v11.20.0 December 07, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Access Control Enhancements
Search AI now automatically enforces RACL rules at design time for all new workspaces, protecting sensitive content. Access is based on the permission settings defined in the connector. Existing workspaces retain full design-time access for backward compatibility, with runtime access following source-system RACL rules. When design-time RACL is enabled, queries made through the public API return only public content. To enable design-time RACL for existing workspaces, contact support.Connector Enhancements
New Axero Connector The Axero connector enables ingestion and management of content from the Axero knowledge base platform, including Pages, Wiki, Discussions, Documents, Articles, Announcements, Blogs, and associated comments. The connector also preserves hyperlinks within content for use in the answers. GitHub Connector The GitHub Connector now supports additional content types, including the conversations on pull requests, issue threads, file names, and wiki pages. It also improves incremental synchronization of ingested content, ensuring that any new, updated, or deleted content is processed incrementally. Custom Connector Search AI enhances the Custom Connector configuration by introducing a new per-header Encoding Format option. Developers can choose whether a header value should be Base64-encoded or sent as plain text, providing greater flexibility for integrations and removing the previous limitation that forced Base64 encoding for all headers.Deprecation and Automatic Migration Notice
To improve platform performance and stability, Search AI is automatically upgrading legacy embedding models, legacy re-ranker models, and the legacy web crawler. Starting December 7, 2025, the following components will be automatically upgraded in applications where they’re in use:- Embedding Models: MPNet, LaBSE, E5, and BGE-M3 V1 are upgraded to BGE-M3 V2. Users must trigger training to apply the changes to their index.
- Re-rankers: MS MARCO Cross Encoder and Mixbread Large are upgraded to BGE Re-ranker. This change applies automatically with no user action required.
- Legacy Web Crawler: The legacy web crawler is replaced with the AI-Powered Crawler. Users must trigger a re-crawl to extract content using the new crawler.
v11.19.1 November 19, 2025
Patch Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Enhanced Default Extraction Strategies
Search AI now offers optimized default extraction and chunking strategies tailored to each content type and source, enabling more accurate, efficient data processing automatically. These enhanced defaults are automatically applied to new apps.View Document Access Controls
Search AI introduces new UI capabilities to provide greater visibility into RACL implementations. These enhancements are currently available only for the following connectors: Google Drive, HubSpot, Jira, Confluence Cloud, Confluence Server, Bitbucket, SharePoint, ServiceNow, Asana, Guru, and JFrog.- Permission Entity Viewer: View the groups, sub-groups, and users assigned to each permission entity.
- Document Permissions: Lists and displays the users and groups who have access to each document.
Connector Enhancements
Slack Connector The Slack Connector now offers enhanced flexibility and control for data ingestion:- API key-based authentication, eliminating the need for channel invitations (required earlier with OAuth) to retrieve channel content.
- Channel-based standard filtering option and advanced date-range based filtering for targeted ingestion.
- Support for crawling message history from the past six months.
- Expanded ingestion support for blogs and spaces, in addition to pages, and comments from both pages and blogs.
- Space-based standard filters for targeted content selection.
- Advanced filtering options for more precise indexing.
- Incremental sync capability and webhook-based deletion handling are available upon request.
v11.19.0 October 25, 2025
Minor Release This update includes only bug fixes.v11.18.0 September 27, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancement included in this release is summarized below.Dynamic Custom Data Support in Business Rules
Users can now define Business Rules using the customData fields sent via the Advanced Search API. During rule evaluation, context.customData is dynamically bound to the API payload, enabling rules to adapt in real-time based on the provided custom data.v11.17.1 September 15, 2025
Patch Release This update includes enhancements and bug fixes. The key enhancement included in this release is summarized below.Index Configuration Enhancements
The Vector Configuration page now offers more flexibility when working with custom embedding models and prompts for both text and visual content. Key improvements:- Separate Models and Prompts for Text & Image: Users can now configure distinct embedding models and prompts for text and image vectors. Image embeddings are managed using the Vector Generation - Image feature, while text embeddings are handled by Vector Generation - Text. The appropriate model is automatically determined based on the chunk extraction method.
- Simplified Prompt and Model management: Users can now configure and edit prompts directly in the Search AI UI, streamlining the setup process.
