You can also browse, search and filter all release notes directly in the application → Open release notes in TAI.
Billing Center: see your credit consumption
Major New Feature. Only Company Admins.
Billing Center in the sidebar opens the real dashboard instead of a 'coming soon' placeholder. It shows your own company only.
Three headline figures — Billable, Included and Total value — over a date range you choose, opening on the last 30 days. The range covers both tabs.
Breakdowns by category (AI compute / Connectors / Storage), capability, identity, connector tier, agent ownership and agent, each a table beside its chart. Regroup the charts daily, weekly or monthly, or narrow the Overview to particular agents.
Top Sessions ranks individual chats and reports by credits. Credits settle once a day, so a session run today shows its figure once the day is confirmed — the Overview includes today's work as it happens.
Download CSV exports the per-session rows. It covers settled days only, so it leaves out today's open sessions and storage — the page says so. A Nezasa admin administering your company sees the same view you do.
The Billing Center dashboard — credit totals, per-agent breakdown, and a weekly trend, scoped to your company over the date range you pick.
Control TAI's recurring emails from My Account
New Feature.
New notification preferences under My Account let you turn off the release-notes summary or the monthly scheduled-reports digest independently — both are on by default. Operational emails, like password resets and the reports your own schedules deliver, are unaffected.
Release notes now land in your inbox too
New Feature.
TAI now sends a short email after each release summarizing what's new — scoped to exactly what you can already see in the in-app release notes. Turn it off anytime under My Account → Notifications.
Docs: eleven more how-to guides now show the screen they describe
Improvement.
Billing, Web access, Connector credentials, Email notifications, Using connectors, Custom agents, and the five built-in agent guides (Generic Assistant, Itinerary Checker, Passolution Agent, Traveller Locator, TripBuilder Ops & BI) each gained a screenshot of the real screen they walk through.
The Analytics guide's screenshot now shows a populated view instead of an empty one.
Company admins can now fix a connector's stale name in one click
Improvement. Only Company Admins.
A connector created as a throwaway draft keeps the name and description it was given, and agents read that text as the connector's scope — so a connector still labelled "QA test" could get its operations declined as off-topic.
When the assistant submits a connector whose purpose no longer matches the catalogue text, the review panel now offers the corrected name and description with an Update connector button. Nothing you typed is ever overwritten without that click.
Submitting a connector no longer fails because of its old name. The independent reviewer that gates activation now judges what the connector actually declares — its operations and the hosts it calls — instead of checking them against a label that may predate them.
The connector test drive now grounds itself in the operations a connector declares rather than its name, so it runs them instead of refusing them as off-topic.
Chat transcript PDFs now keep the reply's formatting
Improvement.
Export PDF printed a chat reply's Markdown verbatim — literal **bold** and - bullets, comparison tables as rows of | pipes, and → and € reduced to ?. Headings, lists, bold, code blocks and tables now render as such, and route arrows and currency amounts survive intact. For a polished report to share, ask the agent for a PDF report rather than exporting the transcript.
Reports author filter now lists everyone who ran one
Improvement. Only Company Admins.
The Reports library's Run by filter now lists every author who has actually created a report or run, instead of the company's first 100 users. A company with more users than that — or a report run by an API key or an integration — previously couldn't be filtered by that author at all.
The filter is now searchable, so a long list of authors is quick to narrow.
An API key's name now shows in the Run by column and on report details, in place of an internal id. An author with no name to resolve — a traveller-facing integration, for instance — still shows its own identifier, as before.
Top Sessions now shows today's credits as they accrue
Improvement.
Sessions run today used to show — for credits — the figure only appeared once the next day's settlement pass reached them. You now see a live estimate right after a chat or report runs, marked with ≈ until it is confirmed (usually within a day), in Analytics → Sessions and the Billing Center's Top Sessions alike. Sorting by credits ranks today's sessions correctly among settled ones too, instead of parking them at the bottom.
Analytics: see which knowledge bases agents use
Improvement.
Knowledge searches now show up alongside skills and connectors across Analytics: a new series on the activity-over-time chart, and a fifth summary tile counting knowledge calls.
A new Breakdowns chart, Searches by knowledge base, ranks your knowledge bases by how many searches actually found something in them — so you can see which ones are pulling their weight. One search can draw on several bases at once, so the bars can total more than the Knowledge calls tile.
Their history differs too: the tile and the over-time series include your existing activity, while the new breakdown fills in from this release forward and starts empty.
Itinerary Checker starts reviewing sooner
Improvement.
The Itinerary Checker now starts reviewing immediately and is less likely to stop early before producing a report. Paste a booking or name a TripBuilder ref as usual.
Docs: Password & security now points to the right avatar
Fix.
The Password & security and Email notifications guides sent readers to an avatar in the bottom-left of the sidebar, then to a menu item that does not exist. Your avatar is in the top-right, and clicking it opens My account directly. Both guides now say so.
Connector Builder self-tests now report what the target really said
Fix. Only Company Admins.
Three defects made a self-test misleading while the assistant built a connector:
Query parameters were dropped on the way to the target, so a filtered call quietly returned everything.
An upstream 404, 401 or 500 came back looking like a success, so the assistant reported a passing operation that had in fact failed.
Every turn ended with a stray "Files that could not be attached: connector.py" notice, even though nothing had gone wrong.
Self-tests now surface the target's real status and body, and the notice is gone.
Top Sessions marks sessions outside your date range
Fix. Only Company Admins.
In Billing Center's Top Sessions list, a session that started before, or continued past, your date range showed turn and token figures for only the part inside it, next to credits covering the whole session — which read as a billing error. Those rows are now marked Partial, and the marker says how much of the session the row actually covers.
Chats that hit a limit now answer with what they found
Fix.
A turn that ran out of steps, hit its token budget, or gave up on a tool that kept failing used to end as a blank message. It now says plainly why it stopped and, where there is budget left to do so, sums up what it gathered. The Incomplete badge and token count show right away instead of only after a reload.
Analytics: crowded breakdown charts are readable again
Fix.
Breakdown charts — by agent, by skill, by connector and by user — now show the top 10 by volume plus a single Other bar, so a busy breakdown no longer compresses every bar and label past legibility.
Each capped chart gains a View all button: the full ranked breakdown as a searchable, sortable, paginated table.
The skill and connector breakdowns also stopped competing for one shared slot budget — a company with very busy skills could push its connectors out of the chart entirely. Both now rank independently.