Answer visitors in the inbox

Every message a visitor sends from your widget lands in one place: Conversations. Setting the product up takes an afternoon; this screen is where your team then spends its days, so it is worth ten minutes.

One thing to know before anything else: the inbox is shared. There is no private queue. Every teammate — operators included — sees every conversation in the workspace, and the filters are how each of you narrows that down to your own work.

What a row tells you

Each row carries who spoke last and a preview of it, when, the conversation's status, who it is assigned to, and whether somebody owes it a reply.

That last one is the one to read first. Needs reply means the customer spoke last in a conversation that is not closed — or somebody on your team flagged it. Replied means the last word was yours.

A third value appears only in workspaces whose AI agent answers visitors: Needs a person. The agent has told the visitor it could not answer and that a colleague is coming, and this badge is where that promise reaches your team. It replaces Needs reply on that row rather than sitting beside it — it is the same waiting said more precisely — and it comes down when an operator replies or the conversation is closed. Without the agent there is nothing to hand over, and rows stay on the two values above.

The assignee cell can name the agent too. In an agent workspace, a conversation the agent is answering with nobody assigned to it reads AI agent rather than Unassigned — nobody has to pick it up, it is being handled. The moment the agent asks for a person, the cell goes back to Unassigned and the row says Needs a person: that one is yours.

An automatic answer is not a reply. If your widget answers visitors automatically (the Auto-reply card on the Widget page, on by default), the row still says Needs reply and still previews what the customer wrote. They have been promised an answer, not given one — so nothing about it moves the badge, the filter or the waiting-visitor alert.

There are three statuses, and only the first two count as open:

Awaiting operator is never set for you. Nothing moves a conversation into it automatically; an operator picks it by hand. A team that never uses it will never see it, and that is a perfectly good way to run an inbox.

Finding the conversation you want

Three filters, and they combine:

• Status — open (Active and Awaiting operator together), or everything.
• Reply state — Awaiting reply, and Needs a person for the handovers above. The second is a narrower question inside the first, so pressing both gives you the handovers.
• Assigned to — Mine, Unassigned, or both at once. "Mine or nobody's" is the useful daily filter, and it is where an operator now lands: both chips are on when you open the inbox, which is the same question the sidebar badge counts. Owners and admins still land on the whole workspace.

Search is not a fourth filter, and the screen says so while you use it: it looks at every conversation in the workspace, and your filters are paused rather than lost. Clear the search and they come back.

The count above the list is that same query counted, so the number and the rows can never disagree. The badge beside Conversations in the sidebar is a narrower question: conversations awaiting a reply that are yours or unassigned — not the whole workspace's. A teammate's conversation going unanswered does not move your badge.

Who a new conversation lands on

When a visitor starts a conversation it is assigned automatically, in rotation, to whoever is signed in to the console at that moment. Turn by turn, so two people online share the arrivals instead of one person taking them all. Role plays no part: an operator is as likely to be picked as an owner.

Two things follow from "at that moment":

• If nobody is online, the conversation arrives unassigned — and stays that way until somebody takes it. It is not held for whoever signs in next. The Unassigned filter is how you find these, and the waiting-visitor alert is what tells you they are there.
• An assignment is not exclusive. Anyone can open and answer any conversation. It says who is expected to reply, not who is permitted to.
• A handover from the AI agent returns the conversation to the queue. The rotation picked its owner before anybody knew whether a person would be needed at all; when the agent says one is, that guess stops being useful. The conversation becomes nobody's, so the whole shift can see it and take it, rather than sitting with a colleague who has never opened it.

Answering

Open a conversation and the thread is there, newest at the bottom, with Load older messages at the top of a long one.

An auto-reply your workspace sent sits in the thread like any other message, tagged Auto-reply so nobody reads it as a colleague's words. It carries no pencil: nobody typed it, and there is nothing to correct.

The same is true of a customer who wrote to your workspace's email address: the first letter of a new thread is answered automatically too, and that answer leaves as a real letter, in the thread they already have open. Only the letter that starts a conversation gets one — a reply inside a thread you are already handling does not, and one correspondent is answered at most once a day, so two autoresponders cannot end up writing to each other.