- Conditional Custom Prompt UI: The updated interface allows for the creation of new custom prompts directly when a custom model is selected for vector generation.
Change Log Support for Public API Events
The enhanced Change Logs now capture events triggered via the public API. This ensures complete visibility into system activity, enabling users to track, audit, and troubleshoot changes with a unified record of events, whether performed through the UI or programmatically.v11.17.0 August 23, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.LLM Stage for Document Transformation and Chunk Enrichment
Search AI introduces a new LLM Stage within the pipeline, enabling the seamless integration of large language models into transformation and enrichment workflows. Users can configure this stage using a model from the configured Models Library and a custom prompt. This feature empowers teams to automate content transformation, improve metadata quality, and add custom enhancements to content with minimal effort.API Stage Enhancements
The API Stage in Workbench has been enhanced to give users more flexibility when integrating with external services during the transform or enrich phase. Users can now include dynamic variables directly in the endpoint URL, in addition to the request body. These variables can be resolved from the document schema or chunk context available at simulation/training time. With this update, the API stage can be used with APIs that depend on document-level details as query parameters.Improved Training Visibility and Control
The training process in Search AI has been enhanced to provide greater transparency and flexibility. Users can now access detailed training logs that provide document-level visibility into the training lifecycle, including insights into errors and failures. This makes it easier to monitor the progress effectively and identify and address issues promptly.Answer Debug Improvements
The Answer debugging feature has been enhanced to provide comprehensive visibility into the retrieval process for effective troubleshooting, result validation, and a deeper understanding of system behavior. Key improvements:- Image Support: Extracted images are also displayed alongside text content in the Retrieved Chunks tab.
- Universal Retrieval Tab: This tab displays the retrieval process, including key metrics for each step in the pipeline.
- Enhanced Error Handling: Provide detailed error visibility in both UI and JSON views.
- UI Improvements: Remove visual clutter and enhance the information hierarchy.
Unified Schema Management Enhancements
Search AI introduces a new Manage Schema interface that gives users enhanced control and visibility over document fields based on the unified schema. With this update, users can:- View key field properties, including field name, description, data type, custom field indicator, and mapped chunk field.
- Create up to 50 custom fields for greater flexibility in data handling.
- Edit descriptions of existing ones to add meaningful references to their system fields.
v11.16.1 August 11, 2025
Patch Release This update includes bug fixes.v11.16.0 July 26, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Advanced RACL Capabilities in Connectors
The following key enhancements have been made to the RACL feature in Search AI connectors, enabling more secure and efficient access control.- Introduction of an RACL scheduler that gives the ability to sync RACL-related data from the connectors independently of regular content sync tasks. This provides greater flexibility to keep access permissions in Search AI in sync with those in the third-party applications.
- Automatic mapping of individual users to the permission entities in Search AI. This automation streamlines the implementation process, allowing enterprises to enforce RACL without requiring additional setup or processing. This has been implemented for the following connectors: Google Drive, Jira, SharePoint, Trello, Miro, LumApps, Workday, OneDrive, YouTrack, Zulip, Box, Shortcut, Zeplin, HelpScout, Slack, HubSpot, Zendesk, Microsoft Teams, Salesforce, Hive, GitHub, Aha, JIRA On-Prem.
- RACL support for the Confluence Server connector.
Simplified LLM Setup within Answer Configuration
Users can now configure the LLM directly within the Answer Configuration page, streamlining the setup process and eliminating the need to navigate to the Gen AI features section. This improvement enhances usability and reduces configuration time.v11.15.1 July 12, 2025
Patch Release This update includes only bug fixes.v11.15.0 June 30, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Multi-Vector Search
Search AI introduces Multi-Vector Search that allows users to associate multiple vectors with a single chunk of data. This advanced technique significantly improves retrieval accuracy by capturing different essential fields of a chunk. Each vector can be assigned a weight based on the relative importance of the fields used in that vector generation, enabling the system to prioritize specific fields during search and retrieval. This leads to more relevant and accurate results. For example, consider a document utilizing a dual-vector approach for information retrieval. One vector, generated from the document’s content and title, captures detailed context. The other vector, derived from questions related to the content, anticipates user inquiries. By appropriately weighting these vectors, retrieval efficiency is significantly enhanced. This is because the vectors encompass various content aspects, such as title, content, and derived questions, resulting in improved alignment with diverse user queries.Enhanced Multilingual Support
Search AI now supports over 100 languages, enabling a seamless multilingual experience across content ingestion, query understanding, and result delivery. Users can:- Add and organize content in these languages.
