When your website chat doesn't know the answer, a Muzau agent says so plainly and gives the part it does know, with instructions not to invent a price or a deadline. If you chose that option, it asks once for the visitor's email or phone number and, if they leave one, passes their question to your team. The team finds it in "To do" with the whole conversation, and replies by email or calls the number back.
A visitor asks the chat something your site doesn't cover: a special case, a lead time for their situation, a price that depends on their request. A badly set up chat makes something up, and your team has to correct it. Or it says "I don't know", and the visitor closes the tab without leaving a trace.
The goal is not for the chat to know everything, but to keep the contact until the person who does know. Your team keeps the answers that need judgment; the agent holds the connection in the meantime.
A Muzau agent's chat answers from your business's Brain and the pages of your site that Muzau has read. When it has only part of the answer, its instruction is to give that part, then say in one sentence what is still open. It must not invent any price, lead time or commitment that is not in the Brain. Every answer is also reviewed automatically before it is shown: if it promises something nothing backs up, it is held back and the question goes to your team. See what this review checks.
When you create the agent, you choose how it behaves: "It takes their address and alerts me" or "It just says it doesn't know". In the first case, the chat adds: "Leave me your email or phone number and someone from the team will get back to you." Once the contact is given, the visitor reads "Noted, we'll reply to …", or "we'll call you back on …" for a phone number. That sentence is written only when the request actually goes to your team, never before.
The request becomes a card, "Someone is waiting for an answer (website chat)", with the email or phone number left and the questions asked. The visitor's later questions in the same conversation are added to it, and a visitor who comes back with the same email rejoins their card while it is still waiting. If you gave an address to be alerted at, your team also gets an email, just one per conversation, with their contact, what they asked and what the agent has already done. With Slack or Microsoft Teams connected, the agent alerts the team there too: see getting alerts in Slack or Teams.
On opening the card, the person taking over sees the whole conversation, the agent's replies included. The agent prepares a reply for them, which they send by email from the business address, to the one the visitor left; for a phone number, they call back. From the same card, they can add the missing answer to the Brain, choosing who is allowed to hear it. If it is open to everyone, the chat can use it for the next question of the same kind; if it is kept for the team, it will never be said to a visitor.
Whatever the tool, a chat that doesn't know stays useful if this moment is handled properly:
The team's reply works in three steps. It restates the question in the visitor's own words, since they may have forgotten what they asked: "You asked in the chat whether …". It answers, or says precisely what is still needed. It ends with a single next step: a time slot, a document or a question.
Review the unanswered questions every week: the ones that keep coming back show what your site is missing, such as prices, lead times or a frequently asked questions page. For the rest of the chat, meeting visitors' needs, see answering your website visitors at any hour.
Autonomous does not mean unchecked. You set the level of autonomy of every agent, its limits cannot be crossed, everything it does is recorded in a plain-language log, and one gesture stops it all.
For the tasks you don't have time for. You give the address of your website, then describe the work in your own words. Nothing to install.