Release 5.14.11
This release focuses on Operator Line: dialog cards now update as soon as a message arrives even on instances with hundreds of open requests, the Actions button works again on custom folders and appeal filters, and the table view keeps its saved columns and client-field values without a page reload. It also closes a serious Scenario Builder case where saving could write your scenario into a different bot, fixes an Instagram story reply shown under another client's name, and makes email cards, Instagram attachments, and digitally signed PDFs behave as expected in the chat.
Operator Line
Fixed
- A dialog card now updates the moment a new message arrives, without switching folders. On instances where an operator sees more than 100 open requests, cards for the "newer" dialogs stayed frozen until the operator switched folders or reloaded the page — in the case that surfaced this, three client messages did not appear in the list for 19 minutes. The card now rises to the top, shows the new text, and increases the unread counter right away; a rapid series of messages is not lost, and a repeated delivery of the same message does not double the counter. The new-message sound (under the older notification rules) plays again for these dialogs too.
- The "Actions" button opens again on custom folders and appeal filters. On instances with the new search engine enabled, the button over the request list responded with an error instead of the bulk-actions menu — only on custom folders and appeal filters, so it looked intermittent. Bulk actions still run over exactly the set visible in the current folder or filter. If the request list for a bulk action cannot be fetched at all, the operator now sees a clear message instead of a button that simply "doesn't click".
- The saved column layout in the table view is restored in full on a cold open. Client-field columns could disappear from the saved layout when the panel was opened in a fresh session, and the two or three surviving columns stretched across the whole table — it looked as if the first column took up the entire screen, and only F5 fixed it. The layout is now re-applied once the full field catalogue is available, and any width, pinning, hiding, or reordering the operator saved is preserved.
- Client-field columns are filled in immediately after switching folders. Columns with client and organization fields used to show dashes (
—) after a folder change (for example, "All open" → "All forms") and only filled in after a page reload. - A photo sent by email is visible in a closed dialog. An email with no text, where the client attached only a photo, appeared as an empty card in a closed dialog — the file was there, but the card collapsed to a sliver. Closed email dialogs now use the same layout as open ones, which also brings back the email subject, read status, and message delivery indicators there. In the "all client emails" view, the operator's own emails from a closed dialog are correctly marked as the current thread instead of being labelled as another thread.
- The "Forward" item no longer appears in the message menu of a closed email dialog, where there is no reply field for it to open into.
- A previous email in a thread collapses under "Show previous emails" in more mail formats. Two real-world formats were not recognized before: the "-----Original Message-----" marker broken across a line break by clients that wrap HTML at a fixed column, and a plain quote block preceded only by a text attribution line ("date, time, From "Name"
:") with no technical markers — seen with ukr.net. The detection is deliberately conservative: an ordinary line of business text that merely contains a date, a time, or an email address will not collapse the operator's own reply. - An Instagram story reply no longer appears under another client's name and avatar. For story-comment conversations, the chat header showed the right person while the message author — name and avatar — belonged to somebody else, so the operator was working the conversation against a stranger's client card. Newly created story rooms are now always linked to the real sender. Rooms created before the update keep their existing link; contact your manager if you need one of them corrected.
- Story preview rows no longer show "No" as the message author. These rows are your own brand's content and have no author, so the label is simply not shown.
Clients
Fixed
- Clicking the "Clients" tab from a segment no longer flashes the login page. Opening Clients, selecting any segment or saved view, and then clicking the "Clients" tab briefly showed the sign-in form and dropped the user on the main page with the filter lost — even though the session was perfectly alive. The same applied to other one-level sections opened with parameters in the address (Settings, Dialog, Process Library). The tab now reliably opens the full client list; segments stay where they were, in the sidebar.
- The "Appeals history" tab on a client card no longer looks empty for a client who does have appeals. On a cold open — a new session in which the operator had not yet searched for chats in Operator Line — the table immediately showed "No chats found" and never sent a request at all. If the data is still being indexed, the tab now says so honestly instead of showing an empty list.
- The channel preview on a client card no longer shows the English phrase "No dialogs found in this channel". All preview messages are now translated in every interface language.
Scenario Builder
Fixed
- Saving can no longer write your scenario into a different bot or delete that bot's sections. Moving between bots inside Scenario Builder — via a link or the browser's Back button — changed the address but did not reload the editor, so it kept showing the previous bot's scenario, and the next "Save" sent one bot's sections into another. The editor now fully reloads when the bot changes, nothing from the previous bot stays in the tab's memory, and "Save" is refused with a clear message ("the scenario is still loading" or "reload the page") while the editor is not ready. The server also rejects such a save outright. The same protection covers activating a version from history.
- "Start run" from a version in history no longer fails with a server error. Viewing a version from the scenario history and starting a test run could end with an error — and when it did succeed, it silently tested the currently published version rather than the one on screen. Running a test from a history version is now explicitly blocked, with an explanation, at every entry point: the toolbar button, the Start node panel, the run dialog, and "Run from node" on the canvas. Stopping an already running test from history mode works as before. If a section has been deleted by someone else, the dialog now says so instead of showing a technical error code.
Added
- The test-run debugger shows more at a glance. Each step now has a short "Result:" preview under its metadata, so you no longer have to expand the step card to see what a node returned. A step error is marked with a warning icon, and a node passed more than once in a cyclical scenario shows an
×Nbadge on the canvas with a "Passed N times" tooltip. A new test run resets the counters.
Client Chat
Fixed
- An Instagram attachment Meta marks as an unsupported type now reaches the operator. A client (typically a blogger sharing collaborative content) could send an attachment that produced no message in the chat at all — the operator saw nothing. Such an attachment is now downloaded where possible and delivered as a photo or a file. When the media genuinely cannot be retrieved, the operator gets the same notice already used for other unsupported message types, so they at least know the client sent something and can view it in Meta Business Suite. Tip: add an
unsupported_messageconstant to your bot to show operators a message in your own language. - The widget accepts a PDF signed with a qualified electronic signature. A signed PDF (a PKCS#7 signature envelope) used to be rejected before the scenario even started, with "Unsupported file type". Such a file is now accepted, stored byte-for-byte so the signature stays intact, and opens for the operator as an ordinary PDF in the browser.
Fastline Pro
Fixed
- AI agents running on the Cursor engine answer again. From 14 September, the model the "auto" setting relied on was withdrawn by the provider, and every request from the AI Assistant — and from the documentation assistant — failed with an engine error. "Auto" now uses the provider's own automatic model selection, and the two withdrawn models were removed from the model lists.
Added
- An optional filter that blocks an agent's internal reasoning or prompt fragments from reaching the client in the widget. In rare cases an end client could see the agent's chain-of-thought or pieces of the system prompt in its answer. The protection is off by default and is enabled per instance, together with the list of phrases to block and the number of retry attempts — ask your manager to turn it on if you need it.
- A shortcut from an agent's tool list straight to the bot behind it. A scenario-based tool row in the agent tab now has an arrow button that opens that bot in Scenario Builder. Built-in and external tools do not get the button.
Stability and internal improvements
- AI quality assessment prompts no longer carry the knowledge-base instructions for checklists that have no knowledge base attached, which cuts the cost of every evaluation call on those checklists. Existing checklists pick this up the next time they are saved.
- Support and implementation engineers have new diagnostic tools for reading a chat's full audit log and the history of its subject, which shortens investigations into "who changed what, and when" in a conversation.
- Preparatory registration of feature switches for the upcoming sprint's functionality: no behavior changes on their own.