- Understand and interpret queries in these languages.
- Search and deliver answers and results in the same language as the user query.
- Markdown extraction support for Hungarian, German, and Chinese, improving the accuracy of content parsing in these languages.
- Image-based extraction and Advanced HTML extraction are now supported for German, enabling richer parsing of complex and visually structured documents.
API Stage in Document Workbench
Search AI now features a new API stage in the Chunk Enrichment process. This enhancement allows users to leverage external services for content transformation. Users can specify a custom POST API endpoint along with the necessary headers and request body to send chunks for enrichment. This integration facilitates seamless collaboration with third-party tools, enabling the classification, tagging, or enhancement of metadata for the chunks.Enhanced Insights Export API
The Analytics Export Public API now supports additional filtering parameters, providing more granular control over exported data. These new filters work in conjunction with the existing group and filter date-based keys to enable more targeted analytics exports. With this enhancement, you can now include the following optional filter in the API request payload:- eventTypes: Filter data by user feedback events such as ‘thumbsUp’ or ‘thumbsDown’.
New Connector Support in Search AI
Search AI introduces out-of-the-box connectors for the following platforms. These connectors enable seamless content ingestion from both cloud and self-hosted environments, expanding integration flexibility across enterprise systems.- Confluence Data Center
- xMatters
- Jira On-Prem
- JFrog On-Prem
- GitHub On-Prem
Enhancements to Existing Connectors
Search AI now offers improved capabilities through enhancements to existing connectors:- Zoom Connector: Added support for advanced filters to refine content retrieval.
- LumApps Connector: Now supports fetching and indexing attachments, enhancing search coverage and relevance.
v11.14.1 June 14, 2025
Patch ReleaseEnhanced Date Filtering for Jira Connector
The Jira connector now supports customizable date filters, offering users greater control over data ingestion. Users can now define custom date ranges for data ingestion as part of standard filters. Once configured, the connector automatically syncs and ingests data for the specified time frame, allowing for more targeted, relevant, and flexible data extraction based on specific requirements.v11.14.0 May 31, 2025
Minor ReleaseAPI Stage in Document Workbench
Search AI now features a new API stage in the Content Transformation process within the Extraction Module. This enhancement allows users to leverage external services for content transformation. Users can specify a custom POST API endpoint along with the necessary headers and request body to send extracted content for enrichment before chunking. This integration facilitates seamless collaboration with third-party tools, enabling the extraction, classification, and transformation of metadata and content.New and Improved Web Crawler
Search AI now includes a new and improved web crawler, designed for better performance and higher-quality content extraction. Key enhancements- Improved Performance with above 80% reduction in crawling time.
- Enhanced Stability and Reliability with complex web pages.
- Optimized JavaScript Crawling & Memory Management with reduced resource usage.
- Enhanced Sitemap Processing with asynchronous sitemap fetching and application of crawl depth limits and URL filters at the fetch stage for faster processing.
- Default Exclude Tags like header, footer, script, style, form, iframe, noscript, etc., to improve content relevance.
- Improved Error Handling in case of crawl failures and timeouts.
Dynamic Prompt and Model Selection in Answer Generation APIs
Search AI answer generation APIs (v1 and v2) now support optional fields in the request to specify the prompt and model for generating answers. This enhancement enables dynamic selection of LLM configurations, allowing users to tailor responses based on context, use case, or audience.Enhanced Multilingual Support
Search AI now offers expanded multilingual support with the addition of Spanish to enhance the customer experience. Users can:- Add and organize content in multiple languages.
- Understand and interpret queries in supported languages.
