Release 5.18.0

A minor release with noticeable changes for the daily work of operators, integrators and administrators. In the operator panel you can now leave a request without a status and hide unneeded request fields, the topic filter shows nesting, and a chat taken from the queue no longer drops out of the "My Chats" list. In Scenario Builder, large schemas can be organized with colored frames, and the events of an action node can be reordered. The AI Assistant gains quality assurance tools: it explains why the AI took points and changes criteria only after confirmation. The operator package limit is now counted with the same number that goes into billing, and MCP access is issued as separate keys with a role and an expiry. Some new capabilities are off by default and enabled on request — see "Please note".

Operator panel

New

A request can be left without a status. In the request details panel, the "Status" field now has a "×" clear button. After you confirm the prompt "Remove the status …? The request will be left without a status.", the status is removed and a system line such as "Olena Koval removed status Resolved" appears in the feed. Previously, a status set by mistake could only be replaced with another one, which distorted reports and status filters. The button is available for requests that are not closed, to users with the right to change the status — in both the new and the previous interface.
Unneeded request fields can be hidden. In the updated Operator Line interface, each request field row now has the same "Unpin field" button as client fields. An unpinned field disappears from the panel, and you can bring it back with the "Show unpinned fields (N)" button. The choice is stored per operator, survives a page reload, applies on other devices, and covers every topic that has this field.
Tickets: you can see that SLA did not start. If the SLA timers did not start when a ticket was created because of a technical failure, the ticket is still created as before, but its SLA section now shows an "SLA not started" badge with the reason and a "Start SLA" button. Time is counted from the moment the ticket was created, not from the button press, so downtime does not extend the deadlines: if a deadline has already passed, the escalation fires immediately. Only the missing SLA metrics are started.

Improved

The topic filter shows nesting. In request filters and saved filters, as well as in the request topic filter of the "Clients" section, child topics appear right under their parent with an indent. Searching by a parent topic's name keeps the whole branch, and the topic tooltip shows the full path — identical names such as "Other" in different branches are now easy to tell apart. In "Clients", the list shows all topics, not just the first 50. What the filter selects has not changed: choosing a parent topic still does not include its children.

Fixed

A chat taken from the queue no longer drops out of the "My Chats" list. After an operator took a chat from the queue, its card could disappear from the list on the very next message (the list header then showed, for example, "4 / 5") and came back only after switching folders or reloading the page. The card now stays in the list, and the chat leaves the "In the queue" folder right away for all operators.
Dialogs that FastLine AI handled and closed on its own can be found by search and filter. Previously, such dialogs could not be found either with the "Executor: FastLine" filter or by searching for a phone number or name, even though they appeared in the statistics export. Users with FastLine permissions now see them both in the filter and in search results. Operators without these permissions still do not see such dialogs.

Clients

Improved

Creating a client — on a single page. The "Create client" panel in the updated "Clients" interface is no longer a two-step wizard: full name, Email, Phone, "Organization" and the additional card fields are all visible at once and are saved with a single "Save" button. If none of full name, Email or Phone is filled in, the system shows a hint, and the organization and additional fields you have already entered are not cleared.

Scenario Builder

New

Grouping nodes with a colored frame. Select several nodes and click "Group" — a colored frame with a title appears beneath them. You drag the frame by its header together with all the nodes inside; it can be renamed, recolored and resized, and "Ungroup" removes only the frame without touching nodes or connections. The frame is purely visual: it does not affect scenario execution, validation, export or billing. Large schemas become easier to read, and a logical block can be moved in one motion.
Event order in an action node. In the action node inspector, each event in the "Events" section now has ▲▼ buttons. The order of rows on the node follows the order in the panel, so outputs can be arranged to match where the target nodes are — and the arrows stop crossing. Only the display changes: the scenario logic and the connections between nodes stay the same.

Improved

Clearer step details in the "Runs" section. An expanded step of a test run now shows two blocks: "Input data" — the action, the step parameters and the variables used with their values before the step — and "Output data" — the result, the block the scenario actually moved to, the changed variables and the error. A node waiting for a user reply is marked "Waiting". The full technical content of the step is available behind the "Show raw payload" switch; the choice is remembered in the browser.
A warning about a topic that was not found, in the scenario tester. If the "Connect to operator" node receives a topic alias that matches no topic, the chat is still connected without a topic as before, but a warning now appears in that node's logs in the scenario tester — so such a configuration mistake is easy to spot.

