Release 5.18.7
Release 5.18.7 brings sending a CSAT survey by email from a closed mail request, a fix for saving CSAT scores from a scenario, a stricter Fast Line right check for Fast Line Pro sections and a number of AI Assistant improvements. One change affects role access — see "Please note".
Please note
- Fast Line Pro sections open only with the Fast Line role right. Previously the right only hid the menu item, while the pages of AI agents, chatbots, knowledge bases, documents and chunks still opened by a direct link. Now a user without this right who follows such a link lands on the panel start page.
- A role with the view right but without the save right sees the section, but changes are not saved.
- A role that has the section right on but the view right off no longer opens the section. If it needs the section, turn on the view right in the role settings.
- Writing knowledge base chunks from external systems and changing the MCP tool binding of agents in Scenario Builder now also require the Fast Line save right.
- Integrations and service scripts that work with Fast Line Pro on behalf of a user (the knowledge base plugin, chunking, the AI Assistant) must use a user whose role has the Fast Line right with view and save. Otherwise their requests are rejected with 403. Check that user's role before updating.
- Roles with the full Fast Line right and root users work as before.
- The CSAT question from a scenario is now asked after every dialog. Previously a client who had rated once did not get the question next time. Ratings given before the update do not appear in the cards of old dialogs; they stay in the statistics.
Operator panel
New
- CSAT by email from a closed mail request. An operator with the "Send CSAT" right can manually send the client a survey by email from a request that arrived by mail (the Email channel) or as a Google Form submission. The "Send CSAT" button is in the "CSAT ratings" section of the details panel. The email contains a rating scale, and the client's answer appears in the client card and in quality statistics.
- Before: for a mail request the button was disabled, and a rating could only be collected by a scenario at the moment of closing.
- Now: the survey can be sent manually after closing, no later than 30 days after. When it cannot be sent (the request is still open, more than 30 days have passed, a survey was already sent, or email sending is not set up), the button is inactive and its hint explains why.
Fixed
- The CSAT score from a scenario reaches the request. When a client rated a dialog in reply to the CSAT question from a bot scenario, the operator did not get the "CSAT results received" notification, and the score was missing in the dialog card and chat history (it was present in statistics).
- Before: the score was not linked to the dialog it belonged to.
- Now: the score appears in the dialog card, in the chat history and in the notification to the operator.
AI Assistant (Instance Agent)
New
- Rating the assistant's answers. Under every finished answer there are "Like", "Dislike" and "Copy" buttons. A rating can include a comment, and text you selected in the answer is sent along with it. The buttons appear when the assistant has finished the answer or you stopped it.
- Closing dialogs by a saved folder or a one-off filter. The assistant can close a batch of dialogs, not only single ones. Before closing it asks for confirmation. Without the bulk actions right in the operator panel, closing does not start, and dialogs that are already closed are not touched again.
Improved
- The assistant is more accurate with parameter names and uses a short instance label. Tool parameters are now named consistently, so there are fewer errors when it performs actions.
- Stopping in the chat produces one message. After "Stop", the text "Execution stopped." is added to the answer already shown instead of appearing as a separate message with a second row of buttons. Stopped answers are now also told apart from successful ones in our assistant quality analysis.
Fixed
- The "Agent", "Plan" and "Ask" modes switch without leftovers of the previous mode. If you moved from "Plan" to "Agent" after a question, the assistant could refuse to act and answer that in planning mode it can only suggest. Now the mode of the current request applies.
- In "Plan" mode the plan map with a "Confirm" button is shown. Previously the assistant wrote that the action was unavailable and advised doing everything by hand.
- An answer is not lost when the assistant works for a long time. If the assistant ran a tool or thought for a long time, the chat could stay empty with no error even though the turn had finished.
- Before: after two minutes of silence the chat counted the turn as successful and showed no answer.
- Now: the connection is kept alive, and if it still breaks, the chat shows a timeout or lost connection message. After a break the request is not repeated automatically, so an action is not performed twice.
Technical improvements
- Background process handlers are more resilient: after a failure they recover automatically.
- ConnectiveOne MCP tools now mark which of them only read data and which delete it, so external MCP clients ask for confirmation before changes more accurately.