- Search and deliver answers and results in the same language as the user query.
v11.13.1 May 17, 2025
Patch Release This update includes only bug fixes.v11.13.0 May 03, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Image-Based Document Extraction
Search AI now supports image-based extraction for complex PDF documents using Vision embeddings. This feature is especially useful for rich-layout PDFs. Each page is converted into an image, and VDR embeddings are generated to capture the visual structure and textual content. This enables accurate and intelligent retrieval from visually complex documents, significantly improving search relevance. Currently, this capability is supported only for PDF files.Enhanced Connector UI
With this release, Search AI introduces a new and improved experience for configuring connectors, making it easier to manage data sources and providing complete control over the type of data ingested and processed, improving the connector configuration experience. Key enhancements- Streamlined Configuration: Set up connectors intuitively with improved navigation and layout.
- Flexible Field Mapping: Gain enhanced visibility and control over how source fields are mapped to the Search AI schema, use the post-processing script to customize mappings, and refer to a detailed field mapping guide that includes schema requirements and sample API responses.
Support for Simulator in Document Workbench
Search AI now includes a Simulator for Content Transformation Stages in the Extraction Module, enabling users to preview and validate how the configured stages affect the ingested data before vectorization. This helps ensure cleaner, more contextual, and optimized data for AI models. The simulator enhances data quality by allowing users to test and refine transformations, improving search results’ accuracy and relevance.New Search AI Connectors
Search AI extends support for two new connectors, enabling seamless content ingestion and retrieval from Teamwork and Opsgenie applications. These connectors enhance enterprise search by integrating knowledge from these platforms. The connectors also enable access control for the content that is ingested from the applications. In addition to the new connectors, the capabilities of several existing connectors have been expanded, enabling them to ingest additional content types from their respective platforms. This enhancement significantly increases the volume and variety of searchable content.v11.12.1 April 19, 2025
Patch Release This release includes some enhancements and bug fixes. The key updates are summarized below.Support for Sending Complete documents to LLM
Search AI now introduces Document-Level Processing, a new configuration that enables sending full documents, alongside relevant chunks, to the LLM for more accurate and context-rich responses. When enabled, Search AI identifies the most relevant chunks and automatically includes their complete associated documents, all within a defined token budget. This feature is especially beneficial for complex queries that require critical information to be distributed across multiple parts of a document. By providing the LLM with a more comprehensive context, it significantly enhances the quality of the generated answers.User Feedback Support
Search AI now includes a built-in feedback mechanism that allows users to rate the quality of answers by providing a thumbs-up or thumbs-down. This helps in capturing valuable user sentiment, enabling continuous evaluation and improvement of the responses generated by the system. Feedback can be submitted via the Web SDK or through the public API.v11.12.0 April 05, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Agentic RAG Architecture
Search AI now features an Agentic RAG architecture that leverages LLMs to improve query understanding and generate optimized retrieval queries for more accurate responses. This system uses autonomous agents to break down complex queries, enable preprocessing, and facilitate multi-step retrieval. The following agents process a user query sequentially to enhance it for effective retrieval and answer generation.- Query Rephrasing Agent: Refines the user query by understanding the context and intent, using the previous conversation.
- Query Transformation Agent: Identifies key terms and removes noise from the query for better retrieval.
- Result Type Classification Agent: Determines whether the user’s query requires a direct answer and search results or only a list of search results in the response.
- Metadata Extractor Agent: Extracts the source and fields from user queries and maps them to standard fields of the source, which helps to apply appropriate filters and enhance retrieval accuracy.
Exclude Stage in Document Workbench
Search AI now offers an Exclude Stage in the Document Workbench to restrict chunk generation from extracted content. This feature provides greater control over content processing, allowing users to exclude unnecessary content, thereby improving search relevance and reducing processing time.Enhanced SharePoint Integration with Restricted Permissions
To address security concerns and provide more controlled access, SearchAI now supportsSites.Selected permission for SharePoint integration. This limits access to only the specific site collections explicitly granted by administrators. This is implemented by introducing the OAuth Client credentials grant type auth mechanism for SharePoint Integration.
Channel Aware Response Formatting
Search AI now supports Channel-Aware Response Formatting to deliver a more engaging and seamless user experience across digital and voice channels. The default prompt has been updated to generate channel-appropriate responses that vary as per the mode of interaction. For voice channels, responses are concise and free of complex formatting, ensuring clear and effective communication. In contrast, digital channels benefit from well-structured responses with enhanced formatting for improved readability and user engagement.Enhanced Multilingual Support
Search AI now offers expanded multilingual support with the addition of the Ukrainian language to enhance the customer experience. Users can:- Add and organize content in multiple languages.
