AI controls

When a chat agent is handling a conversation, the AI can stop responding — either because a teammate stepped in, or because something about the conversation calls for a person.

​

When a teammate replies, the AI stops

The moment anyone on your team sends a message in a conversation from Unibox, the AI goes quiet on that conversation. This is automatic on every channel and there is nothing to configure — a human reply means a human owns the conversation. The AI does not come back on that conversation by itself, and it stays out of it however long the conversation runs — closing the conversation does not change that, and neither does the contact writing in again days later. A human reply means a human owns the conversation until someone says otherwise. When you want the AI to take over again, open the ⋯ menu in the thread header and choose Let the AI reply again. The AI answers the contact’s next message as normal. It applies from that moment on, not retroactively, so nothing you or the AI has already sent changes. The option carries a second line, Paused since a teammate replied, so you can tell what you are undoing. It is only shown to members who can manage Unibox. A view-only member does not see it. The same menu uses the words Let the AI reply again for a second thing: lifting a No AI reply rule on the sender. That option sits lower in the menu, with the sender rules, and removes the rule for every conversation with that sender. See Blocking a sender. If a teammate replies again afterwards, the AI goes quiet again — you do not have to undo anything first, and handing a conversation back once does not opt it out of this permanently.

​

Pause reasons

The AI also stops on its own in a few situations. Whatever the cause, the conversation shows a paused band with the reason: Where you see it:

• A band at the top of the conversation, just under the header, reading AI paused and the reason. It is shown on open and snoozed conversations; a conversation that has ended does not show it.
• A row in the conversation timeline marking the moment the AI stopped — and, if someone has handed the conversation back, a row for that too.

​

Turning the AI back on

It depends on who stopped it, and the difference matters: When the AI stopped itself — a keyword, an escalation, a booking, a message limit, profanity, a stop request from the contact, a reply loop — marking the conversation read is the way back. Click Mark as read in the thread header, and the contact’s next message is answered by the AI as normal. When a teammate replied, marking it read is not enough, because the AI is staying out of a conversation a colleague is handling rather than reacting to something the contact did. Use Let the AI reply again in the ⋯ menu. That option shows on an open conversation that is currently showing the teammate-reply pause. On a conversation that has ended it is not there. A chat conversation does not reopen when the contact writes again: their next message starts a new conversation, and the AI answers it as normal. One exception clears itself. An Out of credits pause lifts as soon as the wallet is topped up. It is a workspace-wide billing state rather than a decision about that conversation, so no one has to reopen anything. Escalations wait for your team. When the agent escalates, your admins get an email and the AI stops replying in that conversation. The contact can keep writing, and the thread waits for a person. The conversation closes 24 hours after the escalation, gets its summary and outcome, and the contact’s next message starts a new conversation that the AI answers. Your team is emailed once per conversation. The escalation email links straight to the conversation in Unibox, so whoever picks it up can reply there. Contact opted out is different. Marking the conversation read does not lift an opt-out. SMS opt-outs block every outbound path to that contact — including the AI — until the contact replies START, UNSTOP, or YES, or someone manually re-opts them in from the contact panel. Marking it read just ends this conversation; the AI still won’t message that contact again until the opt-out itself clears.

​

Reply loops

Some senders are bots. A billing notice or a code sender can answer every message it gets with the same text. Without a guard, the AI would answer back each time, and the two would loop for as long as the bot keeps going. Nedzo watches for this on SMS, WhatsApp, Messenger, Instagram, and email. On SMS, WhatsApp, Messenger, and Instagram, the AI pauses when, inside 10 minutes:

• the AI’s last 3 replies were the same, or
• the contact’s last 3 messages were the same and the AI’s last 2 replies were too, or
• the AI sent 25 replies in the conversation.

On email, the AI pauses after 8 auto-replies in a conversation inside 10 minutes. Case and spaces do not matter when messages are compared. Replies that failed to send are not counted, so a sending outage does not look like a loop. A contact who sends the same message a few times does not trip the guard on their own. When the guard trips, the conversation pauses with the reason AI paused - reply loop detected, and a row in the timeline marks the moment. The AI sends nothing more. Check the thread and reply yourself if a person is on the other end. To hand the conversation back to the AI, mark it read. There is nothing to configure. The guard is always on.

​

Configuring pause behavior

Some pause reasons are configured per agent, in the agent’s Settings tab:

• Protection keywords that trigger a pause
• Profanity detection
• Message limits
• Whether to pause after a completed booking

Three pauses are not configurable. The reply loop guard is always on. Pausing after a teammate’s reply always happens, on every channel. Pausing after an escalation always happens too: the AI stays out of an escalated conversation until it closes, with no per-agent setting to turn that off. See Chat Agents for full configuration details.