Chats

Chats answers the question your team asks first each morning: is anything waiting for me? It gathers messaging conversations, email threads, and calls from every agent into one newest-first list, and marks the ones an agent has handed to a person. Instead of opening each agent in turn to check, you open one screen, see the whole day, and answer from there.

Navigation: Open Chats at the top of the left sidebar. The Inbox tab inside an Email Agent connects and manages that agent’s mailboxes; it is separate from Chats.

Chats appears for anyone who can already read at least one kind of conversation — messaging, email, or calls. It shows exactly the kinds you may read and nothing else, so two teammates with different permissions can see different lists on the same workspace.

Read the waiting count

The number beside Chats in the sidebar is how many conversations are waiting for a person right now. It is the single reason to keep Chats pinned where it is: when it is empty, nothing has been handed over and there is nothing to chase.

A conversation is counted when an agent handed it to a person and nobody has answered yet:

A messaging conversation counts while it is paused by a handover. Resume the agent in the conversation and it stops counting.
An email thread counts while the escalation is recent and no reply has been sent on that thread. Reply to the customer and it stops counting; an escalation nobody acts on also stops counting after 15 days, so an old backlog cannot keep the number permanently high.
Calls are never counted. A call does not record that it was handed over, so no call can be marked as waiting.

The count hides itself at zero. If it shows a number followed by a plus sign, either there is more waiting than the badge can display, or one kind of conversation could not be counted just then and the number is a minimum — hover it to see which.

The badge counts what arrived since you last opened Chats, so opening it clears the number and the next escalation brings it back. The queue itself is not cleared: Needs a human in the filters still shows everything waiting, however long it has been there. The badge is remembered per browser, so opening Chats on your laptop does not clear it on another machine.

Find a conversation

The filter bar above the list narrows it down:

The search box at the top finds a person. Type a name, an @handle, an email address, a phone number, or a call’s conversation id.
The channel buttons under the search box are the quickest way in. Most of the time you are looking for one kind of conversation — the Instagram messages that came in overnight, or just the email — and one click gets you there. There is one button per channel type you use — two Instagram accounts share a single Instagram button — and Email and Calls sit beside them. Pick several to see them together, or All to go back to everything.
Everything else sits behind Filters, which shows how many are active. What you set appears as a row of chips under the search box; remove one with its ×, or use Clear all.
Needs a human shows only the conversations counted in the sidebar. Because calls cannot be marked as waiting, they are left out while this is on, and the list says so.
Unread shows conversations with a customer message nobody has read. Only messaging conversations have a read state, so email and calls are left out while it is on.
Conversations only hides email threads that are a single message nobody ever acted on — useful when a busy mailbox buries the threads that went somewhere. It is off by default.
The agent and channel lists are grouped by kind, so an agent or a connected account can be picked directly. Both belong to one kind of conversation, so choosing one leaves the other kinds out and the list says which.
The status list is grouped by kind, because a status does not mean the same thing across messaging, email, and calls. Choosing one narrows the list to that kind, and the strip above says which kinds were left out.
From date and To date limit the list to a period.

If a channel is disconnected while you have it selected, its button stays where it is so you can click it off — the list would otherwise stay narrowed by something you could no longer see.

A strip above the list always states what the search matched on and which kinds were left out and why. That is deliberate: an empty list should never leave you guessing whether the customer is missing or the search simply could not reach them.

What each kind can be searched by

Messages — contact name, handle, platform id, and any customer name, email or phone the agent captured during the conversation.
Email — the sender’s exact address, or a fragment of a sender identity the agent captured behind a contact form. Written addresses and subjects are encrypted, so nothing else reaches an email thread. A match anywhere in the thread counts, so a thread your colleague answered last is still found by the customer who started it.
Calls — the caller’s number, or the call’s exact conversation id. A call records nothing else about who called.

Phone numbers are matched as digits, so spacing and punctuation never matter: +44 20 7946 0000, +44 (20) 7946-0000 and 0044 20 7946 0000 all find the same person. The country code does matter — a number typed in national form, like 020 7946 0000, will not find one stored internationally. Type a number the way it was captured, in full; a partial number and a partial email address both find nothing.

Each row shows who wrote in, which channel and connected account it arrived on, which agent is handling it, the latest message, and its current state — for example Manager replied, Draft ready, or Escalated. Rows the agent handed over carry a marker of their own.

For email, a status filter asks whether the agent ever did that to the thread, while the row shows the most recent state. Filtering by Escalated can therefore return a row labelled Manager replied — that is the real sequence of events: the agent escalated it and a colleague has since answered.

The list loads in pages and keeps loading as you scroll towards the bottom. If a page fails to load, a Load more button appears at the end rather than the list retrying on its own. Refresh (an icon on a phone) reloads it, and the list also refreshes when you return to the tab.

