You can also browse, search and filter all release notes directly in the application → Open release notes in TAI.
Collapse the navigation sidebar for more workspace on wide screens
New Feature.
You can now fold the left navigation into a slim icon rail to give wide surfaces — Chats, Reports, and data tables — more horizontal room. Use the Collapse sidebar control at the bottom of the navigation to switch between the full sidebar and the icon-only rail; your choice is remembered in your browser and restored on your next visit. While collapsed, hovering over the rail slides the full sidebar back out temporarily, so you can jump to another section without expanding it again, and each icon shows its label on hover.
Passolution Agent now in preview — travel alerts, entry & visa rules
Major New Feature.
The Passolution Agent answers both live travel events and standing travel regulations in one chat: current alerts for a country, airport, or region, plus entry, visa, and health requirements for a destination and traveller nationality — with itinerary lookup by TripBuilder booking reference and downloadable HTML reports.
Travel events and alerts are included with TAI.
Regulations lookups run on a Nezasa-provided Passolution account during preview; your company can add its own Passolution token any time in Settings → Credentials.
Rolling out first to companies in the preview program — reach out if you'd like early access.
Agents now remember key facts across your conversations -- and reply faster
Major New Feature.
Ask an agent to remember something -- a preference, a recurring detail, a fact worth reusing -- and it recalls it in later conversations, no need to repeat yourself.
Memory is scoped per user, per agent: what you tell one agent stays private to you and that agent, and other users' memories never mix in.
Replies are faster, too: agents now pull in the specific skills and connectors a question needs on demand, instead of loading everything up front on every message.
Leaner TripBuilder data fetching — faster agent responses, less credit used
Improvement.
Agents that read a booking from the TripBuilder Cockpit API now pull only the parts they actually need for the task, instead of the whole bundle every time.
Less data carried through the conversation means faster chat and report responses and lower credit consumption — with no change to what the agents can answer.
Applies to the Itinerary Checker, Trip Guard, TripBuilder Ops, and Passolution agents; the full booking is still fetched automatically whenever a task genuinely needs it.
Agents explain a connector credential problem in plain language, not a raw error
Improvement.
When a connector has no credential set, or its stored token is rejected by the provider, the agent now replies with a short, plain-language explanation and points to where a company admin can fix it — no raw error codes, status numbers, or internal details.
The agent doesn't retry a problem that only a credential change can fix, so the answer comes back promptly.
If only one of an agent's data sources is affected, it still answers fully from the sources that work and lists the unavailable one separately, rather than blanking the whole reply.
Submit any form or confirm any dialog with Cmd/Ctrl+Enter
Improvement.
Press Cmd+Enter (Mac) or Ctrl+Enter (Windows/Linux) to trigger the primary action of whatever you're working in — Run a report, Save an agent, confirm a dialog — without reaching for the mouse.
Especially handy in forms with a large text box, like the report prompt: there, plain Enter adds a newline, so previously there was no keyboard way to submit at all.
Works consistently across the Workbench — create and edit forms, confirmation dialogs, and modal panels.
Leaner trip overviews — TripBuilder markup stripped before it reaches chat
Improvement.
The day-by-day trip overview agents can pull for a booking is now delivered as clean, readable text instead of the source's raw formatted document.
Stripping the underlying markup shrinks this data significantly, so it costs less and loads faster in chat and reports — with the same trip detail as before.
Continues the leaner TripBuilder data fetching work from TAI-602.
Cleaner header when your company only uses one language
Improvement.
The language switcher in the top-right corner now only appears when your company has more than one Workbench language enabled — no more opening a menu with a single, already-checked option.
New companies now start with English only by default; more languages can be turned on any time in Settings → Languages.
Click anywhere in a report row to open it
Improvement.
Report rows in the Reports library are now fully clickable — no need to aim for the title. The title text also no longer displays in red regardless of status; only the status badge (Ready / Running / Failed) conveys state now.
Agent skills now bring their required connectors automatically
Improvement.
When you add a skill to an agent, any connector it needs comes along automatically — no need to also add the connector by hand. The agent's Connectors field is now reserved for wiring a connector directly to a built-in agent; for a personal or company agent, adding the right skill is all you need.
TripBuilder agents answer cancellation-fee and traveller questions more reliably
Improvement.
The TripBuilder agents — Ops & BI, Itinerary Checker, and Trip Guard — now read booking data with a shared field guide instead of inferring it from a single response.
Cancellation answers quote the authoritative customer total (
salesCancellationPrice) directly, rather than trying to reconcile it against supplier-level component fees — no more circular arithmetic when the two legitimately differ.A missing traveller field (e.g. a passport expiry that was never captured) is reported as not captured for this booking, instead of the agent wrongly concluding the system can't hold that data.
Reports always render as a polished, printable document
Improvement.
Every report is now prepared as a clean, self-contained HTML document — so even when an agent falls back to plain text, it reads as a styled document in the viewer rather than raw markup.
Because the report is built once, it looks identical wherever you open it — on screen, in the downloaded HTML, and in the exported PDF — and exporting is quicker with nothing re-rendered on the fly.
Agents that specifically need plain Markdown output can still opt in per report channel; polished HTML is simply the default everywhere else.
Faster replies from agents with many skills or connectors
Improvement.
Agents now load a skill's or connector's full instructions only when they're actually needed for the question at hand, instead of loading everything up front.
You'll notice it most as a snappier first response from agents that carry a large skill or connector library -- there's simply less for the agent to read before it can start answering.
Custom agents can query connectors from a built-in skill again
Fix.
Custom agents that add a built-in skill — such as the TripBuilder Data Warehouse skill — can now reach the connector that skill relies on. Previously the connector was withheld from these agents at run time, so the agent had no way to run the query and would stop with a message about the tool not being available.
Built-in agents were never affected. Now a custom agent built on the same skill behaves the same way and can answer connector-backed questions end to end.
Reports grouped by agent now show the agent's name, not an internal id
Fix.
In the Reports library, the group-by-agent headers now read the agent's name (like Itinerary Checker) instead of an internal identifier. Built-in agents were previously shown as a raw code.