When a customer asks for a live person
This page covers the Conversations section: where to see people waiting for a human, and how to reply to them.
The assistant answers by itself for as long as it can. But in the chat window on your site and in your own app, a visitor can tap “Talk to a person”, and in Telegram, Viber and WhatsApp they can ask for a person in words — and then the conversation waits for you.
The reply goes back where the call came from
A customer from a messenger gets your reply in that same messenger; tap “End” and they learn that the assistant is answering again. If delivery fails (the channel is off, the customer blocked the bot, the 24-hour window in WhatsApp has passed), the reply box will say why — the reply won't pretend it was sent. What exactly the customer hears when they call for a person — Channels → When a customer asks for a live person.
Where to see them
The Conversations section at the top of the dashboard. On the left — the list of requests; on the right — the selected conversation in full: what the customer asked, what the assistant answered, what you answered.
While the section is open, the number of people waiting is shown in the tab title — so you don't have to sit in it all day. When the dashboard is closed, the request comes to you by itself — in a messenger and by email: see So requests reach you outside the dashboard.
Everyone you've invited can see the section — even the member role can reply to customers. See Team.
Assistant drafts
If training mode is on, there's a “drafts waiting for you” bar above the list of requests: the assistant has written a reply, and the customer gets it when you send it — as is, edited, or your own. In such a conversation there's a draft card instead of the usual reply box.
The conversation status is shown at the top
| What it says | What it means |
|---|---|
| Customer is waiting for a person | they tapped the button, the assistant went quiet, there's no reply yet |
| A person is handling this — the assistant is silent | you or someone on the team is replying |
| Handled by “name” | a colleague has taken the conversation |
| The assistant is answering now | an ordinary conversation, nobody called you |
What you can do
- Take — the conversation becomes yours, and colleagues see that someone's on it. A customer waiting for a person sees this right away: the chat window header says “An operator is with you”, and the conversation shows the line “An operator has joined the conversation”, even if you haven't written anything yet. Return the conversation to the team — and they see “Waiting for an operator…” again.
- Take over — the same, but a colleague is already handling the conversation. The reply box is closed here not with a lock but on purpose: taking over should be a deliberate step, and your colleague will see that the conversation has passed to you.
- Return to team — you're no longer handling it; the request is free again.
- End — the conversation is closed, and the assistant answers the customer again from then on.
You can also write without taking the conversation: as soon as you send a message, the assistant goes silent in this conversation. Two can't reply at once.
You can also write first — in a conversation the assistant is handling (the “Handled by assistant” view): the customer didn't call you, but they'll see your message. Not instantly, as in a conversation where a person was called, but within a few seconds — up to half a minute if their chat window is minimised: then a number appears on the chat button. An assistant conversation open on your screen keeps loading by itself while the customer keeps writing.
If the assistant was in the middle of replying when you stepped in, its reply will still reach the customer — it was already being written. The transcript shows everything in the order it happened: the question, your message, the assistant's late reply.
The same in words
Without opening “Conversations”, you can tell the builder:
| Say | And it will |
|---|---|
| “what's going on in conversations?” / “who's waiting?” | list conversations where a customer is waiting for a person, drafts to review and actions awaiting approval |
| “tell him we'll deliver tomorrow” | write a reply in your words, show you the exact text and send it after your “yes” — in your name, and that's how it appears in the transcript |
| “assign this conversation to me” / “return it to the team” / “end the conversation” | do the same as the buttons above |
| “approve” / “decline, because …” | approve or decline an assistant action awaiting approval — only the one you named, and after “yes” |
The builder here is your hands, not another assistant: it replies to a customer and approves actions only at your direct request, never on its own. The text that goes to the customer is exactly what it showed you; if you want it different, say so, and it will ask again.
