You can also browse, search and filter all release notes directly in the application → Open release notes in TAI.
Company admins can now store their own Passolution API credentials
New Feature. Only Company Admins.
Connectors that support it (starting with Passolution Travel Regulations) now show a Connection section on their page where a company admin can store the organization's own Passolution access token — the token you create in your Passolution account under Account → API Access.
The token is verified against Passolution on save, encrypted at rest, and never displayed again — remove it any time to fall back to the Nezasa-provided default key.
During private preview the Nezasa default key applies automatically; bringing your own key becomes relevant in beta.
Follow TAI release notes in the Help Center — one article per release
New Feature.
Every production release now publishes its customer-visible release notes as an article in the TAI Help Center's Release notes collection — the same notes you see in the in-app feed, available where you already look for how-to guides. Each article groups changes into New, Improved and Fixed, and links back to the in-app page where you can browse, search and filter the full history.
See a report's live progress and inputs, and re-run failed reports in one click
Improvement.
Working with reports just got clearer. While a report is generating, the run now shows what the agent is doing and how long it has been running, instead of a bare spinner. Every finished report displays the input parameters it was generated from, so you can see its scope without re-opening the agent. And reports that failed are now openable from the library: the detail view explains why the run didn't complete and lets you re-run it with the same parameters in one click — or edit the parameters first.
Usage stats show clean model names like "Sonnet 4.6" instead of raw ids
Improvement.
Model names across the usage statistics — the by-model breakdowns and the sessions and events tables — now read as friendly names ("Sonnet 4.6", "Haiku 4.5") instead of raw provider ids like eu.anthropic.claude-sonnet-4-6-20250930-v1:0.
The same model invoked under different provider ids (regional profiles, dated snapshots) now counts as one row instead of splitting into several.
The raw ids stay available on hover for auditing.
Agents recover instead of failing when deep thinking uses up the token budget
Fix.
On a hard prompt, an agent with extended thinking could spend its entire token budget reasoning and emit no answer — surfacing a “token limit exceeded before any response was generated” error. Raising the limit only pushed the ceiling higher; a tough enough query hit it again.
The agent now detects that case and automatically retries the step with thinking switched off, producing an actionable answer instead of an error. Recovery is transparent and streaming responses are unaffected in the normal case. A reply may still be cut short if your Max Output Tokens is set low — that's a truncated answer, not a failure.
Cockpit deep links now open the right domain in the Terminal, PDFs, and reports
Fix.
Handoff links an agent generates (e.g. "open this itinerary in Customer Care") now point to your actual cockpit domain — white-label domains included — no matter where the answer ends up.
Previously they only worked when the chat was embedded live in the cockpit; in the Tai Terminal, downloaded HTML reports, and PDF exports they pointed at the wrong host or produced no clickable link at all.
Embedded chat uses the domain you're already on; other surfaces resolve your cockpit domain from TripBuilder automatically.