This is a preview page for the upcoming 2025.42 release to announce the necessary infrastructure changes early before the release as can be seen in .

📌 Release Overview

• TBA

• Unique hosted: October 14, 2025

• Self hosted: October 20, 2025

Chat Platform

🚀 What’s New (Business/End-User Focus)

User Steps: Steps generated during execution of the user’s prompt are now visible if the message log record is populated during the execution.

Admin Language Model Whitelisting: We’ve introduced a new GraphQL endpoint that allows organizations to whitelist language models either company-wide or for specific users. This feature ensures tighter compliance and data protection controls over which models can be accessed.

type mutation {
  createManyEntityLanguageModel(input: [EntityLanguageModelCreateInput!]!): Boolean!
}

How It Works

• Once configured, only models included in the whitelist will be accessible to the company or specified users.

• If a whitelist is set for a specific user, that user will not inherit the company-level whitelist — enforcement is handled independently.

• Permissions Required:

  • ADMIN_SPACE_WRITE

  • SPACE_MANAGER

Important Notes

• This system is a whitelist, not a blocklist.

• Once any EntityLanguageModel is configured, whitelisting becomes active.

• If you plan to whitelist models for specific users, ensure there’s also a company-level whitelist to prevent access issues.

🔧 Improvements

User Assistant Message Updates from SDK: The UI now automatically refreshes when assistant messages are updated through the SDK.

This ensures users always see the latest assistant responses in real time — without needing to manually reload or refresh the page.

Why It Matters

• Provides a smoother, more dynamic user experience.

• Keeps the assistant interface synchronized with backend message updates.

• Reduces manual refreshes and potential confusion from stale content.

🐞 Bug Fixes

User Banner notifications: Banner colors now persist across updates and after reset or dismissal actions. This ensures visual consistency and preserves your intended design or status indicators, even when banners are refreshed or reinitialised.

User Chat Knowledge Scope: Resolved an issue where previously selected files were automatically reloaded into the chat window when returning to the same conversation.

Outcome

• Files now remain cleared unless intentionally reattached.

• Prevents confusion and accidental resubmission of old files.

• Ensures a cleaner, more predictable chat experience.

RAG

🔧 Improvements

Admin Making private folder creation the default in the Knowledge Base: Previously, newly created folders were Shared by default. We introduced a new environment variable (see Infrastructure Changes) that allows changing this behavior so that new folders are Private by default instead.

User Excel-Ingestion: Improved Excel file processing to support embedded form data (e.g., dropdown selections, checkbox responses, or text input control). With this update, the ingestion system now detects and extracts values from embedded form fields in Excel sheets, ensuring all user-entered data whether typed directly into cells or entered via form controls is accurately captured.

Example:

• A form-based Excel file where users select a department from a dropdown or enter comments in text boxes will now have those selections and comments correctly ingested into the system.

🐞 Bug Fixes

User Excel ingestion breaking with large text cells: Fixed an issue where Excel ingestion failed if a cell contained too much text, causing embedding requests to exceed their limits. To preserve the full row context, the solution duplicates the affected row and splits the oversized cell content into two parts.

Admin View Chunks page crashing: Fixed an issue that the View Chunks page was crashing for some files.

Agentic Table

🔧 Improvements

User Removed Answer Agreement Status icons: The check mark icons for Answer Agreement Status have been removed for a cleaner, simpler interface, while backend logic for answer matching remains intact for future enhancements.

User Highlight and copy enabled in history panel: Users can now select and copy text from the information history side panel in the Agentic Table.

🐞 Bug Fixes

User Added questions now appear correctly in focused cell: When using Add Single Question, the entered question is now properly inserted into the focused cell without being overwritten or causing agent errors.

User Answer cell edits now persist when assigning users: Edits made in the Answers column are now properly saved even when an assignee is added to the same row.

Unique AI

🔧 Improvements

User Sub-Agent References: References generated within sub-agents are now properly transferred to the main agent and included in the final answer. This improvement ensures that all supporting sources and citations from sub-agents are accurately reflected in the overall response. See the documentation for more details on how to configure this.

🐞 Bug Fixes

User Image Upload: A bug has been resolved where chats with uploaded images in Unique AI did not trigger follow-up question generation. Additionally, previously uploaded images are now correctly retained and taken into account during ongoing chat interactions.

Language Model

GPT-3.5 Retirement

Admin Due to the upcoming retirement of GPT-3.5-Turbo-0125, all spaces and AI templates previously using GPT-3.5-Turbo-0125 have been migrated to GPT-4o-1120 (AZURE_GPT_4o_2024_1120).

Retirement Notice

Admin The o1-mini model will officially retire on 17 November 2025. We recommend migrating now all use cases currently running on o1-mini(AZURE_o1_MINI_2024_0912) to gpt-5-mini (AZURE_GPT_5_MINI_2025_0807). Clients will have two weeks to validate the impact and adjust prompts if needed. After this period, all remaining spaces and templates using o1-mini will be migrated to gpt-5-mini as part of Release 2025.44.