Human handoff and escalation

The agent is built to know when it is not the right answer. It hands a conversation to your team on its own judgment, the moment a visitor asks, and whenever something goes wrong. Your team picks up in the inbox with the whole conversation already there.

Conversation statuses

Your team can move a conversation between statuses from the inbox. Setting an escalated or resolved conversation back to unresolved puts the agent back to work in it.

When the agent escalates by itself

• The visitor asks for a person, for example "I want to talk to a human". The agent escalates immediately.
• The visitor accepts an offer of help. When the agent cannot find an answer in your content it says "I don't have specific information about that in our knowledge base. Would you like me to connect you with a human support agent?" A yes escalates.
• Something failed. If a tool or the model itself fails, the agent apologises and escalates rather than leaving the visitor hanging. If a reply cannot be generated at all, the visitor sees "Sorry — I'm having trouble replying right now. A member of our team will follow up here as soon as possible."
• A frustrated visitor is offered a person; the choice stays with them.
• The Free plan's monthly AI replies are used up. New conversations are escalated with "Thanks for your message! Our team will reply to you right here as soon as possible."

Replying to an unresolved conversation from the inbox also escalates it, so the agent does not talk over a colleague.

What each side sees

In the widget, escalation is quiet by design: the typing indicator stops, the agent stays silent, and the visitor can keep writing. Your replies appear live in the same chat window. They are shown with the same avatar as the agent's messages, so it helps to introduce yourself in your first reply ("Hi, this is Sarah from support").

In the inbox, the conversation moves to escalated, the preview reads "Conversation escalated to a human operator", and the full transcript is there to read before you answer. See The team inbox.

Visitors are not notified outside the chat

The widget does not send emails or show unread badges. A visitor who leaves the page sees your reply when they open the chat again in the same browser, within their 24-hour session. For anything time-sensitive, ask for an email address in your reply, or connect an integration such as Zendesk or Slack so the follow-up happens where your team already works.

Resolving

The agent resolves a conversation when the visitor signals they are done ("That's all, thanks"), with a short goodbye. Your team can resolve from the inbox at any time. A resolved conversation shows "How was this conversation?" with 👍 and 👎; the votes feed Analytics.