What the customer left in the chat window
A customer in the chat window doesn't just write. They can leave a contact (the “Leave contact” button: name, phone or email, a note) and rate the assistant's answer — “helpful” or “not helpful”. All of this appears in the conversation itself, in its place, without reloading the page:
| What the customer did | What you see |
|---|---|
| left a contact | a line “Customer left a contact” where they did it: the name, the contact itself with a “Copy” button, and whatever they added. The latest contact is also shown in the conversation header — while it's open, it's right in front of you |
| rated an answer | a caption under that very answer: “Customer rated the answer: not helpful” (or “helpful”) |
In the list, such conversations are marked: “left contact” and “dissatisfied” (the customer's latest rating is “not helpful”; a later “helpful” removes the mark). Below the search box there are two buttons — “Left contact” and “Dissatisfied”: they keep only such conversations in the list, in any view and together with search. Pressing the same button again brings back all of them.
The notification about a new contact (see When something breaks) includes a link to the conversation where it was left.
The customer doesn't see these lines: they only see “Thanks for your rating” and “Thank you! We'll get in touch”.
All contacts together
The list of all contacts customers have left and a summary of ratings are with the builder: “show the contacts”, “how are answers being rated”. Only the assistant's answers can be rated: there are no rating buttons in the chat window under your messages or under replies you approved in training mode.
What you see when someone else closed the conversation
- A colleague ended it — the conversation on your screen closes by itself, and the reply box becomes unavailable.
- The customer erased the conversation in their chat window (the bin in the window header) while waiting for a person or talking to one — the chat window first asks them again and says the operator won't be able to reply in this conversation any more; if they confirm, the request is closed, and the last line of the transcript has the note “The customer erased this conversation in their chat window and started a new one”. You can no longer reply in it: the customer doesn't have it.
- A question the assistant didn't answer — it crashed mid-turn or hit a plan limit — stays in the transcript without an answer. If the customer calls for a person after that, you can see what they asked.
- The assistant called a person by itself — it writes the reason as a note in brackets: “(Handing the conversation to a person. Reason: …)”.
While a person is talking, no credits are spent
Neither your messages nor the customer's cost anything during this time. The assistant comes back to the conversation after “End”.
So requests reach you outside the dashboard
You have to open the “Conversations” section — but customers call for a person when the tab is closed. So the request goes by itself to where you already are. Nothing to turn on: every route you have works.
| Where | What arrives | How to reply from there |
|---|---|---|
| Your chat with the assistant in Telegram (or another messenger), if you've linked it | a card with the request number: what the customer needs, their contact, the latest messages and a link to the conversation | start your message with the number: 123456: your text — it goes to the customer from the operator. беру 123456 (“taking”) marks the conversation as yours; готово 123456 (“done”) ends it. These two commands are recognised in Ukrainian |
| Support in Telegram — a separate bot for people on duty (below) | the same message to everyone currently on duty | by replying (Reply) to the message; the “Take” and “End conversation” buttons |
| Email — to the owner at the account address, to the agent's notification address and to team members who turned on emails for themselves | an email with the customer's question and a link to the conversation | if the agent has a mailbox it reads by itself — just reply to the email; otherwise — via the link in the dashboard |
| Your own ticket system (below) | an event about the request | through the operator API |
Where requests will reach you specifically is shown in the first line of the “Conversations” section — “We notify you about requests: …”. If there's no route at all, there's a warning instead: you'll only find out about new requests from this page. The builder says the same as soon as you turn on calling a person, and if you ask directly:
Where will I be notified when a customer calls for a person?
Everyone turns on emails for themselves. A team member: “Conversations” → “Change” (next to the notifications line) → “Email me”. The owner gets them by default and can turn them off in the same place or in words: “don't email me about requests”. Every email has a “stop sending” link.
One request — one message. If the customer keeps writing while waiting, the new messages arrive in the chat and in support as a short line, and by email only to whoever has already replied by email. If nobody takes or answers the request within 10 minutes, one reminder comes — and no more. You change the time in words:
Remind me after 5 minutes if nobody takes it
Don't remind me about requests
Quiet at night. Tell the builder when not to disturb you:
Don't disturb me from 10 pm to 8 am
During these hours nothing rings about requests. A customer who called for a person hears an honest reply — “it's outside working hours now, we'll reply after 08:00” — and when the quiet hours end, a request that's still waiting arrives as one message. In the dashboard's “Conversations” it's visible the whole time. To go back — “disturb me around the clock”.