Before you pick one

The right-hand side starts empty, and it tells you where the work is: a line says how many conversations are waiting for a person, and pressing it filters the list down to exactly those. It is the fastest way to start a shift. When nothing is waiting, it says that too.

This line and the count on the sidebar’s Chats entry are deliberately different numbers. The sidebar counts only what arrived since you last opened Chats — it is a notification, and opening Chats clears it. This line is the whole queue, the same set the Needs a human filter shows, which is why pressing it takes you straight there.

The line counts the whole conversation list, so it only appears while no filter is set — with a channel button or a date range active it would be counting something other than what you are looking at. If a kind of conversation could not be counted, the number is shown as a minimum rather than quietly leaving that kind out.

Read a conversation

Select a row and the conversation opens beside the list: the whole exchange in order, with the customer on one side and everything sent from your side on the other. On a phone — held either way — or a tablet held upright, the list fills the screen, and the conversation replaces it when you open one so you have room to read and reply; the arrow at the top-left of the conversation (Back to conversations) returns to the list. Photos, videos, voice messages and files a customer sends appear in the thread itself, so you can look at them without switching to the agent’s own inbox. The agent’s replies and a colleague’s are shown separately and labelled — in messaging conversations a colleague’s reply carries their name, and simply reads Staff where the record does not hold one — so you can see at a glance whether a person has already answered. New sessions, “your team was notified”, a handover the AI made without replying to the customer, and delivery failures appear as their own lines between the messages rather than as replies — see See why the AI handed a conversation to your team. A message the customer later removed on Instagram stays in place with a Deleted by customer label under it — see See when a customer removed a message. The date you are reading stays pinned at the top while you scroll. When a sender is blacklisted, the header says so with a red Blacklisted button and a line under it. Read that line: unlike pausing the AI on a messaging conversation, a blacklist applies to that sender everywhere — every thread, every mailbox in the workspace.

The contact’s name is at the top beside their picture, with whatever address the agent captured underneath it, when there is one. A contact nobody could identify at all is simply named Unknown. Below that a short strip gives what the conversation records, starting with the agent that owns it: then the channel and account, whether automation is paused and why, any phone number captured, the mailbox and classification for email, or the duration, numbers, outcome and summary for a call. The actions you can take are described under Act on a conversation.

Very long conversations open at the recent end, and the pane says so when older messages are not shown.

The address bar keeps the conversation you are reading, so reloading the page returns to it and the link can be sent to a colleague. Someone opening that link sees the same conversation even if their filters would not have listed it.

What the outcome badge means

Once a messaging session has ended and been reviewed, a badge beside the contact’s name says how it went, and a Summary line in the strip below gives the gist in a sentence or two. It saves reading the whole exchange to find out whether the customer was actually helped.

The badge names the session it belongs to, so on a conversation that has started again since, it describes the last completed session rather than the one in progress. Hover it to see which session it is, who decided the outcome, and — where the conversation was handed to a person — why and how that part ended.

The outcomes are:

Resolved — the customer got what they came for and said so.
Assumed resolved — the customer left after an answer that closes the question, without saying it helped and without complaining.
Handed to team — the conversation ended in your team’s hands: the agent handed it over, a colleague stepped in, the agent told the customer to contact your team another way, such as by email or phone, or it promised that someone would follow up using contact details the customer had already left. Only asking for contact details does not count.
Left unresolved — the customer left dissatisfied with the agent’s replies: refusing what it asked, such as their contact details, complaining about its answers, repeating themselves or saying that was not what they asked, even if the agent then told them to contact your team another way. A complaint that is the request itself, such as a late order, does not count.
Abandoned — the customer never received a real answer, for example they said hello or asked something and left after the agent asked a question back, such as for their contact details.
Not a request — nobody is asking for anything: a story mention, a bot, a lone emoji. A greeting and an offer of a partnership or a sale are requests.

A badge ending in · check is one nobody has confirmed yet: the review was unsure, or something about the conversation is worth a second look. It is a suggestion to glance at it, not a problem.

Correct an outcome

If a badge is wrong, click it and set the right one. Your correction replaces what the review concluded, is what the Conversation Quality page counts from then on, and clears the · check mark.

Two fields appear only where they apply. Choosing Handed to team asks why the agent could not finish the conversation, and — where a colleague replied — how their part ended. Where nobody replied, it is recorded as unhandled without asking.

Moving the outcome away from Handed to team is allowed even on a conversation that was handed over: the handover stays on the record as something that happened, and the outcome describes how the conversation ended, which can be a different answer. Its handover-only fields are cleared when you save.

Correcting an outcome needs permission to manage messaging agents. Without it the badge is still shown — it just cannot be clicked.

Open a completed callback from Agent Tasks

When a Voice Agent callback is complete, open Agent Tasks and select See in Chats on its task card. Chats opens that call directly, where you can listen to its recording and read its summary alongside the rest of the conversation history.

