: Non-present metadata in document template.
• Fix for: HTML in retrieved sources, when using sources from the RAG pipeline for generation with formatted answers.
• Fix for: Docling API in Kubernetes did not work with EasyOCR.
• Fix for: The AI Answer Component with a defined prompt from the pipeline could not be edited.
• Fix for: Chat constraints were incorrect when switching models/pipelines.

25.2.0.346

• Added: Metadata can be requested and resolved in the index servlet response events of “Index Status”.
• Added: Action to duplicate folders in the File Manager of the Mindbreeze Management Center.
• Added: New index configuration option to include ignored terms (e.g. stop words from ReplacementTransformer) while transforming queries to similarity expressions.
• Added: New index configuration option to specify the fallback region size if not explicitly set in the query.
• Added: Support for HTTP Security Headers: Strict-Transport-Security, X-Frame-Options and X-Content-Type-Options.
• Added: Support for Confluence LTS 9.2.0.
• Added: Automatic resize functionality for standalone preview and chatframe component.
• Fix for: Possibility to set custom "data-app-fragment-ref-names" for 360-degree view action "showAllResults".
• Fix for: JavaScript crawling in the Web Connector could not be used in Kubernetes.
• Fix for: Deleting indexed documents in case of Kerberos Login Exceptions in the Microsoft File Connector.
• Fix for: Exporting properties as JSON in Import/Export dialog.
• Fix for: The date and modification date was not set correctly when using the REST API in the Microsoft Dynamics CRM Connector.
• Fix for: Missing synthesized properties in some NLQA answer cases.
• Fix for: Old app.telemetry logpools were deployed with a Snapshot deployment. Logpools are now always the current logpools of the installed version.
• Fix for: Special character handling for text fragments in "Answer" results.
• Fix for: Handling of expired change token in the Sharepoint Online Crawler.
• Fix for: The crawling of sites with special characters in their relative path failed.
• Fix for: ACL Precomputation treated missing ACLs as INCONCLUSIVE and early rejected objects that needed external authorization.
• Fix for: Issue in the PDF preview where in special setups the preview could not be opened properly.
• Fix for: Content Filter with mixed case enabled extensions are not mapped to documents with those extensions.
• Fix for: Admin API did not start on the first startup.
• Fix for: Height in standalone PDF preview and AI Chat. Adapted height relative to viewport on initialize and resize.
• Fix for: SharePointOnlineACL changes of document libraries broke delta crawls.
• Improved: Robustification of handling invalid/truncated UTF input in UTF8TermStore.
• Changed: InSpire LLM uses OpenAI API per default for newly created Pipelines. Already existing InSpire LLMs still use Huggingface TGI.
• Changed: Increased the read timeout for Service Gateway to one hour.