Release Notes for Mindbreeze InSpire
Version 23.7

Innovations and new features

Mindbreeze InSpire AI Chat

With the Mindbreeze InSpire 23.7 release, Mindbreeze InSpire offers a new way of gathering information. Customers can now use the Mindbreeze InSpire AI Chat to obtain relevant facts in natural language. In addition, a link to the source of the information is provided so that the user can find out additional context for the answer. With these changes, the Mindbreeze InSpire AI Chat can answer user’s questions and immediately offer summarised information in natural language.

Furthermore, the Mindbreeze InSpire AI Chat can be customised to meet various needs and requirements. Administrators can create individual models and pipelines with just a few clicks and precisely control them with constraints. This allows administrators to create customised models for different applications and edit existing models and pipelines.

The technological basis for the Mindbreeze InSpire AI Chat is a Generative AI. The required facts are collected from Mindbreeze InSpire using Retrieval Augmented Generation (RAG). A Large Language Model (LLM) then processes the collected facts and generates answers, taking into account the user’s access rights. This allows the user to receive information quickly and intuitively in natural language. This process is illustrated in a diagram below.

Mindbreeze InSpire AI Chat Pipelines based on Insight Services for Retrieval Augmented Generation

Mindbreeze InSpire Insight Services for Retrieval Augmented Generation (RAG) is a new functionality which enables Mindbreeze InSpire AI Chat. Support of multiple connectors ensures the recency and relevancy of the answers. Customers can immediately use configured connectors and find pre-indexed content in the Mindbreeze InSpire AI Chat. The individual information landscape is also mapped using semantic links. These enrich existing information with additional context, providing users with more comprehensive answers. The issue of data security is addressed through various authentication options. Users are only shown information for which they have the appropriate access rights.

The Mindbreeze InSpire AI Chat is therefore able to offer users a robust basis for finding information. Insight Services and RAG ensure that the information provided by an LLM is valid and up-to-date. As a result, Mindbreeze InSpire AI Chat enables users to find information quickly and intuitively, while also ensuring that it is based on facts, up-to-date and understandable.

Administration of the Generative AI - Creating and Managing AI Chat Pipelines

With the user-friendly administration of Mindbreeze InSpire AI Chat, data sources can be managed and models can be adapted for their respective requirements. Data sources which are configured in the Mindbreeze Search Client can be seamlessly transferred and immediately used in the Mindbreeze InSpire AI Chat. It is also possible to add or exclude data sources. Administrators can easily create various customised models. For example, models can be specialised for different departments, as shown in the following screenshot. To reuse older models, model versioning is also available.

Through various options administrators can perform fine-tuning. For example, pipeline constraints, tokens, and prompts for a Large Language Model (LLM) can be defined. Configurable data sets are also available. The Mindbreeze InSpire AI Chat is available to Mindbreeze customers on-premise, as SaaS, and in the cloud.

Link to the documentation

New and advanced connectors

Options for constraining the crawler in Microsoft Teams

Microsoft Teams is used in many companies for communication, as well as for the storage of documents. With the Microsoft Teams connector, users of Mindbreeze InSpire have quick and easy access to the documents from their respective Teams channels. With the Mindbreeze InSpire 23.7 release, administrators can control the crawler more precisely and adapt it to their needs, as well as determine exactly which Teams channels and documents are to be indicated and found by the search client.

Various constraints can be placed on the crawler. Using a regex pattern, Teams channels are excluded from the crawler on the basis of their name. Archived or private Teams channels can be excluded using the appropriate setting. By specifying the Teams ID, you can exclude Teams channels based on their ID.

Link to the documentation

Editorial maintenance of answers in Best Bets

With the Mindbreeze InSpire 23.7 release, the functionality of the Best Bets connector has been extended in terms of answer handling. This means that the editorial management of answers is now possible. Here, an administrator checks the correctness of the generated answers. If the answer is factually correct, it will be marked as a "curated answer" in the search results.

Link to the documentation

Technical extensions

Extended authentication and identity management with SAML attributes

With the Mindbreeze InSpire 23.7 release, the support for Single Sign-On (SSO) via SAML is extended. It is now possible to replace the identity of users with supplied SAML attributes or to add additional roles and aliases. This provides extensive configuration options to improve the control of user identity and permissions.

