SquidSupport
Product reference / Inbox

Inbox

In the app: Inbox

Needs the 'View Conversations' permission.

The Inbox is where your team reads and answers customers. Email, live chat and WhatsApp conversations from every store arrive in one list. The page has four parts: the views on the left, the conversation list, the open conversation with its reply box, and a details panel with the customer, their orders and what the AI did.

Views

The left column lists views. A view is a filter on the conversation list; clicking one shows the conversations that match.

The number beside a built-in view counts matching conversations across every store you can open and every channel. Saved-view counts can include conversations in stores you cannot open, so their numbers can exceed the tickets available to you. It does not follow the store filter or the channel tabs above the list, so the list can be shorter than the number. A view with nothing in it shows no number, and "Flagged by AI" never shows one. The numbers refresh every minute.

The built-in views:

  • "All Open": every open conversation.
  • "Unassigned": open conversations with no assignee. A conversation that only has a team, and no person, is still in it.
  • "Assigned": open conversations with an assignee, whether that is a person or the AI agent.
  • "Snoozed": every snoozed conversation. When snoozed conversations without an assignee exist, "Unassigned Snoozed" appears under it.
  • "Assigned to Me": open conversations assigned to you.
  • "My Snoozed": snoozed conversations assigned to you.
  • "AI Drafts": open conversations where the AI wrote a reply that waits for a person to approve it.
  • "Closed, AI-handled": closed conversations that the AI answered, that no person closed and that are not assigned to a person or handed to a team. A ticket you closed yourself is not in it, even when the AI replied on it.
  • "Flagged by AI": open conversations the AI left open for a person to look at instead of answering. The AI marks them with the tag "ai-flagged"; removing that tag takes a conversation out of the view.
  • "AI Outage": conversations where the AI could not respond because of an outage. They can recover on their own when processing resumes.
  • "Closed": every closed conversation, resolved ones included.
  • "Resolved": closed conversations whose latest close was a resolve. They are in "Closed" too.
  • "Failed Emails": conversations with a reply that bounced or was rejected by the customer's mail server.
  • "Trash": deleted conversations. Shown to people who can delete, restore or permanently delete conversations.

A download icon appears when you hover over "All Open", "Unassigned", "Assigned", "Snoozed", "Assigned to Me", "My Snoozed", "Closed" or "Resolved". It downloads that view as a CSV file, for every store you can open and every channel, whatever the store filter and channel tabs say. One exception: the icon on "Assigned" downloads every open conversation, unassigned ones included. A file holds at most 10,000 tickets. It needs the 'Export Conversations' permission.

The arrow at the top collapses the views column; "Expand sidebar" brings it back. Drag the edge of a column to resize it. Sections and folders can be folded, and the app remembers this in your browser.

Saved views and folders

Needs the 'Personal Views' permission.

Under the built-in views are "Shared Views", which the whole team sees, and "My Views", which only you see. Using saved views needs the 'Personal Views' permission. Creating, changing or deleting a shared view or shared folder also needs the 'Shared Views' permission.

The plus button opens a menu with "New view" and "New folder" for each of the two groups.

"Create New View" asks for:

  • "View Name";
  • "Private" or "Shared with team";
  • "Folder (optional)", shown when a folder of the chosen kind exists;
  • "Filters": one or more conditions, all of which must match. "Add Filter" adds a condition. A view needs at least one filter. Conditions that compare a value need that value before you can save. Conditions such as "Is empty" and "Is not empty", and the AI enabled, escalated and ignored conditions, need no value.

A new view starts with the conditions of the list you have open: the status and assignee of the built-in view, the channel tab and the mailbox you picked, or the conditions of the saved view you are in. The store filter above the list is not copied. When nothing can be copied, it starts with one condition: status is open.

You can filter on: channel, language, the mailbox a ticket was received at, store, assignee team, assignee user, customer, tags (contains all of, one of, or none of), priority, snooze, status (open, closed or resolved), the dates a ticket was created, updated, closed, last had a message or last received a message (before, after, or more or less than a number of minutes, hours, days or weeks ago), message count, AI Agent feedback (positive or negative), CSAT score, whether the AI is switched on for the ticket's store, whether the AI escalated the ticket, whether the AI ignored it, hallucination detection (none, rewritten, blocked) and your custom ticket fields.