The composer has two tabs, and the difference between them matters more than anything else on this screen:

• Reply goes to the customer.
• Internal note does not. It is your team talking to itself on the thread — every teammate sees it, the visitor never does, on any channel.

The rest of the composer, briefly:

• Enter sends; Shift+Enter starts a new line.
• Up to 5000 characters per message.
• Up to 3 attachments per message — dropped on the panel, pasted, or through Attach — uploaded one at a time. Up to 5 MB each: JPG, PNG, GIF, WEBP, PDF, DOC, DOCX, XLS, XLSX, CSV and TXT. Anything else is refused; there is no archive format on that list, because nothing scans inside one yet.
• Notes are text only. Drop a file on a note and the screen tells you to switch to Reply.
• A saved answer can be dropped into the box from your canned responses, and is ordinary editable text once it is there.

All of it is live. Their reply appears without a refresh, you see when they are typing, and they see when you are.

Fixing a reply you have already sent

For 15 minutes after sending, you can edit your own reply or your own note. After that the pencil is gone — the thread is the record of what was actually said to a customer, and an unbounded edit would make yesterday's conversation something anyone could rewrite today.

Three limits inside that window:

• Your own messages only — not a colleague's, and never the customer's.
• Not a reply that went out as email. That one is already in somebody's mailbox, and editing the copy here would leave the two of you holding different versions of the same answer, with only theirs being real.
• An internal note on such a thread stays editable, because it never went anywhere.

Handing a conversation over

Transfer moves it to a named teammate. Return to queue leaves it unassigned for anyone to pick up. Take it claims it for yourself — it is beside Transfer whenever the conversation is not already yours, and it is the one to use on a conversation the agent handed over. Each of the three leaves its own line in the thread — "Anna transferred this conversation to Boris", "Anna returned this conversation to the queue", "Anna took this conversation" — which your team sees and the customer does not.

Taking a conversation is not answering it. A row the agent handed over keeps saying Needs a person until somebody replies or closes it; claiming it says who will write that reply, not that it has been written.

Only an active teammate can be a transfer target; somebody deactivated is not on the list. And when a teammate is deactivated, their open conversations return to the queue on their own. Their closed ones keep their name, because there it is the record of who dealt with it.

Flag, close, reopen

Flag means "come back to this". A flagged conversation counts as awaiting a reply and wears a chip in the list, however long ago the last message was.

It comes down in exactly three ways: you unflag it, you reply, or you close the conversation. Adding a note does not lower it — a note written mid-investigation is precisely what "I will come back to this" looks like.

Closing is not deleting or archiving. The conversation stays, readable and searchable; it just stops counting as waiting on you. Reopening it is setting the status back to Active — but the flag does not come back with it, so a reopened conversation does not resurrect a reminder your team had already put down.

Closing also stops the AI agent answering in that conversation: an answer it was still composing when you closed the thread is thrown away rather than written into a conversation you have finished.

What the visitor sees

None of your bookkeeping. Statuses, assignments, flags, transfers and internal notes are all yours alone. The visitor sees the conversation, your replies and your attachments, in the widget where they started.

Come back later in the same browser and their earlier conversations are under History in the widget. There is no sign-in on their side, so clearing site data or moving to another device starts them fresh.

The other channel: email

A conversation can also arrive as a letter rather than from the widget, and such a row is badged Email in the list. Answering one works like any other conversation — with the edit caveat above, since that reply really does leave as mail.

Your workspace has an address of its own for this, shown on the Integrations page. A letter sent to it becomes an ordinary conversation in your inbox, and your reply goes back to the customer as an email from that address.

The point is not that customers write to it directly — it is that you forward your existing support mailbox into it. The Integrations page walks the three steps: turn on forwarding in the mailbox your customers already write to, point it at the address above, then send yourself a test letter. The card turns green the moment the first letter lands, so you can tell whether forwarding actually works rather than guessing.

Two things to know before you announce it to anyone:

• Replies reach your customer from the Replium address, shown under your workspace name, not from your own domain — we cannot sign mail for your domain yet. Their mail client still files the answer into the thread they started, so the conversation reads normally on their side.
• Files a letter brings come through into the message. Anything the workspace will not accept is not dropped in silence — the operator is told what did not make it.

If the Integrations page says receiving mail is not switched on for this installation, there is no address to hand out and nothing to configure; write to us.