Release 5.16.0

The headline of this release is Context API, second generation: an integration no longer has to assemble scattered client data on its own — it requests a compact pack instead: what is happening now, which events are already registered, what is known about the person or subject, and which appeal snippets it can cite. Two built-in scenario actions now work with this context without touching the CRM card. Separately, the release closes a long-standing Telegram channel issue: when a group is converted into a supergroup, messages now go to the correct chat id instead of getting lost. On top of that, Scenario Builder's test run sidebar is easier to search.

Integrations

Added

Compact context for one entity. Context API can be requested for a client, subject, operator, or skill group. The response includes the current state, a journal of already registered events, allowed fields, and short appeal snippets you can cite. The pack does not include the full chat archive.
Event and state writes stay outside CRM. An integration can register, for example, "checkout submitted" and a purchase stage in its own data group. The client card, appeal status, and queue do not change.
A retry does not duplicate the event. The same write command can safely be sent again and still produces one record. A different body under the same key is rejected.
MCP tools for reading context. They read the same pack for a client, subject, operator, and skill group. There is no write path in MCP.

Scenario Builder

Added

The "Register client event" action. From a scenario step, it records a client event in Context API and optionally updates state. It is not added to existing bots automatically — a scenario developer adds it where it's needed.
The "Load external context" action. Calls an allowed external system and stores the response in the client's context — unlike an ordinary request to an external service, this action puts the result straight into context, ready for later reads.
The "AI QA assessment completed" internal event in the Start node. Previously, to react to the result of an AI quality assessment, an integration had to poll assessments separately or wait for the chat to close. A Start node can now subscribe to an event that fires right after a successful AI assessment, carrying the score and a short summary. The event does not fire on a failed assessment; a second successful assessment of the same dialog fires it again.
Search in the test run "Execution details" sidebar. Finding a specific step in a long test run used to mean scrolling the list by hand. There is now a search field above the execution sequence that matches by step name or ID (a query like #12 matches the ID exactly), a filter by block type, next/previous navigation through matches — by arrow buttons or Enter/Shift+Enter — that accounts for repeated nodes, and a separate marker for steps whose block no longer exists in the scenario.

Channels

Fixed

Telegram: correct handling of a group-to-supergroup conversion. When an admin converts a Telegram group into a supergroup, the messenger changes the chat id. Previously the system kept sending messages to the old id (the chat showed Telegram service errors), and incoming messages from the new id created a duplicate client and dialog in the operator panel. The system now recognizes this conversion and moves the dialog to the new id without losing history, leaving a system message in the dialog that explains why the id changed.

If a duplicate dialog had already appeared before this fix (the group was converted earlier), the old dialog card is closed cleanly with a system message linking to its successor dialog — without notifying the client about the closure and without affecting that dialog's quality score.

Settings

Improved

The Context API and new search readiness page now shows the actual connection state, not just that the setting is on. You can see whether client, subject, operator, and skill-group context are ready, and whether AI features are connected to them (typically not yet).

Stability and internal improvements

Reduced load and background log volume on instances where the new search and context features are only partially enabled or not enabled at all — background processes now more precisely detect when there is actually work to do.

Please note

Context API and the new search are turned on by instance settings. Until the context store is ready, requests fail — there is no silent fallback to the old format.
Copilot and Fast Line Pro do not call Context API by themselves. Connecting AI features to context is a separate decision; this release does not turn it on.
A client message like "I paid" stays an ordinary chat snippet. It becomes a confirmed payment event only when an integration or scenario registers that event separately through the new action.
Duplicate client cards created by a Telegram group conversion before this release are not merged automatically — use the Client Deduplication workbench if needed.
Search in the Scenario Builder test run details is available immediately, with no separate setting to turn on.