Release 2025.36

This is a preview page for the upcoming 2025.36 release to announce the necessary infrastructure changes early before the release as can be seen in 2025.36 Infrastructure Changes.

📌 Release Overview

• v 2025.36-48df6
• Unique hosted: September 3, 2025
• Self hosted: September 8, 2025

Chat Platform

🔧 Improvements

• Admin Enhanced form validation in the tools settings for easier configuration.
• Admin Terms and conditions functionality will be disabled if not set, simplifying user experience.
• Admin You can now choose which model to use in Unique AI for better customization.
• Admin The “Select files” button is now only visible in Unique AI, streamlining the interface. It can still be configured for DocumentSearch modules if needed.

🐞 Bug Fixes

• tech Tool Choices events are now correctly included in external module event payloads.
• User Resolved reliability issues with assistant access for user counts exceeding 1000.
• User Fixed the issue preventing pasting into the chat input on mobile devices.

RAG

🔧 Improvements

• User Added support for latest OpenAI embedding models: We’ve added support for the latest OpenAI embedding models, text-embedding-3-small and text-embedding-3-large. A new embdeddingDimension parameter allows customizing embedding size, enabling trade-offs between accuracy, performance, and storage.
• User Changed Chunk metadata view: The metadata in the Chunk view page is now displayed in form of the tags and replace the previous presentation as a simple json object.

🐞 Bug Fixes

• User Improved chunking of .txt files: We’ve improved how .txt files are processed during ingestion by making chunking token-aware. Instead of splitting .txt files only by character length, chunks are now sized based on tokens. This ensures more consistent chunk sizes, prevents “payload too large” errors, and makes downstream processing more efficient and reliable.
• User Fixed issue with copy-protected files: Some copy-protected could not be ingested anymore. With the latest change, these files can be ingested again.

Agentic Table

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

• User Added Owner display to overview and Agentic Table pages: The owner’s name and icon now appears in the overview table and on the Agentic Table sheet page.

🔧 Improvements

• User Improved performance of row status transitions in Agentic Table: Optimized status updates (e.g., Needs Review → Verified) so changes persist faster and the UI updates more smoothly.
• User Grouped file import buttons into a single dropdown in Agentic Table: All file import options are now available under a single “Add/Import” dropdown, making the menu bar cleaner while keeping existing functionality:
  • Add Sources from Knowledge Base – Select files from the Knowledge Base to be used when generating answers in the Agentic Table.
  • Add Single Question – Add a new question to the sheet or choose from predefined questions.
  • Import Questions
    • From Knowledge Base – Import questions from files uploaded in the Knowledge Base.
    • Upload Files – Upload files directly to add their contents as questions.

🐞 Bug Fixes

• User Fixed Agentic Table report generation failure caused by text in non-configured columns: Fixed an issue where adding text in columns beyond the valid configuration caused DOCX report generation to fail.
• User Fixed error when adding answers to rows without questions in Agentic Table: Users can now enter content in the Answer column of a row even if the Question column is empty, without triggering an error.
• User Ensure “Updating” label shows for all background edits in Agentic Table: Fixed an issue where edits to certain cells (e.g., Department or Section) triggered a background update but did not display the “Updating” label in the UI.
• User Fixed display issues with the first row in Agentic Table sheets: The first row now always shows correctly, preventing missing or duplicated rows when creating or opening a new sheet.

Unique AI Chat

Admin Model Selection: Supported LLMs for Unique AI can be selected directly on the main configuration page via a dropdown lost. Only models available for the respective client are displayed. Additional and custom models can be configured in the advanced settings.

Admin Smarter Token Usage: Token usage by the orchestrator and tools is now automatically adapted to the selected language model’s context window. Models with larger windows include more search context in each step, making full use of their capabilities. The percentage of tokens used for internal and web searches remains configurable in the settings.

User Chat-uploaded documents prioritised: Documents uploaded into the chat are always processed in the first orchestrator iteration and included in the response context.