Channels

New

WhatsApp (Maytapi): a default topic for messages from the manager's phone. When a manager writes to a client directly from the WhatsApp app and the system creates a new chat from that message, the chat now gets the channel's topic right away — and lands in that topic's skill group instead of "Unassigned". The topic is chosen in the WhatsApp Maytapi or Custom Channel settings in Scenario Builder ("Default subject" → "Save subject"); there is no need to reconnect the channel. Chats that are already open do not change their topic. The feature is optional, off by default, and enabled on the instance on request.

Fixed

Email (Gmail): letters from different people in one thread are no longer merged. If different people sent letters with the same subject (for example, from a feedback form) and Gmail combined them into one thread, the system treated them as a single client — an operator could see another person's email in the chat. Now each sender in such a thread gets a separate conversation, a separate chat and their own client card, and the chat shows the actual sender's address.

Settings

New

Skill group "All topics, including future ones". In a skill group's settings ("Topics" section), a root administrator can turn this option on — the group immediately receives all existing topics and then picks up every new or restored topic on its own. This is convenient for general groups such as "Support": new topics no longer have to be added by hand for their requests to reach operators. While the option is on, the group's topic tree is locked and the form shows a "Topics in group: N" counter. Turning the option off does not remove topics already added. Other roles do not have access to this option.

Fixed

A dialog status can again be linked to several bots. On some instances, a status with two or more bots in the "Bots" field could not be saved — a duplicate error appeared. Such statuses are now saved and work; the data structure is updated automatically during the instance update.

Limits & usage

Changed

The operator package limit is counted the same way as in billing. The check when an operator is created or activated now compares the package limit with the number of operators who accepted more than one dialog in the current calendar month — the same number that goes into billing. Previously the check relied on a snapshot that was updated with a delay, so early in the month it could let a mass creation of operators through. The new counter may lag behind the actual data by up to an hour. The change takes effect immediately on instances where billing limits are enabled and an operator limit is set.
The operators tile in the "Limits & usage" section shows this same number with the caption "N of M billing operators".

New

Restricting going online when the operator package is used up. When the package is full, an operator who is not among the package operators for the current month cannot switch to a status that receives chats, take a new chat, accept a transfer, or get a chat through automatic connection. They see a banner in Operator Line and a message when changing status. Package operators, root users and test accounts work as usual, open dialogs are not interrupted, and reconnecting to a dialog the operator has already taken part in is not blocked. The feature is off by default and enabled on the instance on request.

Statistics

Improved

The "CSAT" filter — a choice from a list. On the "Dialogs" tab of native statistics, the "CSAT" filter is now a multi-select list instead of a text field: "Has rating", "No rating", and the scores that occur in the data for the period, in numeric order. The table and both charts on the tab now use the same filter, so their numbers match. With no CSAT selected, the numbers do not change.

Fixed

A dialog transferred to a skill group that nobody took is no longer credited to the operator who transferred it. If nobody from the group connected after the transfer before the dialog closed, the closure and its CSAT rating used to go into the transferring operator's statistics. Now such a dialog is recorded without an operator: it is not counted in any operator's "Accepted" or CSAT and counts as missed for the operators who were online when it closed. The change applies to dialogs closed after the update.

Quality assurance

Improved

AI assessment has become more consistent. The base instructions of the AI assessment agent have changed in three places:
- the comment is now written before the verdict, and the score and result must match it — no more cases where the comment says "no points deducted" while the criterion is failed;
- the criterion takes priority over the knowledge base: each criterion assesses only its own topic, and a single general rule from the knowledge base no longer takes points across many unrelated criteria;
- a critical criterion counts as failed only when the AI gives a verbatim quote of the operator's message with the violation; without a quote the criterion is passed.
The new rules apply to a checklist after its AI agent is updated — when the checklist or a criterion is saved, or during an update carried out by the ConnectiveOne team.