Act on a conversation

You answer from Chats — there is no need to open the agent first. On a phone, on most tablets, or in a narrow browser window the conversation header keeps its main control on the row — AI Active / AI Paused for messaging, AI Active / Blacklisted for email — and the rest of the actions sit under the three-dot button at the right of the header (More actions).

Below the messages is a reply box. Type your answer and press Enter to send, or Shift+Enter for a new line. On channels that reply from a connected account, the box names it above the text (“Replying as …”) so you always know which identity the customer will see. Long answers are easier to write with a taller box: drag the handle above it.

The reply box appears only where you have permission to send. If it is absent, you can read the conversation but not answer it — ask an administrator for the matching permission on that agent type.

Two things are worth knowing about sending:

Your reply is kept if sending fails. The box does not clear, so you never have to retype an answer because a channel was briefly unreachable.
An email reply that neither succeeds nor fails clearly is not sent again for you. You are told to check the mailbox before sending it a second time, because an automatic retry is how a customer receives the same answer twice.

When the agent has already written a draft

On an email thread where the agent prepared a reply and left it for review, the box shows AI draft ready with a Use draft button. The draft is offered, never loaded — the box stays empty until you press it, so nothing goes out under your name that you have not read. Press it, edit as much as you like, then send.

Sending your own reply closes the agent’s draft in the mailbox, so a stale answer is not left behind for someone to send later by accident. On threads answered before drafts were tracked individually, the draft cannot be found automatically; you are told so and can delete it in your mail client.

What else you can do from here

The strip above the messages carries the actions that make sense for that kind of conversation:

Messaging — pause or resume the agent (AI Paused / AI Active), Exclude from AI to stop the agent answering this contact anywhere (Excluded from AI once it is), Ignore auto-pause so the agent keeps answering even after a person replies (Follows the agent setting when it does not), and New session to close the current session and start a fresh one.
Email — AI Active to blacklist the sender (Blacklisted once it is; pressing it again releases them), Classify (Classified: … afterwards), AI reply to have the agent write one now — drafted or sent according to that category’s reply mode — CRM lead to create the opportunity, and, once the agent has replied, QA feedback to open that reply with its feedback form.
Calls — a call is read-only. Its recording plays in the pane; there is nothing to answer.

You are only offered what you have permission to perform: an action you may not take is absent rather than shown greyed out. Some actions ask you to confirm first.

Opening a messaging conversation here marks it as read, exactly as opening it in the agent’s own inbox does, so the two never disagree about what your team has seen.

To reach the agent’s own screen — its settings, its full history, everything Chats does not show — use the arrow icon at the end of the strip above the conversation; hovering it reads Open in agent inbox. It opens in a new browser tab, so the conversation you were reading stays where it is. For an email thread that is the Email Agent’s inbox, which looks and works the same as this page, scoped to that one agent. When it is greyed out, the reason appears at the foot of the pane, below the reply box. The usual cause is a shared mailbox connected to more than one Email Agent: with no single owning agent there is no single place to send you, and the row shows the mailbox address instead of an agent name.

Troubleshooting

Ask AI is missing in Chats

The Ask AI support widget is hidden while you work in Chats so it does not cover conversation controls. To ask for help, open another page such as Knowledge Base, then select Ask AI. Returning to Chats hides the widget again.

A conversation stopped being counted after someone replied

That is deliberate. The count is conversations an agent handed to a person that nobody has answered yet, so a colleague’s reply takes one off it — the agent stays paused on that conversation either way, because a person has taken it over. If the customer writes again and still has no answer, it counts again.

The waiting count does not match the list

The two answer different questions: the sidebar count is what arrived since you last opened Chats, while Needs a human lists everything still waiting. A filter list longer than the badge is normal — that is the older backlog. If the list is shorter, check whether a date range or a channel button is still active.

Part of the list is missing

A red strip appears above the list when one kind of conversation could not be loaded, naming the kind and offering Retry. While it is there, the list stops growing as you scroll — continuing would put older conversations above newer ones once the missing ones come back. Retry, or use Refresh.

An email thread is not listed

Chats lists an email thread once the mailbox is connected to an Email Agent and a customer has written in. Outbound-only threads — announcements and campaigns nobody replied to — are not conversations and are left out. If a mailbox was connected recently, its first sync may still be running. Check whether Conversations only is on: it hides single messages nobody acted on.

An email reply is labelled “Sent from this mailbox”

On threads answered a long time ago, the record does not say whether a particular reply was the agent’s or a colleague’s. Rather than guess and put words in someone’s mouth, Chats shows that it was sent from your mailbox and leaves the author unnamed. Recent replies are always attributed.

Searching a name finds no email

A name finds an email thread only where the agent captured who really wrote in, such as the sender behind a contact form. Written names and subjects are encrypted, so for ordinary mail use the exact address.

A call is not listed

Calls whose Voice Agent has since been deleted are not listed here. Find them under Phone → Call Logs & Recordings.