Right-click a saved view, or use its "More options" button, for: "Edit" (name, sharing and filters), "Export CSV", "Rename", "Move to" a folder or "No folder", and "Delete". A folder has "Rename" and "Delete". "Delete" acts at once, without a confirmation, and cannot be undone. Deleting a view removes the filter only; the conversations stay. Deleting a folder keeps the views that were in it.

Switching "Newest" / "Oldest" in the list while a saved view is open stores that order with the view. The stored order is used when you export the view with "Export CSV". It is not applied again when you open the view later; the list keeps the order it had.

Conversation list

The middle column lists the conversations of the chosen view and loads more as you scroll.

Above the list:

  • The store filter ("All stores" or one store) limits the list to one store. The number on it is that store's open email and WhatsApp conversations; live chats are not counted, although the hint on the number says "Open email conversations". The choice is remembered in your browser and is the same one Home uses. A teammate with access to some stores only sees those stores here, and the first option reads "All your stores".
  • The plus button ("Create ticket") starts a new email to a customer. It is shown to people with the 'Create Conversations' permission.
  • The channel tabs "All", "Chat", "Email" and "WhatsApp" limit the list to one channel. The WhatsApp tab is shown when a store you can open has WhatsApp connected, or while WhatsApp is the selected channel, including from a saved view.
  • "Mailbox" limits the list to tickets received at one of your email addresses. It is shown only when there is more than one mailbox to choose from, and not on the Chat and WhatsApp tabs.
  • "Newest" / "Oldest" switches the order. The list is ordered by the time of the last message.
  • "Export" downloads the tickets in the list as a CSV file; see "Export tickets".

In Trash none of these controls are shown except "Newest" / "Oldest". The store you chose before opening Trash still limits the list there, although the store filter is not visible.

Clicking a row opens the conversation. Opening it marks it as read for you only; a teammate still sees it as unread until they open it.

When the list is empty, the page says why for that view, for example "Everything has an owner" in Unassigned or "No AI drafts waiting for you" in AI Drafts.

Badges and icons in the list

Each row shows the customer's name, the subject, the ticket number and how long ago the last message was. Around that:

  • Small avatar on the customer's picture: the teammate the conversation is assigned to.
  • Dot on the picture (live chat only): whether the visitor is on your site now or has left.
  • Eye chip with a name or number: teammates who have this conversation open. It turns into a red pen chip when one of them is writing a reply.
  • "SLA" in red: an SLA target was missed. A time in amber: the deadline is close. "SLA paused": the clock is paused, for example while snoozed.
  • "Escalated": the AI handed the conversation to a person. It also shows on any conversation that carries the tag "escalated". "· needs a human" is added while no person is assigned.
  • Robot icon: the AI is handling the conversation. It goes away once the AI has escalated the conversation.
  • Merge icon in place of the channel icon: the ticket was merged into another one. Merged tickets are left out of the list, so you will rarely see it.
  • Envelope, chat bubble or WhatsApp icon: the channel.
  • "Draft": you have an unsent reply saved for this conversation in this browser.
  • Coloured labels: the first three tags, then "+N" for the rest. On an escalated conversation the "escalated" tag is left out, because the "Escalated" mark already shows it.
  • Red bar on the left edge: the conversation has the tag "active_dispute", which Squid adds while the order behind it has an open payment dispute and removes when the dispute ends.
  • Red envelope with a warning: an email to the customer failed. Red triangle: a WhatsApp message was not delivered.
  • Blue dot on the right: you have never opened this conversation, or you marked it unread. It goes away when you open it. A new message alone does not bring it back.

Bulk actions

Needs one of these permissions: 'Assign Conversations' or 'Delete Conversations'.

Tick the checkbox on one or more rows to open the bulk bar above the list. It shows how many are selected. The checkbox at the left of the bar selects every conversation loaded in the list so far, not the ones further down that have not loaded yet; the cross clears the selection. The bar is not available in Trash.