AI Assistant

New

Finding out why the AI took points. Ask the AI Assistant to explain a quality assessment — it compares the criterion, the dialog, the knowledge base and the manual correction and tells you where the mistake is: in the AI, in the criterion text, in the knowledge base or in the manual correction — or that the penalty is justified. It adds verbatim quotes to the verdict, points out where to look in the interface and suggests a draft fix. During the analysis it changes nothing. For this, the assistant gained knowledge base search and, in read-only mode, access to quality assessments within your role's permissions.
Changes to criteria and AI assessment settings — with confirmation. When the assistant changes an AI QA criterion or one of the AI assessment settings, the confirmation card shows the path, what exactly changes, the full "Before / After" with copy buttons, and the effect of the change. The change is written only after "Only this time"; out-of-range values are rejected, and if the value has changed since the card was shown, nothing is written. AI assessment settings can be changed only by root, a criterion — by a user with the right to save checklists; without the required right, the assistant shows no card and replies with text describing what to change and where, by hand. A draft fix from an assessment analysis can be applied through the same card.
Questions to the "AI judge". At your request, the assistant can ask the checklist's AI agent itself why it passed or failed a specific criterion, and show the saved verdict and the answer side by side — quotes from the dialog and the knowledge base lookups. The assessment does not change. Each question is a separate paid request to the checklist's AI agent, so a daily limit per instance applies (30 by default). The feature is off by default: root turns it on in the instance settings, and roles with the right to edit assessments can ask.
Exporting statistics to a table. On request, the assistant builds a table of key metrics or statistics in CSV or markdown format and shows it in the chat — you can copy the content and save it as a file. There is no separate download link.

Improved

Any data change — only after confirmation. When the assistant is about to write something on the instance, a card with the buttons "Only this time", "For this chat, for actions like this" and "Reject" appears in the chat. For changes to a Fast Line Pro agent's configuration, as well as for deleting and restoring data, only "Only this time" is available. The assistant will not perform a sharp shortening of a long Fast Line Pro agent prompt at all (new text shorter than half of the current one) — this protects a working prompt from being overwritten by accident.
Targeted edits of a large Fast Line Pro prompt. To change a fragment of a large agent prompt, the assistant now finds the exact place and replaces only that fragment instead of rewriting the whole text. If the fragment does not occur in the prompt exactly once, the replacement is not made.
Resilience to failed changes. If a write fails, the assistant does not repeat the same change in a loop and stops after several identical errors. If a reply was cut off by a timeout, the assistant warns plainly that the change was not confirmed and the settings should be re-read.

Fixed

Message text in nodes created by the assistant is no longer lost. When the assistant or another MCP client added or changed a node with a message, that message could arrive empty in the test chat. The text is now stored as a section translation and displays correctly. Nodes saved before the update stay as they are until they are saved again.

Fast Line Pro

New

Context between agents in one scenario. If a dialog in a scenario moves from one Fast Line Pro agent to another, the next agent can now receive a short summary of what the client discussed with the previous ones. The summary is written by a separate summarizer agent that you choose in the bot settings (the "Context between Fastline Pro agents" card); the scope and the scenario constants to pass along are set there too. Each summary is a separate agent call that counts toward AI usage just like a regular answer. If the summarizer does not respond, the next agent answers as usual. The feature is optional, off by default, and enabled on the instance on request.

Integrations and MCP

New

Separate MCP keys with a role and an expiry. The "MCP Token Configuration" page now manages separate keys: you give a key a name and choose a role and an expiry (30 days, 90 days, 1 year or no expiry), and the list shows the author, last use and status of each key; a key you no longer need can be revoked. The secret is shown only once — at creation. A key's access is determined by the current permissions of the chosen role, including the bots available to it: changing the role or revoking the key takes effect from the next request, without reissuing the key.

Improved

A longer token lifetime for integrations. For integrations that obtain an access token programmatically, the maximum token lifetime can now be extended for your instance — up to 30 days — without changing the length of operator sessions. The ConnectiveOne team configures this on request. The message about an invalid lifetime now states that the value is in seconds and gives the allowed range.
A clear response to an invalid filter. List requests (clients, tickets, Custom Data and so on) with a filter the system cannot process now receive an "Invalid filter" error response instead of an internal server error.

