You can also browse, search and filter all release notes directly in the application → Open release notes in TAI.
Connectors: reach APIs that sign in with a client ID and a private key
New Feature. Only Company Admins.
A connector can now reach an API that issues tokens to a client which proves its identity with its private key, with no person signing in. Enter the provider's token URL, client ID, audience and your private key, and click Test + Save: TAI requests a token at once, so wrong settings are caught and nothing is saved. The connector then renews its token by itself. If the provider later refuses the client, company admins get an email to check the client settings.
Learn more in Credentials.
Ops & BI Agent: exact TripBuilder commission by category and rate
Improvement.
Agency commission broken down by commission category or rate now counts each commission item under its own category and rate: a room at 10% and its amenity at 20% no longer land together under the room's. Zero-rated items and negative adjustments are included, and amounts stay in each service's own currency.
Harness: advisor runs no longer fail on a malformed reply
Improvement. Only Company Admins.
Agent advisor runs and the Prompt Advisor now hand back their result in a checked format instead of writing it out as one long document by hand. A slip such as an unescaped quote inside a quoted excerpt is caught and corrected in the same run, so the run finishes instead of failing after a long repair attempt.
Connectors: the list now points to drafts in progress in Connector Lab
Improvement. Only Company Admins.
A connector that has been started but not finished is a draft, which the Connectors list leaves out. The list now says how many of your company's drafts are in progress, with a link to continue them in Connector Lab. A connector's page also shows its ownership, as agent and skill pages do.
Connectors: see which agents and skills use a built-in connector
Improvement.
A built-in connector's page now has a Used By section listing the agents and skills that use it, with via and the skill's name when an agent gets it through a skill. Used By on every connector, skill and knowledge page now names only what you can open yourself. Learn more
Used By on a connector: agents show which skill adds it.
Prompt Advisor: read a proposed prompt side by side with Split view
Improvement.
A prompt the advisor proposes in the agent editor (and, for admins, on an advisor run) can now be compared with Split: the current prompt on the left, the proposal on the right, each changed line opposite its replacement. Unified stays the default, and your choice is remembered for the next diff. Learn more.
Harness: long runs no longer fail with a false 'too large' error
Fix.
Agents with large skill sets, such as Itinerary Checker, stopped as 'too large' after a few steps. They now run to the end.
Prompt Advisor: fewer failed runs, and a content-filter stop now says so
Fix. Only Company Admins.
A finished analysis is no longer thrown away because a quoted excerpt in it kept its line breaks. These runs used to fail as "unparseable result" after an extra repair step of up to 90 seconds; they now complete straight away.
When the AI provider's content filter stops an advisor run, the run now fails with that reason, in your language for live advisor answers, instead of the misleading "unparseable result".
Knowledge: a new source's status updates once indexing ends
Fix. Only Company Admins.
A source you just added no longer shows Pending on the Knowledge list after it is ready: the list refreshes while a source is being indexed.
Connectors: no more glued sentences or duplicate versions in the Builder
Fix. Only Company Admins.
Replies that continue after a tool call are spaced properly, a repeated identical submit leaves one version, and catalogue changes are described as proposals.