To turn off request notifications completely — “don't notify me about requests” (or the switch in the same “Change” window): then they're only visible in “Conversations”.
The customer doesn't see who replied or from where
A reply from the chat, from email or from the dashboard looks the same to the customer: it was sent by “Operator”. Your messenger number and email address never reach the chat window on your site.
Support in Telegram for people on duty
When it's not one person replying but a shift, connect support in Telegram:
Connect support in Telegram, here's the bot token: 123456:AA…
This is a separate bot from the one your assistant answers in: there the assistant answers, here live people write. You get the token from @BotFather, just like for the assistant.
Then say who's on duty:
Olena is on duty Mon–Fri 9 to 6, and Petro on weekends
You won't be asked for numeric IDs: have everyone who's going to be on duty send the bot “/start” — and the builder will see them by itself. You can ask what the current setup is any time:
Who's on duty in my Telegram?
Connected doesn't mean “working”
A bot with no one on duty accepts requests and shows them to nobody. The builder will say so plainly rather than report “done”.
What the operator sees while the customer is typing
While a person is handling the conversation, the operator sees a draft — what the customer is typing, before it's sent. This helps prepare a reply in advance, and the customer is told so plainly in the chat window when they call for a person.
You can turn it off in the conversation: “don't show operators what the customer is typing”.
Find the right conversation
There's a search box above the list. It searches the whole correspondence, not just titles — because a complaint arrives as content, not a number:
a customer says the bot talked nonsense about returns
Type “returns” — and you'll see the conversations where that word really came up, including from the bot. Search works together with the filter: you can search among closed ones too. Conversations where a customer left a contact or was dissatisfied are found not by a word but by their mark.
If a customer sent a file
The chat window on your site has a paperclip: the customer attaches a photo of a receipt, a document scan or a screenshot right in the chat — or pastes it from the clipboard (Ctrl+V) or drags it onto the window (how exactly). While the file sits in the input box, it's not sent yet — the cross really removes it. It goes together with the message (or on its own, without a caption — that works too).
The same comes from messengers: in Telegram the customer just sends a file.
A “Customer sent files” block appears above the feed — each attachment opens separately, in a new tab.
The assistant also sees that a file was sent
It sees photos by itself and replies to the point. It reads documents too: PDF, Word, Excel, CSV and plain text — short ones in full, long ones it reads on and searches through; a scan without text — up to ten pages. Whatever it couldn't read (a password-protected file, an archive, a scan that's too large), it says so plainly. More — Photos and files.
Files don't go the other way
The operator's reply box has no paperclip: you can't send a customer a file from here. Things the agent should send customers regularly (a price list, terms, instructions) go into agent files — then the assistant sends them by itself.
If a customer asks to erase their data
An open conversation has “Erase data”. The correspondence of this conversation disappears, along with the contacts the person left and the assistant's memory of them.
This is irreversible — for you and for us
Erased data can't be restored. The dashboard will name the consequence before doing it, and will only do it after your confirmation.
You can also tell the builder — “a customer asks to delete their data”, naming the conversation: it will state the consequence and erase after your “yes”.
If you and your customers speak different languages
The bot replies in the customer's language by itself. But your replies as an operator go as written by default. When support writes in Ukrainian and customers write in English or Polish, tell the builder:
Translate my replies into the customer's language
From then on, each of your messages is translated before delivery; the customer sees it in their language, and you write in yours. You turn it off the same way (“don't translate my replies”).
It's off by default, for a simple reason: translation is a model call for each of your messages, which means credits. For most support teams that speak their customers' language, it adds nothing.
If you already have your own ticket system
A conversation handed to a person can be passed to your system — Zendesk, your own backend, anything:
Send operator handoffs to https://crm.example/hooks/youselfbot
A POST goes there for every handoff and every customer message during it, signed with a key shown in the dashboard. Your system can reply through the operator API.
Next
- When something breaks — to learn about requests without sitting in the tab.
- For developers — if requests need to be handled in your own support system rather than here.