This is a preview page for the upcoming 2025.40 release in order to announce the necessary infrastructure changes early before the release as can be seen in .

📌 Release Overview

• TBA

• Unique hosted: October 2, 2025

• Self hosted: October 3, 2025

Chat Platform

🔧 Improvements

User Display sub agent answers in the main agent: Sub agent answers can now be displayed before the main agent’s response. Assessments (such as hallucination checks) from sub agents can be consolidated and shown in the main agent. See the documentation for more details on how to configure this.

User Favicons for external website references: The globe icon formerly used for external website references has been replaced with the website’s favicon.

Admin Hide Sub-chats & Sub-agents created by agent to agent communication: Chats automatically inherit this behavior whenever they are started from an assistant that has its “Sub-agent” status enabled. Chats from agents created as sub-agents are no longer visible to users.

This setting can also be turned on from the “Advanced Settings Configuration” of custom spaces

Admin Space 2.0 Product Metrics Export: The feature allows admins to export Product metrics as CSV. Zitadel role chat.feedback.read is needed to be able to export Product metrics

🐞 Bug Fixes

User Service Users: Unsynchronized service or machine users in Zitadel are now displayed with clear, meaningful names.

Admin Reference stat analytics export failing: Customers with large content and scopes could not export reference analytics. We’ve improved the export algorithm there and affected customers can now export the data

Admin Group search in Space Share: The search now supports groups (e.g. “All company”) on the share modal window.

Admin Spaces 2.0: advanced settings theming fix: Resolves the issue where the space advanced settings was not respecting the configured theme.

Admin Mistral models tool calling error: Tool calling and message history structure were broken earlier. A fix has been deployed along with a version update in unique_toolkit.

RAG

🔧 Improvements

User Microsoft Document Intelligence (MDI) ingestion performance: Improved the PDF ingestion performance with MDI by changing the processing method from a page-by-page approach to a batch-based approach. PDFs are now split into batches containing multiple pages each. Each batch is processed collectively by MDI, resulting in faster ingestion and more efficient handling of large documents.

User Recalculate sharing toggle position based on folder selected: We added a privacy toggle for new folders that automatically matches the parent folder’s setting. By default, new folders inherit their parent’s privacy (Private or Shared). If a user changes the parent folder during creation, the toggle recalculates in real-time to reflect the new parent’s setting, while still allowing manual adjustments.

🐞 Bug Fixes

User CSV Ingestion: Resolved an issue with CSV ingestion where data was being imported as plain text rather than structured tables. The fix ensures that CSV content is now properly interpreted and imported as tables, preserving the correct structure and improving the accuracy of the resulting data chunks.

User Microsoft Document Intelligence (MDI) Ingestion: Resolved an issue with MDI PDF ingestion where HTML tables were not being displayed correctly. The fix ensures that tables are rendered as intended, without any manipulation, which improves the accuracy and integrity of the data chunks generated during ingestion.

User Fix incorrect uploader attribution for files in a folder: Resolved an issue where re-uploading a file with the same name caused all files in the folder to display the new uploader’s name. Now, only the re-uploaded file correctly reflects the uploader, while existing files retain their original attribution.

Agentic Table

🔧 Improvements

User Repositioned About and Warning messages: About and warning messages on the sheet overview page now appear above the table list to provide immediate visibility to the user.

🐞 Bug Fixes

User Replace failed sheet error with guidance banner for new sheets without questions: Users can now successfully create a sheet when providing just a sheet name and sources. Users will now receive contextual guidance banners instead of an error about missing question files.

User Failed ingestion files no longer selectable for import: Files with Ingestion failed status in the Knowledge Base can no longer be selected or counted during sheet creation, ensuring accurate file counts and consistency.

User Resolved assistant access error under concurrent executions: Running the agent on multiple accounts at the same time no longer triggers the User does not have access to assistant error, ensuring stable and correct access handling.

User Resolved cell edits outside columns in library eliciting an error on verification status change: When a cell is edited in the library sheet outside of the specified columns and then the status of the row is set to verified, there is no longer an error in ingestion.

Retirement Notice

Admin The GPT-3.5 model will officially retire on 11 November 2025. We recommend migrating now all use cases currently running on GPT-3.5 to GPT-4o-1120 (AZURE_GPT_4o_2024_1120). Clients will have two weeks to validate the impact and adjust prompts if needed. After this period, all remaining spaces and templates using GPT-3.5 will be migrated to GPT-4o-1120 as part of Release 2025.42.

New Anthropic Models

Admin New Anthropic models are now available: the Claude Sonnet 4.5 model and the Claude Opus 4.1 model. Both can be accessed through LiteLLM using the following enums:

• litellm:anthropic-claude-opus-4-1

• litellm:anthropic-claude-sonnet-4-5

Unique AI

🔧 Improvements

Admin Crawlers:

• External crawlers (Firecrawl, Tavily, and Jina) are now marked as Experimental. These crawlers require an API key to be enabled (not provided by default, as your consent is required). If you’d like to test one of them, please coordinate with Unique.

• Expose the list of unwanted URL Content/type. This would help in having a quick control for the URLs you don’t want to crawl. Web Crawler Configuration