  • The assign menu: assign the selected conversations to a teammate or to a team. A conversation is skipped when the teammate has no access to its store. Assigning to a team removes the person the conversation was assigned to. Needs the 'Assign Conversations' permission.
  • "Unassign": removes the person assigned to the selected conversations. Any assigned team stays. Needs 'Assign Conversations'.
  • "Send to AI": hand the selected conversations to the AI agent, after you confirm. There is no field for extra context here; that exists only when you assign one conversation to the AI. Conversations that cannot be handed over are reported back with the reason. Needs 'Assign Conversations'.
  • "Delete": move the selected conversations to Trash, after you confirm. A merged ticket is skipped. Needs the 'Delete Conversations' permission.

Assigning and unassigning from the bulk bar act at once. They do not ask for the takeover form or the escalation rating that an open conversation asks for, they do not reopen a closed conversation, and they add no line to the conversation. Assigning to a person creates an assignment notification for that person. These actions apply to the selected tickets for the whole team and can be changed again. They do not themselves send a customer reply; assignment automation rules can also run.

After each action a message says how many conversations were changed and how many were skipped, with the reason. A conversation with a delivery re-check pending is skipped with "Open the conversation to reassign it". With "Send to AI", a ticket already held by AI keeps its re-check instead of being skipped. See "Snooze, close, resolve and reopen".

Export tickets

Needs the 'Export Conversations' permission.

"Export" above the list opens the "Export tickets" dialog. It starts with the current view and filters. You can change the store and creation-date limits, including clearing them to include more tickets:

  • "Store": one store or "All stores". It starts on the store the list is filtered to.
  • "From" and "To": limits the export to tickets created in that date range.
  • "Include conversation history": adds the messages the customer could see as one extra column. Internal notes and internal lines such as assignments and snoozes are never included. History is cut off at 32,000 characters per ticket, followed by a truncation marker. If history cannot be loaded, the file still downloads with empty history cells and a warning.

The dialog estimates the number of tickets and the file size. An export holds at most 10,000 tickets, the first 10,000 in the order of the list; narrow the date range to export the rest in parts. The file is in the order the list is in. Exporting needs the 'Export Conversations' permission; without it the button is greyed out.

Create ticket

Needs the 'Create Conversations' permission.

The plus button above the list opens "Create ticket": you write to a customer first instead of waiting for them. Fill in:

  • "Store" and "From": the store and one of its verified sending addresses. A store without a verified address cannot send; the dialog links to Settings → Email.
  • "To", one or more addresses, and optionally "CC" and "BCC";
  • "Subject" and the message, both required;
  • "Tags" and your custom ticket fields, optional;
  • attachments, up to 10 MB per file.

"Create" queues the email for delivery to the recipients and opens the new ticket. Delivery happens in the background; "Ticket created" does not confirm it reached the customer. The ticket is assigned to you and gets the tag "human-created-ticket", so tickets your team started can be told apart from the ones customers started. The email cannot be unsent. Needs the 'Create Conversations' permission; without it the plus button is not shown. Creation is also refused while billing is inactive or the trial ticket limit is reached, and temporarily if the trial limit cannot be checked. To send the same template to many customers, use Outbound.

Snooze, close, resolve and reopen

The buttons at the top of an open conversation change its status. Each change adds a line to the conversation that says who did it, and the new status shows for the rest of the team at once. None of these buttons sends the customer an email or a WhatsApp message. On a live chat, "Close" and "Resolve" do tell the visitor's widget that the chat has ended.

  • "Snooze" hides the conversation until a time you pick: "1 hour", "4 hours", "Tomorrow" (24 hours), "2 days" or "1 week". It moves to the Snoozed views and its SLA clock pauses. It comes back as open when the time is up, when the customer writes again, or when someone on the team replies to it. Live chat conversations cannot be snoozed; email and WhatsApp can. Needs the 'Snooze Conversations' permission.
  • "Unsnooze" on a snoozed conversation brings it back now. Needs the 'Reopen Conversations' permission, not 'Snooze Conversations'.
  • "Close" ends the conversation. It moves to "Closed". Needs the 'Close Conversations' permission.
  • "Resolve" closes the conversation and records that the issue was solved. It shows in "Resolved" as well as "Closed", and a chat visitor's widget is told the chat was resolved. Needs 'Close Conversations'.
  • "Reopen" on a closed conversation makes it open again. Needs the 'Reopen Conversations' permission. A new message from the customer also reopens a closed conversation. A reply from the team on a closed conversation does not: it stays closed.

A button you have no permission for is greyed out and names the permission when you hover over it.

