Release 2026.26

5 min read

📌 Release Overview

  • 2026.26

  • Unique hosted:


Chat Platform

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

Admin Experimental Model Usage Limits: Administrators can now set daily spending limits on model usage, giving organisations control over how much individual users can spend per day. Navigate to Settings → Cost Management → Limits to configure limits at three levels:

  • Company default — a fallback cap that applies to all users with no other limit set

  • Group limits — per-group caps; if a user belongs to multiple groups with limits, the highest one applies

  • User limits — per-user caps that always take priority over group and company limits

Limits are denominated in USD and reset automatically each day at midnight in the user's local timezone. When a user hits their limit, further model calls are blocked until the next day.

Read more here: Model Usage and Cost Management .

This feature is disabled by default and must be enabled via feature flag FEATURE_FLAG_MODEL_USAGE_LIMITS_UN_18889. Note that the the feature also depends on the following feature flags FEATURE_FLAG_SAVE_MODEL_USAGE_UN_12832 and FEATURE_FLAG_SAVE_MODEL_USAGE_DASHBOARD_UN_18889.

image-20260614-210047.png

🔧 Improvements

User Remove uploaded files: Uploaded files can now be removed from a chat at any point before sending including while they're still being ingested rather than being stuck once attached. Hover over the file and click the remove icon to take it out before sending your message.

User SwappableIntelligence space icon: SwappableIntelligence spaces now display a dedicated waypoints icon instead of the generic Unique logo.

User Experimental Wider tables and charts in chat: Tables and code execution charts in chat responses can now expand beyond the chat message width (up to the full width of the window) so wide tables and graphs are easier to read. The chat message width itself stays the same, and anything wider than the window remains horizontally scrollable.

This feature is disabled by default and must be enabled via feature flag FEATURE_FLAG_ENABLE_FULL_WIDTH_SCROLL_UN_19453.

The wider rendering for code-execution charts/images additionally requires code execution and its fence rendering to be enabled (FEATURE_FLAG_ENABLE_CODE_EXECUTION_UN_17498, FEATURE_FLAG_ENABLE_CODE_EXECUTION_FENCE_UN_17972). These two feature flags are not available in the feature flag management service and must be turned on as environment variables. Contact the Unique team for support turning these on.

🐞 Bug Fixes

User Code-execution card scroll jump: Fixed a visual jump that occurred when inline HTML code-execution cards finished loading — the card now shows a smooth skeleton shimmer until the iframe content is fully ready, eliminating the layout shift.

Admin Analytics assistant dropdown empty: Fixed an issue where the assistant dropdown on the analytics page appeared empty when data was served from the browser cache.

Admin Banner notification date validation: The banner notification date pickers now prevent selecting past start dates and invalid date ranges, allow same-day start and end dates, and normalise times to start/end of day.

Admin Banner notification date normalisation: Creating a same-day banner was flaky because dates were not normalised. Start dates are now set to the beginning of the day (00:00:00) and end dates to end of day (23:59:59), ensuring banners display reliably for the full intended period.

RAG

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

Admin Content Metadata Filter in Advanced Scope Rules Supports List Intersection: List intersection is now supported when using scope rules to filter content using an overlaps / notOverlaps operator which can take a list of values and return a match if there is any intersection with the content metadata field.

image-20260623-134530.png

🔧 Improvements

User Reject slashes in folder names: Folder names can no longer contain / to avoid ambiguity with folder paths. Users will now see validation before saving unsupported names

Agentic Table

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

User Experimental Can Answer (Assignee) role: Introduces a new share role for Agentic Tables that scopes access to assigned rows only. Can Answer users see and edit cells, status, and assignee on their own rows, can comment, give feedback, export, and share at or below their access level — but cannot re-run AI, add or delete rows or sources, edit sheet metadata, or view other users' rows. Restrictions are enforced in both the UI and on the backend. This feature is behind the feature flag FEATURE_FLAG_AGENTIC_TABLE_CAN_ANSWER_ROLE_UN_21308 and must be enabled for your organization. See the Agentic Table role permissions matrix for the full breakdown across all roles.

Admin Experimental Optional sheet metadata form: Space admins can configure a custom metadata form that appears when users create a new sheet. Input fields – free text, dropdown, and date – capture structured context (e.g. fund, investor, reporting date, or frequency) up front. All fields are optional, and entered values are saved as sheet metadata which is editable and visible in the table instructions dropdown. This feature is behind the feature flag FEATURE_FLAG_AGENTIC_TABLE_METADATA_FORM_UN_17903 and must be enabled for your organization.

See the Sheet Metadata Form documentation for setup details.

Screenshot 2026-06-22 at 15.33.16.png

🔧 Improvements

User Answer Library file naming: When rows from an Agentic Table sheet are ingested into the Answer Library, files are now named {OriginalSheetName}_{YYYYMMDD}_{HHmm}_{index} (e.g. DDQ_Q1_2026_20260520_1430_1). This ties each library entry back to its source sheet, shows when it was ingested, and keeps names unique when the same sheet is ingested multiple times. Existing library entries keep their current names.

User Auto-adjusted Excel export columns: Exported Excel files now auto-size column widths to fit their content again. A recent change that stripped HTML tags had left columns at a fixed width, making longer answers harder to read.

🐞 Bug Fixes

User Duplicate row positions causing ERR cells: Fixed an issue where duplicate (sheetId, rowOrder) rows in the agentic-table library caused ERR cells to appear in the UI when multiple rows shared the same position.

User Hidden Excel content excluded from question extraction: When importing questions from an uploaded Excel file, hidden sheets, rows and columns are no longer pulled in by default as spurious questions. This requires the config flag use_xlsx_preprocessing (see Default Module Configuration -> Question Extraction) to be enabled on the space. To include hidden material, set extract_hidden_xlsx_fields to false.

User Comments on empty cells: Prevent users from leaving comments on empty cells beyond the table's column boundaries, which previously succeeded silently with no visible indication.

Admin Chat disabled bypass on Agentic Tables: Fixed a security issue where the API still accepted chat requests and allowed data to be written to a sheet after the "Can chat" permission was disabled in Space Management. The UI correctly blocked interaction, but direct API calls could still inject prompts and update sheet content; access is now enforced on the backend when chat is turned off.

Unique AI

🐞 Bug Fixes

User Fix Markdown LaTeX Rendering: Resolved an issue where currency values wrapped in Markdown formatting (for example, $10 to $20) could be incorrectly detected and rendered as inline LaTeX formulas.

Last updated