Feature Updates
- Administrators can pause new AI sandbox sessions during maintenance, with an optional notification, while allowing existing conversations to finish. When the service is unavailable, affected AI features show a clear notice before users submit requests.
- Agents can select a model tier when creating or updating scheduled tasks. Existing selections are retained unless explicitly changed, and subsequent messages use the selected tier.
- Event-triggered scheduled task chats receive titles that reflect the triggering event without overwriting titles edited by users. Run details also show why a scheduled task started.
- Sign-in protection is now enabled by default and locks accounts after consecutive failed sign-in attempts. Administrators who do not want account lockout must explicitly set
SIGNIN_ACCOUNT_LOCKOUT_ENABLED=false. - Free workspaces now receive a one-time welcome award of 200 credits instead of 200 credits each month. Existing free workspaces receive the one-time award, while new accounts receive it for their automatically created workspace.
- After a Business plan expires, custom domains remain available during a one-week grace period. Owners are notified when the plan expires and when domain resolution is disabled; renewing during the grace period keeps the domain available.
Bug Fixes & Improvements
- Fixed an issue where chained calculation formulas failed when blank text results were treated as numbers. Blank text now remains blank instead of becoming
0. - Comments on deleted records are no longer accessible. When a record is permanently deleted, its comments and subscriptions are also removed.
- Shared grid views and expanded record cards now show all visible fields, including columns that have not yet been scrolled into view.
- Fixed an issue with Base editing permissions when a user's space role outranked their department's Base role. Record visibility and field permissions are also more consistent in wide tables with multiple permission roles.
- Search results for users with multiple permission roles now respect both the search term and row-level permissions. For large tables, searches may pause if there is no safe scan path or usable index, until a table administrator configures searchable fields.
- Fixed inconsistent display of large numbers in cells, record details, copied values, and API results. Small Number and Percentage fields now preserve visible digits beyond JavaScript's safe integer range.
- Summary fields now recalculate for both the original and new linked records when relationships change. Refreshes of dependent fields are also more reliable after relationship conversions.
- Improved manual row ordering in large grids, including during concurrent changes and when records are restored. Creating or copying a view no longer performs unnecessary full-table sorting, and copied tables retain correct sorting behavior.
- Fixed an issue where a table could be hidden but not deleted after deletion failed, and improved recovery when deletion fails partway through.
- Fixed an issue where moving and reordering tables in the sidebar did not persist after a refresh. Failed table entries that cannot be used no longer appear in the sidebar; import errors remain visible in the import flow.
- Fixed an issue where statistics in nested grouped views could remain stuck, particularly when using date-based formula fields. Changing a field's summary method no longer causes unrelated results to flicker.
- Fixed an issue where linked records with titles briefly appeared as “Untitled” in linked-record cells after bulk updates.
- Improved the reliability of large CSV and Excel imports, reducing import failures and temporary errors in related requests.
- Fixed blank thumbnails for SVG attachments in grid views.
- Improved AI chat queue visibility, message ordering, and recovery after refreshes or reconnections. Queued messages immediately show a pending status and are removed if rejected.
- AI conversations render more smoothly during streaming. Base and space pages load certain previews and selectors only when needed, improving load speed.
- Improved AI field autofill performance during large batch operations and fixed generation failures for some existing fields using space-level custom providers.
- Improved App Builder recovery after preview restoration fails or a workspace becomes unavailable, reducing the likelihood of imports and publishing remaining blocked.
- BYODB space usage and row-limit checks now use real-time row counts, reducing inaccurate totals and incorrect quota decisions.
- The organization user API now returns only the departments to which the specified user belongs. Integrations that need all departments should use the dedicated department list API.
- Improved account, license, integration, and sign-in security. Adding a space already linked to an organization no longer returns an unnecessary error.