Surveys that go out after a ticket closes are set under Settings → CSAT.

When the organization has ticket fields that must be filled in before closing, the close is refused until they are filled in; see "Custom fields and required fields".

Delivery re-check. On a store that only accepts a return once the parcel is delivered, a customer can say the parcel arrived, or ask to return it, while tracking still shows it on its way. For email and WhatsApp, the AI then tells the customer it is confirming the delivery, snoozes the ticket and checks the tracking again 6 hours later. Live chat cannot be parked this way. Such a ticket shows "Snoozed until … for the delivery re-check". Unsnoozing it, replacing its snooze or giving it to someone else would cancel that check, so Squid asks first: "Delivery re-check scheduled". "Keep the re-check" leaves everything as it is; confirming cancels the re-check and leaves a note on the ticket.

Assigning a conversation

The assignee control under the subject shows who owns the conversation, and the team it belongs to as a label.

  • "Assign to me": takes the conversation yourself. People with the 'Claim Conversations' permission but without 'Assign Conversations' see only this button, and only on an open conversation that has no assignee. On any other conversation they see who it is assigned to and cannot change it.
  • The assign menu (with 'Assign Conversations'): search or pick under "Teams" and "Agents". "Assign to me", "Unassign from agent" and "Unassign from team" are at the top. The AI agent is not in the list; use "Assign to AI" for that. A teammate who has not accepted their invitation is marked "Pending invite" or "Invite expired"; work assigned to them may sit untouched.

What assigning does:

  • A conversation has a person or a team as its owner, not both. Assigning a person removes the team. Assigning a team removes the person, so the conversation shows in "Unassigned" until someone on that team takes it. "Unassign from agent" and "Unassign from team" each remove both.
  • Assigning a team also takes the conversation away from the AI: a reply the AI was working on is cancelled, and the AI does not pick the conversation up again when the customer writes.
  • Assigning a closed conversation to a person or a team reopens it. "Assign to me" for people who can only claim is refused on a closed conversation.
  • When the individual assignee changes, a line in the conversation records who assigned it to whom, or who unassigned it. Changing only the team does not always add a line.
  • A teammate you assign a conversation to gets a notification. You get none for assigning to yourself.
  • A conversation can only be assigned to a teammate who has access to its store.
  • The customer is not told.
  • It can be changed again at any time.

While the AI holds the conversation, or while an escalation still needs a rating, the assign control is locked; see the next section.

Replying can also assign: when "Enable Auto-Assign on Reply" is on under Settings → Organization, sending a reply to the customer assigns the conversation to you, also when it was assigned to someone else. An internal note does not.

AI handling, take over and escalations

A label under the subject of the conversation says what the AI is doing in it, together with the number of AI turns: "Waiting to respond", "Generating response", "Error occurred" or "AI did not respond", with a "Take over" button beside it. When a draft waits for approval the label reads "AI draft ready". When the AI has handed the conversation over, a line under the header reads "AI escalated this conversation to a human".

While the AI holds a conversation, a person cannot reply to the customer or change the assignee. The AI holds it when it is assigned to the AI agent, the AI has not escalated it, and the last message in the conversation is the AI's own. When the customer has written since and the AI has not answered yet, the reply box is open. Internal notes always work. To step in, use "Take over":

  • AI processing stops and any reply the AI was about to send is cancelled.
  • The conversation is assigned to you, or to the teammate you were assigning it to. The AI does not pick it up again when the customer writes.
  • A line "… took over from AI" is added to the conversation, with the reason when you gave one.
  • When "Require an explanation on manual takeovers" is on under Settings → Organization, you first rate the AI's response and pick a reason.
  • Needs the 'Assign Conversations' permission.

"Assign to AI", in the "More actions" menu, hands a conversation back to the AI agent. The dialog has an optional field for context, up to 1,024 characters, which is added to the conversation as a line "… added context: …" that the AI reads with the rest of the conversation. Needs 'Assign Conversations'.

What happens next: the conversation is assigned to the AI agent and its team is removed, it is reopened if it was closed or snoozed, the "Escalated" mark is removed, a draft of the AI that was still waiting is discarded, and a line with your name, "… assigned this conversation to AI", is added. When the last message is from the customer, the AI starts on a reply at once, without the store's reply delay. Otherwise a line says "AI will reply when the customer writes" and the AI waits for the customer's next message.

