You can also browse, search and filter all release notes directly in the application → Open release notes in TAI.
Agents can now show polished, dashboard-quality charts inline in chat
New Feature.
When you ask for a chart worth showing someone, an agent can now render a polished data chart right in the conversation — crisp, on-brand, and adapted to light or dark mode. It builds on the fast charts from the last release: the agent still just supplies the numbers, so a dashboard-quality chart appears in seconds, not minutes.
Quick inline charts stay the default; the polished renderer is used when you want the extra polish (for example, a chart for a report or a summary).
Same on-brand colours across every chart, so a conversation reads as one product.
Takes effect for chats started after this release.
A polished donut chart rendered inline in chat — each slice labelled with its share, adapting to light or dark mode.
New built-in agent: Traveller Locator finds who's in a crisis zone
Major New Feature.
When an incident hits a destination, the Traveller Locator agent answers the first question — who do we have there? Give it a destination and a date and it returns your confirmed bookings with travellers in that location, sorted by who leaves soonest.
Confirmed bookings only — one row per booking, no drafts, no cancelled itineraries.
Reply
detailand it drills into every booking at once for lead passenger and co-travellers, contact email and mobile, the hotel they're in right now, and first and last flight with PNR.A priority flags table calls out who needs attention first: no contact on file, minors travelling without a contact, accommodation gaps.
Every row links straight into Cockpit customer care so you can act without searching twice.
Ask for the list as a file and you get one: Excel, Word, or a printable HTML report to hand to whoever runs the response.
No setup required — Nezasa owns and maintains the agent. Find it in the agent picker under Chats. Rolling out first to companies in the preview program.
Decide which websites your agents may search and read
New Feature. Only Company Admins.
Agents with the web permission can now search the public web and read pages, and Settings has a new Web access page where you set the rules for the whole workspace.
One switch for the workspace. Turn web access off and no agent can search or fetch, whatever its own permission says.
Allowed sites. Leave the list empty for the open web, or name the sites your agents may reach.
Blocked sites. Name the sites they may never reach. A blocked site wins over an allowed one.
Site matching is exact, so iata.org does not cover data.iata.org: list each host you need. Nezasa additionally blocks a short list of link shorteners for every workspace, shown on the same page. When a site is refused the agent is told why and asked to use a different source, so a blocked link does not sink the whole task.
Ask the Ops & BI Agent how much agency commission you paid
New Feature.
The TripBuilder Ops & BI Agent can now report agency commission across your bookings from the data warehouse — questions like "how much commission did we pay on bookings made last month?" or a breakdown by destination or agency are now answerable as a single number, trend, or table.
Each itinerary carries its current commission as one figure (in EUR), so period totals are accurate — no double-counting across booking changes.
Booking and cancellation commission are reported separately when you want the split (cancellation commission is the commission computed on the cancellation cost).
For the commission on a single booking, the agent still drills into that booking's live detail rather than the warehouse.
Keep editing your agent while the Prompt Advisor is working
Improvement.
While the Prompt Author & Advisor runs — a Draft or Improve can take a minute or two — the rest of the agent editor no longer locks up. The progress dialog stays visible so you can cancel at any time, but it no longer traps focus or blocks the fields behind it, so you can keep working while you wait.
Prompt Advisor now points you to the built-in agent for a capability it can't attach
Improvement.
When you draft, review, or improve an agent — or ask the Prompt Author & Advisor for skill advice — it no longer tells you a capability is missing when it actually exists as a built-in.
Some skills are reserved to a built-in agent and can't be attached to a custom agent you're building. Previously the advisor simply didn't see those, so it could claim the capability wasn't available. Now it names the capability and points you at the built-in agent that provides it — so if what you're building already exists, you'll hear about it up front instead of hunting for a skill you can't attach. Reserved skills are still never suggested as an attachment.
Steer Draft with AI and Improve with your own instructions
Improvement.
Draft with AI and Improve used to run straight off the agent's Description field with no way to guide them. They now ask you what you want first.
Draft with AI asks what the agent is for — describe the agent in its own step and the advisor drafts from that, independent of the Description field.
Improve takes optional suggestions — add a steer like "make it stricter about refunds", or leave it blank to run the automated review exactly as before.
You can see the advisor working, and stop waiting — an advisor turn takes a minute or two, so it now runs in a visible step you can cancel instead of leaving you guessing whether anything is happening.
Draft with AI now asks what the agent is for and drafts from that — independent of the agent's Description field.
Improve lets you add optional suggestions to steer the rewrite, or run it automatically by leaving the box empty.
While the advisor works you can see which action is running and stop waiting at any point.
Links in Chats and Reports now open in a new tab
Fix.
Following a link an agent gave you — an itinerary deep link, say — no longer navigates you out of the conversation or report you were reading.
Undo in the Prompt Author & Advisor survives a follow-up Review
Fix.
After you apply a drafted or improved system prompt in the Prompt Author & Advisor, the Undo now stays available even if you then run a Review or ask for Skill advice.
Previously any follow-up advisor run cleared the Undo — and because applying a prompt also replaces the editor's text programmatically, the browser's own Ctrl/Cmd+Z couldn't bring it back either, so the original prompt was lost. Now only a run that proposes a new prompt (Draft or Improve) replaces the Undo; Review and Skill advice leave it untouched.
Travel alerts now include incidents already under way when the trip starts
Fix.
Asking the Passolution agent or Trip Guard about a travel window could return an all-clear while alerts were active in that destination. Only events that began inside the window were counted, so an ongoing wildfire, strike, or closure that started days earlier was left out — along with open-ended events that carry no end date at all.
Travel dates are now treated as a window: every event active at any point during it is reported, whenever it started. Results also page to the end rather than stopping at the first 25 events, and an incomplete check is flagged as incomplete instead of reading as 'nothing found'.
Reports: a run that outlives your session now returns you to sign-in
Fix.
When a report run crossed a session expiry, it used to surface an error on a session that was already dead — leaving you re-clicking Run with no way forward but a manual reload. A report run now sends you to the sign-in page when your session has expired, the same as everywhere else in the Workbench.
Documentation section links no longer silently go nowhere
Fix.
Links that jump to a section further down a page — like the ones in the Chat UI SDK developer docs — could quietly do nothing.
In the Help Center they never could work: it rewrites every heading's identifier when an article is published, so a link and its destination can never match. Those now read as plain text rather than inviting a click that goes nowhere.
In the documentation inside TAI they now work. Section identifiers follow the same rule the docs use everywhere else, which repaired two links that were landing at the top of the page. A third link pointed at a section name used twice on the same page, so it could have started opening the wrong one at any time — that section was renamed. A new check now keeps every section link honest from here on.
My Account no longer reports a saved profile change as failed
Fix.
Saving your name — or, for Nezasa staff, the acting-as tour operator — could show Failed to update profile even though the change had gone through, sending you back to redo an edit that had already been saved. The save and the follow-up refresh are now reported separately, so the confirmation reflects what actually happened.
Company admins: language settings apply without a page reload
Fix. Only Company Admins.
Turning a language off left the header switcher, the invitation Language menu and My Account still offering it until the page was reloaded. Saving now updates them in your own session immediately — other people already signed in still pick the change up on their next page load.
Emails now follow your company's enabled languages
Fix.
Turning a language off in Settings → Languages switched the Workbench back to English but left emails arriving in the old language. Every email now checks the current list before it sends. Your choice is kept, not erased — re-enable the language and your emails go back to it. The language picker when inviting a user now offers only the languages your company has enabled.