Instance update notifications

Improved

The "Update in progress" indicator is more reliable. The notice about maintenance on the instance is now visible on the sign-in page too, and while the system is temporarily unavailable — without any extra instance configuration. The system rejects a notice whose expected end is in the past, so the indicator no longer shows "the update is taking longer than expected" right after the work starts.
System notifications no longer cover the notification bell — they now appear at the top center of the screen.
After an update finishes, open tabs do not all reload at the same moment — the reloads are spread out over time, which reduces the load on the instance right after the update.
The ConnectiveOne team can now automatically accompany a scheduled instance update with a notice at the start of the work and remove it once the update completes successfully.

Moving off the old design

New

Automatic move to the new design. An instance can have a mode turned on in which every old-design user — both someone already working in an open tab and someone who has just signed in — is automatically moved to the new design in the same tab and to the same page, without signing in again. The "Go to old design" menu item then disappears from the new design. The mode is off by default and is turned on by the ConnectiveOne team.

Stability and internal improvements

A gradual growth in memory usage after rare failures of background processes (schedules, tickets, broadcasts) has been eliminated, and the service step that updates the database during deployment no longer needlessly starts the whole server.
The database components have been updated, and a mechanism for pre-reviewed changes to the database structure has been prepared; this does not affect how instances work.

Please note

Features that are off by default. The going-online restriction based on the operator package, the default topic for WhatsApp (Maytapi), context between Fast Line Pro agents and the automatic move off the old design are enabled on the instance on request — contact your ConnectiveOne manager. Questions to the "AI judge" are turned on by root in the instance settings. Until these features are enabled, the system behaves as before.
Operator package limit. The new counting rule takes effect right after the update on instances with billing limits enabled and an operator limit set. At the start of the month the counter resets, so the check lets more through, while towards the end of the month it may block where it did not before. The "Add operator" button and the threshold banner still rely on the previous counting method, so occasionally the button may be active while creation is rejected, or the other way round.
Going-online restriction (if enabled). An operator who was already in a status that receives chats when the package filled up stays in it but gets no new chats; the banner appears after Operator Line is reopened. An operator who accepted only one dialog in the month is not part of the package.
MCP tokens obtained before the update no longer work for MCP connections. Create a new key on the "MCP Token Configuration" page and replace the token in your MCP client's settings (for example, Cursor).
AI quality assessment verdicts will change once the checklists' AI agents are updated: fewer failed criteria are expected, and a critical criterion without a quote of the violation will be passed. If additional instructions for AI assessment on your instance define violations beyond the criteria text, check the results on a few dialogs.
Operator statistics. Dialogs transferred to a skill group that nobody took before closing no longer count toward the transferring operator's "Accepted" and CSAT, and the "Missed" count of online operators grows accordingly. If you measure operators by "Accepted" or CSAT, take this into account. Data for past periods is not recalculated.
Email (Gmail). A new letter from the same client in a new Gmail thread now opens a separate chat. A client's reply to a letter sent before the update may open a new chat until an operator replies in that dialog after the update. Client cards that were merged earlier are not split automatically.
Capabilities that are now always on. A number of capabilities that used to have separate switches and were already on for most instances are now permanently enabled: the updated settings pages (users, roles, client and user fields, the operator panel settings sections), the process library mode for bots, some Scenario Builder and AI Assistant capabilities, and binding scenarios as tools to Fast Line Pro agents. If any of these was turned off on your instance, it will start working after the update.
Automatic move off the old design (if enabled). Unsaved changes in the old design are lost at the moment of the move; tabs opened before the update move to the new design after the page is reloaded.
Removing a status from a ticket works the same as changing it to any other status: if you remove a status that marks the ticket as resolved, the ticket is considered unresolved again.
Hiding request fields and creating a client on a single page are available in the updated interface; these screens have not changed in the previous interface.

Release 5.18.0

Operator panel

Clients

Scenario Builder

Channels

Settings

Limits & usage

Statistics

Quality assurance

AI Assistant

Fast Line Pro

Integrations and MCP

Instance update notifications

Moving off the old design

Stability and internal improvements

Please note

Related documentation