"Assign to AI" is refused in these cases, each with its own message:

  • The conversation is merged: "Cannot assign a merged conversation to AI".
  • The AI escalated it and nobody rated that yet: "Assign to AI" is left out of the menu until "Rate AI before replying" has been answered.
  • The AI is switched off for the store: "AI is not enabled for this store. Enable AI in store settings first."
  • The AI is working on the conversation right now: "AI is already processing this conversation". Try again when it has finished.

One case asks instead of refusing. When the AI promised the customer to check a delivery again and that check is still scheduled, Squid asks "A delivery re-check is scheduled at …, reassign anyway? Reassigning cancels the re-check." If the AI itself still holds the conversation while it waits for that check, "Assign to AI" keeps the re-check and says so.

For a selection in the list, the same rules are checked per conversation and the ones that could not be handed over are reported back with their reason. Three differences: there is no field for context, a conversation with a scheduled delivery re-check is skipped with "Open the conversation to reassign it", and the rating of an escalation is not asked for.

When the AI escalates a conversation, it steps back and the conversation waits for a person, marked "Escalated". Before the first reply or assignee change, Squid asks one question: "Rate AI before replying". You rate how the AI handled the ticket up to that point; a note is optional, except for a bad rating, where it is required. Submitting unlocks replies and assignment for everyone. If no person owns the ticket yet, submitting also assigns it to you; an existing human assignee keeps it. It is asked once per ticket, and answering it needs the 'Send Messages' permission. Assigning from the bulk bar does not ask for it.

When the AI did not run, a notice in the conversation says why: the AI is not set up for the store, it is switched off, it has no active prompt, or it stopped with an error. The notice links to the AI settings for people who can open them; others are told to ask an administrator.

Priority, tags and more actions

  • Priority: the label under the subject sets "Low", "Normal", "High" or "Urgent". A new conversation is "Normal". Priority can be a condition of an SLA policy and decides the order of "My Tickets" on Home. Needs the 'Change Priority' permission; without it the label shows a lock.
  • Tags: under "Conversation Tags" in the details panel. "Add tag" opens a list you can search; typing a name that does not exist offers to create it. The cross on a tag removes it. Needs the 'Manage Tags' permission. Tag colours are set under Settings → Tags.
  • "Summarize": asks the AI for a short summary of the conversation and shows it in a "Ticket Summary" panel. It is only made when you click, from the last 50 messages at most; nothing is sent to the customer and the summary is not saved with the ticket. After a failure the button shows a short wait before you can retry.
  • The "More actions" menu holds "Merge ticket", "Assign to AI", "Flag for QA" and "Delete conversation". On a narrow window "Snooze" and "Summarize" move into this menu too.
  • "Flag for QA": adds the ticket to the QA review queue. It is offered on closed conversations only. Needs the 'Submit QA Reviews' permission.
  • "Delete conversation": moves the conversation to Trash after you confirm. Needs 'Delete Conversations'.

Small labels under the subject add context: the store, in an organization with more than one; the status ("Open", "Snoozed", "Closed" or "Resolved"); "Escalated"; "Merged"; "Follow-up of #…" on an email ticket that continues a chat, and "Follow-up ticket #…" on that chat; the SLA state; and, when an A/B test is running, which test arm answered this conversation.

Merge tickets

Needs the 'Merge Conversations' permission.

"Merge ticket" in "More actions" joins two tickets of the same customer into one. Tickets of different customers or of different stores cannot be merged, and neither can a ticket that is already merged or in Trash.

  1. Pick the other ticket from "Suggested merges", each with a confidence of "High", "Medium" or "Low", or search for it.
  2. Under "Merge direction", choose which ticket stays as the main ticket.
  3. Choose which subject to keep, which answer to keep for each custom field where the two differ, and who is assigned when the tickets have different assignees: the assignee of either ticket, yourself or nobody. The AI cannot be made the assignee through a merge.
  4. Optionally add a note with the reason, then "Merge Tickets".

What happens: the messages of both tickets are combined on the main ticket, with internal notes and attachments unchanged, and the tags of the other ticket are added to it. The main ticket keeps its status. The other ticket is closed and hidden; it is listed under "Merge History" in the details panel of the main ticket. Any AI work that was pending on the other ticket is cancelled. When you change the assignee in the merge, the AI also stops on the main ticket. The customer is not told.