With this extension, additional roles or alias names can be sent directly in the attributes of the SAML XML file when a user logs in to the client service. This allows the addition of specific user roles and alias names based on the SAML attributes, but also the overriding of the identity of the user. In addition, pre-defined rules can be used to define obligatory attributes. This means that authentication is only successful if certain attributes or values are present in the SAML attributes.

Link to the documentation

Optimised user experience through more precise filter names in the Insight App Designer

When creating Insight Apps, setting filters is an important building block for users. They provide various restrictions for search results and speed up the process of finding information. The Mindbreeze InSpire 23.7 release optimises the user experience of the Insight App Designer by making the names of the filters easier to understand. This allows users to better understand the function of each filter and to use them in a more confident and targeted manner.

Link to the documentation

More precise customisation of answers with text segmentation profiles

With the Mindbreeze InSpire 23.7 release, the precise customisation of answers from Natural Language Question Answering (NLQA) to their respective use-cases is now available. This is made possible through text segmentation profiles, which allow you to control the exact text segmentation and specify with a standard profile whether an answer should consist of one or more sentences. In addition, custom profiles can be configured with which overlapping sentences can be controlled.

Link to the documentation

Security relevant changes

23.7.0.488

• Updated: Chromium to version 120.0.6099.109 (CVE-2023-5217 CVE-2023-5346 CVE-2023-5218 CVE-2023-5484 CVE-2023-5475 CVE-2023-5476 CVE-2023-5474 CVE-2023-5486) (CVE-2023-5996 CVE-2023-6348 CVE-2023-6347 CVE-2023-6346 CVE-2023-6350 CVE-2023-6351 CVE-2023-6345 CVE-2023-6508 CVE-2023-6511 CVE-2023-6512 CVE-2023-6702 CVE-2023-6703 CVE-2023-6704 CVE-2023-6705 CVE-2023-6707).
• Updated: LibreOffice to version 7.5.7.
• Updated: CoreOS to version 38.20231027.3.1 (CVE-2023-5345, CVE-2023-42754, CVE-2023-42756).
• Updated: MomentJS to version 2.29.4.

Additional changes

23.7.1.498

• Improved: Performance of the Highlighting for sample texting, if search terms are not contained in the content. Introduction of a new Client Service configuration option called "Disable Subword Highlighting".

23.7.0.488

• Added: Insight Services Logpool to app.telemetry in new Application Group Mindbreeze Insight Services.
• Added: The setting "Add Single Term Alternatives as Alias" in the Vocabulary Synonym Transformer can be enabled per default.
• Fix for: In the Windows installer the default install path is not suggested if Mindbreeze was not installed.
• Fix for: Open Redirect allows phishing attacks in the Management Center (mod_auth_oidc).
• Fix for: The folder “/icons/” in the Management Center is listable without a login.
• Fix for: Incorrect offset of the highlighting in the PDF preview. The pages are processed twice then.
• Fix for: WaitForEventServlet is stuck in endless conditionalWait.
• Fix for: Due to an incorrect STX offset of the highlighting in the PDF preview, a performance regression occurred.
• Fix for: URLs in the Sitemap that contained "null" values.
• Fix for: URLs that pointed to direct messages and missing URLs.
• Fix for: A regression in delta crawling which was introduced with a json library upgrade.
• Fix for: Index status on master if index is running on different node (enterprise environment).
• Fix for: app.telemetry server may start on a consumer node.
• Fix for: Sentence extraction if a span element starts at the same position as the previous sentence ends.
• Improved: The servlet „find“ and „aggregate“ set the expandable flag to false in the query expression, but can be activated via the URL parameter "expand_query=true".
• Improved: Robustness against uploading the wrong files in the license upload form.
• Improved: Error handling by the client service in case of SSL certificate issues.
• Improved: Minor Sentence Transformer for Semantic Search.
• Improved: Overview of dynamically aggregated metadata filters, by hiding internal filters in the Insight App Designer.
• Improved: Atlassian Confluence Connector -  handling of special access control involving space permissions with global permissions in special use cases.
• Changed: Updating the app.telemetry dashboards is more robust against unusual referrers.