- Understand and interpret queries in supported languages.
- Search and deliver answers and results in the same language as the user query.
Secure One-Time URLs for Uploaded Documents
To enhance security, the citations or references in the search results and answers will now have one-time-use URLs, which expire after a single use or within 15 minutes, whichever occurs first. This feature applies to all uploaded documents and is enabled by default. It ensures controlled and temporary access to cited documents and aims to prevent the unauthorized sharing of URLs.New Search AI Connectors
Search AI extends support for eight new connectors, enabling seamless content ingestion and retrieval from PagerDuty, Figma, LumApps, Zoho CRM, TestRail, DataDog, Jenkins, and Zeplin. These connectors enhance enterprise search by integrating knowledge from various collaboration and productivity platforms. They also enable access control for the content ingested from the applications. Listed below are the connectors and the type of content that can be ingested from the corresponding applications.- DataDog - Metrics, Dashboard, and Monitors
- Figma- Figma Files
- Jenkins - Dashboard, Builds, Jobs, and Plugins
- LumApps - Pages, News, Custom Objects, and Community Posts
- PagerDuty- Ingests Escalation Policies and Schedules
- TestRail - Test cases
- Zeplin - Screens
- Zoho CRM - Leads, Accounts, Contacts, and Deals
v11.11.1 March 15, 2025
Patch Release This update include only bug fixes.v11.11.0 March 04, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Salesforce Connector Enhancements
The Salesforce Connector has been enhanced to support custom object ingestion in addition to the capability of ingesting default objects like Knowledge Articles, Cases, Documents, Opportunities, Leads, Contacts, Accounts, and Tasks. During configuration, users can select the type of content to be ingested, providing greater flexibility. Additionally, the filtering feature has been enhanced, allowing users to refine document ingestion based on specific fields.New Search AI Connectors
Search AI extends support for five new connectors, enabling seamless content ingestion and retrieval from Guru, Miro, Help Scout, Wrike, and Zulip. These connectors enhance enterprise search by integrating knowledge from various collaboration and productivity platforms. The connectors also enable access control for the content that is ingested from the applications.- Guru - Ingests knowledge cards for centralized access to organizational information.
- Miro - Enables searching the Miro boards.
- HelpScout - Enables search across knowledge base articles.
- Wrike - Provides access and searchability for tasks.
- Zulip - Integrates messages from channels into search results and answers.
v11.10.0 February 12, 2025
Minor Release This update includes enhancements and bug fixes. The key enhancements included in this release are summarized below.Custom Embeddings Support for Enhanced Vector Generation
The Custom LLM feature now supports Vector Generation in Search AI, allowing users to leverage custom embedding models for improved accuracy and relevance. Users can define custom requests, pre-processors, and post-processors, specify input and output keys, and seamlessly integrate with existing features. Search AI triggers vector generation during indexing and user query processing, while the document browser displays embedding status and allows filtering. This enhancement provides greater control, customization, and monitoring capabilities for delivering tailored search experiences.Custom Extraction
The new Custom Extraction feature in Search AI enables a tailored approach to extracting content from sources. It sends ingested content to a third-party service, which processes and returns the extracted data in a structured chunk format. This enhances flexibility in handling diverse content extraction needs and ensures improved data retrieval and indexing within Search AI.Answer Insights
The new Answer Insights feature provides comprehensive data about each query-response interaction, enabling a better understanding of system performance and easier troubleshooting. Key features- View a list of all answers for a grouped query.
- Search logs by answer and channel.
- Filter logs based on channel
- A detailed view of each answer, including query overview, answer debug information, and LLM request-response details (if available).
Search Results and Facets
The new Search Results feature, along with faceted search capabilities, allows retrieving and displaying multiple search results ranked by relevance for each query. Key changes- Search Results now use the same pipeline as Answers for improved accuracy and consistency.
- Search Results Configuration - Enable/disable the feature and set the number of search results to display.
- Faceted search offers default filters, customizable filter creation, and static and dynamic filter types.