Undo. A merged ticket shows how many tickets it contains ("2 merged") and "Review Undo merge". The dialog shows what will be restored for the ticket you pick: its original messages return to it, messages added after the merge stay where they were added, and edits made after the merge are listed under "Keep these later edits", which you confirm before continuing. "Restore this ticket" brings the ticket back. AI drafts and runs that the merge cancelled are not restarted.

Merging and undoing a merge need the 'Merge Conversations' permission. Reviewing ticket fields in a merge also needs 'View Custom Fields'. Changing the subject in a merge also needs 'Edit Subject and Dismiss Bounces', changing the assignee needs 'Assign Conversations', and changing a custom field needs 'Edit Custom Fields'.

Trash

Needs one of these permissions: 'Delete Conversations', 'Restore Conversations' or 'Permanently Delete'.

Deleting a conversation moves it to the "Trash" view and adds a line "Conversation moved to trash". It leaves every other view and count, and nothing is sent to the customer. A merged ticket cannot be deleted.

In Trash, an open conversation has two buttons and no reply box:

  • "Restore" puts the conversation back with the status and assignee it had. Needs the 'Restore Conversations' permission.
  • "Delete Permanently" removes the conversation and its messages for good, after you confirm. Needs the 'Permanently Delete' permission.

"Empty Trash" at the top of the list permanently deletes every conversation in Trash at once, after you confirm. That covers every store you can open, not only the store the list is filtered to. It needs 'Permanently Delete' and cannot be undone.

The confirmation text when you delete says "You can restore it within 30 days". Squid does not empty the Trash by itself after 30 days: a deleted conversation stays in Trash until someone restores it, deletes it permanently or empties the Trash.

Message thread

Needs the 'View Messages' permission.

The middle of the page shows the conversation, oldest message at the top. Long conversations load in parts: "Load earlier messages", "Load later messages" and "Jump to latest".

What you can see in it:

  • Customer messages on one side and replies on the other, each with sender and time. An email also shows who it was sent to and, on a reply, the address it was sent from.
  • Who wrote a reply: a teammate's name; the AI agent; "AI draft · approved" for a reply the AI wrote and a teammate approved; "Saved FAQ" for a saved answer from the chat widget's FAQ buttons, which no AI wrote; "Templated outbound" for a message sent from an outbound template.
  • "Internal Note": a note for the team, in its own colour. The customer never sees it.
  • System lines: who assigned, snoozed, closed or reopened the conversation and when, and when a CSAT survey was sent or shown.
  • Quoted text: the earlier messages an email carries along are folded away. "Show quoted text" unfolds them.
  • The signature that was added to a sent email, shown under the message.
  • Attachments: images show as pictures you can open; other files as download links.
  • "Remote images blocked · Show images": pictures that an email loads from the internet are blocked until you choose to show them. This protects you from tracking by the sender.
  • Thumbs up and down under an AI reply: rate the reply. They are shown to people with the 'Manage Drafts' permission; without it there are no thumbs under the reply.
  • "Practice reply": a reply on the untouched example conversation that Squid puts in a new inbox, addressed only to its sample contact. It is saved so you can see how a reply looks and is never sent to anyone.

Reading the messages needs the 'View Messages' permission.

Reply composer

Needs the 'Send Messages' permission.

The box under the thread is where you answer. Its tabs: "Reply" (or "Reply All" on an email with other recipients), "Internal Note" and "Manual refund".

Addresses (email). "From" is the store's sending address. When the store has several, you choose; the composer starts with the address the customer last wrote to, and asks "Send from different address?" before sending from another one. "To" is the customer. "Cc/Bcc · Edit" opens the recipients so you can add or remove addresses.

Writing. For email the toolbar has bold, italic, underline, bullet list, numbered list, quote, link, image, "Insert Macro", undo and redo. Chat and WhatsApp replies are plain text. The store's email signature is shown under your text as a preview and is added when the email is sent; it is set under Settings → Email. The expand button lifts the composer into a large window; Esc collapses it. Drag its top edge to resize it.

