Inbox settings

In the panel, Inbox › Inbox settings sets how conversations run: when they close, whether a quiet customer is asked once, the customer reminder, translation of incoming messages, the spam filter, blocked visitors and your agents' saved replies.

These settings apply to every channel, not just the bubble: the chat bubble, email, WhatsApp, Instagram and Telegram. The cards appear in the panel in this order: Closing conversations, Customer reminder, Incoming message translation, Spam filter, Blocked visitors and Saved replies. Until 30 October 2026 these settings lived on the Widget Settings screen; their values moved over unchanged.

Defaults: the spam filter, asking the AI about suspicious messages, translating messages in other languages automatically and saved replies are on; until you pick one, the translation language is each agent's own panel language. Auto-close, checking in when the customer goes quiet and the customer reminder stay off until you turn them on.

Closing conversations

You choose what happens when the customer goes quiet: ask once, close the conversation automatically, or both. The two settings look at the same silence.

Auto-close idle conversations

Closes open conversations with no new message for the time you choose. The options: Off, 12, 24, 48 or 72 hours, 5 or 7 days. Conversations an agent has taken over, and escalated ones, are left alone; a conversation reopens by itself if the customer writes again. Auto-closed conversations are counted separately in reports and don't enter the resolution rate.

Check in when the customer goes quiet

If you spoke last and the customer stays silent for the time you choose (15 or 30 minutes, or 1, 2 or 4 hours), the chat gets one message: “Is there anything else I can help you with?” It works in the chat bubble and on WhatsApp, Instagram and Telegram; on WhatsApp and Instagram only while the 24-hour reply window is open. It isn't sent by email (there is the reminder email for that), and never on handed-over conversations.

The question goes out in the customer's language (in the chat bubble, one of the bubble's nine languages; on WhatsApp, Instagram and Telegram, Turkish or English) and shows in the chat as Automatic check-in. If the customer writes again, it is asked once more after the next silence. The options make sure the question always goes before the conversation closes: the longest check-in is 4 hours, the shortest auto-close 12 hours.

Customer reminder

If you spoke last and the customer hasn't come back within the time you choose (12, 24, 48 or 72 hours, 5 or 7 days), they get one reminder email. Only for customers who left an email; never on handed-over conversations (there the reply is owed by you). The reminder is written into the conversation as a line and is sent at most once per conversation.

The time must be shorter than the auto-close time; otherwise the conversation closes before the reminder goes out, and the card warns you about it.

This is the reminder that reaches the customer. The reminder an agent sets for themselves is separate: Conversations › Remind me.

Incoming message translation

Which language ⋯ › Translate on a customer message in Conversations translates into. The default is The agent's panel language; pick one of 39 languages and the whole team uses it. An agent can pick another language for a single message; this setting does not change. Translations are shown only to the agent, never sent to the customer.

When Translate messages in other languages automatically below the card is on (the default) and a customer writes in a language other than your team's, the translation of their messages appears in Conversations by itself. Messages already in your team's language aren't translated; each message is translated once and kept. An agent can turn automatic translation off for a conversation or pick another language for it. Details: Conversations › Message translation.

Spam filter

Screens out ads, gambling, backlink and scam messages before they reach your inbox. You can turn it off with the switch on the right of the card; while it's off, every message is processed normally. While it's on, the settings below it are:

• Ask the AI about suspicious messages (on by default): a message containing one of the signal terms is judged by the AI. When off, only the certain terms block; suspicion alone never does.
• Words never counted as spam: if a word on the filter's list relates to your business, add it here; a message is no longer treated as suspicious or blocked because of that word (for example “lottery” if you sell lottery tickets, or “bitcoin” if you work with crypto). Other words in the message and signals such as a pile of links are still checked.
• Your own blocked terms: a message containing any term you add here is blocked outright. One term per line, up to 60 terms.
• Show the built-in list: lists the terms that block on their own and the terms that only raise suspicion (the AI decides).

Terms are matched as whole words: the term “crypto” does not match inside “cryptography”. A visitor who has written to you before is never blocked. Every blocked message is written to the Audit log; the card says how many messages were blocked in the last 30 days and links to the Audit log.

Blocked visitors

Visitors you blocked in Conversations are listed here: the type of block (session, email, IP), the masked value, the date and who blocked them. Remove a block with Unblock; it is saved right away and lands in the Audit log. How a block is placed and what it covers: Conversations › Block a visitor.

Saved replies

Saved replies are the answers your agents keep writing in Conversations. The agent who has taken over a conversation picks one from the Saved replies button on the composer's toolbar (the lightning icon) or by typing / in the box: keep typing to search (e.g. /refund), pick with the arrow keys, press Enter or Tab to insert, Esc to close. The text goes into the box, and the agent can edit it before sending. The customer sees it as the agent writing, not the assistant. Details: Conversations › Composing tools.

The Let agents use saved replies once they join a chat switch at the top of the card is on by default. The list and the suggestions only show while it's on; turning it off doesn't delete the saved texts.

Each saved reply has a title (the short name an agent sees in the list), content (the text that goes into the chat) and a category. The built-in categories are General, Shipping, Return, Order and Payment; choose + New category in the Category picker to add your own (up to 50 characters; typing an existing name adds to that category). Categories only organise the list; the customer never sees them. The variables {{ad}} (name), {{eposta}} (email) and {{temsilci}} (agent) are filled from the conversation when the reply is inserted; a field with no value stays as-is and you get a warning.

Every reply in the list has its own Active / Inactive switch: an inactive reply stays in the list but is hidden from agents, a way to shelve one without deleting it. New saved reply adds one. While the list is empty, the screen shows the first three steps and Add sample replies adds five examples (welcome, shipping status, return policy, payment declined, closing the conversation); you can edit or delete any of them afterwards. The samples are a starter set; they aren't added while the list has replies.

Language versions

When adding or editing a reply, Add language version lets you write the same reply in another language: pick the language (the 39 languages of message translation) and enter the text. A reply can have up to 12 versions; an empty version isn't saved. In Conversations, when an agent picks the reply and a version exists in the customer's language (for a customer writing from the chat bubble, the bubble's language), that version goes into the box and the panel says which language it inserted; otherwise the reply's own text is used. Variables work in versions too. In the list, the languages appear next to the title of a reply that has versions (e.g. EN · DE).

Suggested saved replies

Suggested saved replies below the list shows the replies your agents wrote in chats, the most repeated first, with how many times each was written and which question it answered. Add to saved replies lets you edit and save one; Dismiss hides it (that reply isn't suggested again). A suggestion that may carry personal data is marked May contain personal data; remove names, order numbers and emails before saving. Nothing is added until you save it. Add to saved replies in the ⋯ menu of an agent's message in Conversations adds to the list too.

When is it saved?

The closing, reminder, translation and spam settings, and the switch that turns saved replies on and off, are saved together with the save bar at the bottom of the page: once you change something, “N unsaved changes · Discard · Save” appears at the bottom, the changed card is marked Unsaved, and the panel asks before you leave the page without saving. Changes in the saved replies list (adding, editing, active / inactive, deleting, saving or dismissing a suggestion) and unblocking a visitor are saved right away with their own buttons.

Who can change it?

Anyone whose role has view or edit access on the Widget settings row can open the page. Changing the closing, reminder, translation and spam settings needs edit access on the Knowledge assistant row; changing the saved replies list and the blocked visitors needs edit access on the Widget settings row (Users & roles). A role that can only view (for example Support agent) reads the settings; the fields are locked and the server refuses changes too. A Support agent uses saved replies in chats.