You can also browse, search and filter all release notes directly in the application → Open release notes in TAI.
Chats and Reports now say when content is AI-generated
New Feature.
Chats: every chat says you're chatting with AI, in all 14 languages: under the empty-state message and as a caption under the message box. Chats embedded through the Chat UI SDK show it once the host app updates to the SDK's next version; embedders can reword it but not remove it.
Reports: the viewer, the library and the standalone view label every report "AI-generated". Exported PDF and HTML files, including the PDF attached to scheduled report emails, end with a footer saying so in your language.
If you publish a report as-is on a matter of public interest, keep these labels in place.
Learn more in Using chats, Using reports and EU AI Act.
The AI notice under the message box of a chat.
A report labelled AI-generated next to the agent's name.
Agents: scannable prompt advisor feedback, plus one-click Add all skills
Improvement. Only Company Admins.
Company admins can now scan the prompt advisor's feedback and act on its skill suggestions faster.
Findings are short, one problem each, with a brief summary.
Suggested skills say Add skill (or Remove skill) instead of Apply, and Add all skills adds every suggested skill in one click. Removals stay one at a time.
Billing: budgets without early alerts now explain their On track badge
Improvement. Only Company Admins.
A budget with no alert threshold below 100% never shows Warning, so it reads On track until it's used up. Hover the badge to see why.
Chats accept photo-heavy PDFs and explain when a file is too large
Improvement.
PDFs are sized by page count, not file size, so a photo-heavy scan now goes through. A PDF of more than about 60 pages is too large for one conversation; split it into parts.
CSV and text files are sized the way the model counts them. A CSV over about 300 KB or a text file over about 500 KB is refused up front.
A file too large for one conversation now gets a clear reason instead of "conversation too large".
Reports: failure, capacity and input messages now follow your language
Improvement.
A failed report run, a launch refused because too many reports are already running, or a rejected input (a missing field, a bad date range, an unrecognized upload) now renders in your Workbench language instead of always English — matching how a report's own content already localizes.
Passolution Agent answers now link the official entry and visa pages
Improvement.
The Passolution Agent's entry and visa answers now link straight to the official page behind the claim — an e-visa application form, an entry-registration portal — copied exactly as Passolution provides it, never constructed or guessed. Where Passolution gives no link, the answer names Passolution as the source instead of showing one that doesn't work. Learn more.
Chats: open an HTML report right in the conversation instead of downloading it
Improvement.
Clicking an HTML report card in a chat now opens it in a near-full-screen preview right there in the conversation, instead of immediately downloading it. Download and Export to PDF (when available) stay one click away from the preview's header, and a small download icon next to the card keeps the raw file just as quick to save as before. Learn more.
Click an HTML report card to open the report right in the chat
Chat, Reports and Prompt Advisor errors now appear in your language
Fix.
When a chat response, a report run or the Prompt Advisor fails, times out or arrives incomplete, the notice is now shown in your language instead of a fixed English line, and screen readers announce chat error notices. The standard report input of built-in agents is now labelled in your language too, on the agent's page and on the run and schedule forms.
Agents that follow the company default no longer fail on Balanced
Fix.
Agents following the company default could fail every chat turn on Balanced. The model list now only offers models that are actually available to you, and newer ones appear once they are.
Prompt Advisor Apply no longer gets stuck after an interrupted apply
Fix. Only Company Admins.
An apply cut off mid-way now settles itself: Apply comes back if the change didn't land, and stays done if it did.
Prompt Advisor runs keep updating while you work in another tab
Fix. Only Company Admins.
A finished run now shows as finished when you come back, not Queued. If updates fail, the notice offers Refresh instead of a page reload.
Prompt Advisor runs now say why items were left out
Fix. Only Company Admins.
A run's detail page no longer blames every left-out item on a malformed reply: items that were well-formed but could not be applied get their own notice.
Deleted chats no longer reappear in Chats or on the Dashboard
Fix.
A deleted conversation now stays gone everywhere. The Prompt Advisor no longer lists it, and no longer evaluates it even if it was picked before it was deleted.
Settings and Prompt Advisor: a mistyped detail URL now opens its list
Fix. Only Company Admins.
A Users, API Keys or Prompt Advisor run URL with a malformed id — a typo, or a list-style path like …/all — now returns you to the list, not an error.