Macros. "Insert Macro", or typing a semicolon, opens your saved replies. Picking one inserts its text with the variables filled in. Macros are made under Settings → Macros.

Attachments. The paperclip attaches files: images, video, audio, PDF, Word and Excel documents, plain text, CSV and zip files, up to 10 MB each.

Sending. The send button has three options: "Send" keeps the conversation open, "Send & Close" closes it after sending, and "Send & Resolve" closes it as resolved. The option you pick becomes your default in this browser. Cmd or Ctrl + Enter sends with the default. Adding Shift sends and closes when your default is "Send", and only sends when your default is "Send & Close" or "Send & Resolve". "Send & Close" and "Send & Resolve" need the 'Close Conversations' permission; without it the reply is refused and you are told to choose "Send".

What sending does: an email reply is queued for delivery from the chosen address, a chat reply appears in the visitor's widget, and a WhatsApp reply is queued for delivery to their phone. A reply appearing in the thread does not confirm that an email or WhatsApp message reached the customer. The reply is added to the thread for the whole team. Replying to a snoozed conversation brings it back to open; replying to a closed conversation leaves it closed. A sent reply cannot be recalled or edited.

Unsent text is kept. What you type is saved in your browser per conversation, so switching conversations or reloading does not lose it. When you come back the composer says "Unsent reply restored" with "Discard it". Saved text is kept for seven days, for up to 50 conversations, on this browser only, and is removed when you sign out. This needs working browser storage; text beyond 200,000 characters across the reply and note is not saved.

When you cannot reply. The composer says why: your role is read-only (replying and adding notes both need the 'Send Messages' permission); the store has no verified sending address, with "Set up sending" for people who can change the email settings; the AI holds the conversation, with "Take Over from AI"; or the AI's escalation still needs a rating, with "Rate AI to start replying".

"Manual refund" logs a refund your team makes for this ticket's customer, so it is counted in the Refund Hub and the refund sheet. "Fill in the form" opens the refund form. Submitting it refunds the customer in Shopify; offers that need QA are paid when a reviewer approves them. The tab is shown to people with the 'Process Refunds' permission; submitting the form also needs 'Edit Subject and Dismiss Bounces'. The Refund Hub explains the form.

Internal notes

Needs the 'Send Messages' permission.

The "Internal Note" tab writes a note instead of a reply. A note is saved in the thread with "Add Note", is visible to your team only, and is never sent to the customer. It does not count as a reply for SLA targets, does not change the status of the conversation and does not change who it is assigned to.

Type @ and a name to mention a teammate. They get a notification. Mentioning yourself sends nothing.

Notes work when replies do not: while the AI holds the conversation, before an escalation has been rated, on a store that cannot send email, and outside the WhatsApp reply window. Adding a note needs the 'Send Messages' permission.

AI drafts

Needs the 'View Drafts' permission.

When a store's AI works in drafts, the AI writes the reply and a person decides. A conversation with a waiting draft is in the "AI Drafts" view, and the draft opens in the composer under "AI draft — review before sending".

  • Read it first. The approve button unlocks once the draft has been on your screen. On a long draft, "Show me the draft" scrolls to it; on a very small window "I've read the draft" confirms it.
  • Edit the text freely before approving.
  • "Approve & Send" sends a chat reply to the widget or queues an email or WhatsApp reply for delivery. It leaves the ticket's status unchanged, including when it is already closed or snoozed. "Approve & Close" and "Approve & Resolve" also close it, and need the 'Close Conversations' permission. The default is "Approve & Close", and your pick becomes your default in this browser. When the draft asks the customer a question, Squid switches to "Approve & Send" and shows "Stays open", because the draft expects an answer; picking another option from the menu overrules that.
  • "Reject" discards the draft. Nothing is sent to the customer, and the conversation stays as it was, still waiting for an answer. You are asked why; the reason is optional unless Settings → AI Feedback makes it required or switches the question off.
  • Thumbs up or down rates the draft.
  • Rewriting most of a draft before sending can make Squid ask what was wrong with it. This is off until it is switched on under Settings → AI Feedback, which also sets how much of the draft must change (70% unless changed) and whether the note is required. The reply is sent either way unless the note is required.

Approval cannot be undone, and rejection has no undo button. A sent draft is shown in the thread as "AI draft · approved" with the name of the person who approved it.

Draft mode notice. In drafts the AI does not carry out refunds, discounts, cancellations or photo requests itself. When a draft promises one of those, the information button beside "AI draft — review before sending" reminds you to do it yourself after sending. "AI tool side-effects on this conversation" lists five kinds of record linked to the conversation: refund requests, automatic refunds, discount codes, Basecamp tickets and supplier tickets. It is not a full list of what the AI did: address or customer detail updates, return registrations, photo requests and tags do not appear there, and a refund request is listed whoever made it.

If a draft arrives while you are writing your own reply, your text is kept and you choose between "Use my reply" and "Discard mine".

Seeing drafts needs the 'View Drafts' permission; approving, editing and rejecting need 'Manage Drafts'. Without them the composer says a draft is waiting and that you cannot review it.

Details panel

"Details" at the top right of an open conversation opens and closes the panel on the right. The panel reads the customer's record, which needs the 'View Customers' permission; without it the panel stays empty. It has four tabs: "Details", "QA review", "AI activity" and "AI QA". Conversations normally open on "Details". A link from QA can open the corresponding review tab instead, and "N sources" opens "AI activity".

The "Details" tab shows, from top to bottom:

  • "SLA deadlines": the SLA policy on this ticket, or the reason why none applies; whether its clock runs in business hours or around the clock; and the "First response" and "Resolution" deadlines, each marked "Pending", "Completed", "Completed after deadline", "Deadline missed" or "Paused". A target without a deadline says "No deadline set". The times are in your own display timezone.
  • "Contact": the customer's email address and phone number, when known, and "Customer since", the day this customer was first stored in Squid. It is not the date of their first order.
  • "Conversation Tags": the tags on this conversation. They are added and removed here; see "Priority, tags and more actions".
  • "Store": the name of the store the conversation belongs to. "Stores:" under it lists every store this customer has written to, out of the stores you can open.
  • "Shopify": see "Orders and order actions".
  • "Customer Tags": the tags kept on the customer's record in Squid, when there are any. The customer's tags in Shopify are shown under the figures of the "Shopify" block instead.
  • "Custom Fields": see the next section.
  • "Conversation History": the customer's other conversations, the 20 most recent at most, with ticket number, subject, status, channel and the time of the last message. Five are shown at first, with "Show N more...". Clicking one opens it. For a customer without an email address, the history holds the conversations of every customer record with the same phone number.
  • "Merge History": on a ticket that was merged; see "Merge tickets".
  • "Notes": the note kept on the customer's record, when there is one.

The store card. In an email conversation the store's name, in the address lines of a message and in front of the From address of the reply box, opens a card about the store: "Added to Squid" with the date, "Website", "Return portal" and "Store emails on Squid". Each address shows "AI replies", "AI drafts" or "Human only", and "Receive-only" on an address that is set up to receive mail only; the address of this ticket is marked "This ticket". "Open store settings" goes to Settings → Stores. "Manage emails" goes to Settings → Email and is hidden from people without the 'View Email Config' permission. The card's details need the 'View Brands' permission.

The "QA review" tab holds the one QA review of the ticket: an "Overall score" from 1 to 5, a score from 1 to 5 on each point of your QA rubric and a comment, with the reviewer's name and who handled the ticket.

  • The tab can be opened once the ticket is closed, by people with the 'View QA Reviews' permission. Before that it says "Available once the ticket is closed".
  • "Add QA review" needs 'Submit QA Reviews'. A ticket can be reviewed when it is closed and has at least one reply from a person or the AI; a saved FAQ answer of the chat widget does not count. Only the overall score is required. The comment is required when the overall score is at or below the low-score limit of your QA settings, which is 3 unless changed and never lower than 3.
  • You cannot review a ticket you handled yourself: "You cannot QA a ticket you handled yourself". The handler is the teammate who sent the latest reply written by a person or, when no person replied, the assignee. A ticket answered by the AI alone has no handler.
  • "Edit" changes an existing review. It is offered to the person who wrote it and to people with 'Manage QA Reviews', and it also needs 'Submit QA Reviews'.
  • "Delete" removes the review after you confirm with "Delete review": its scores and notes are removed from the ticket and cannot be brought back. Deleting needs 'Manage QA Reviews'; the person who wrote the review cannot delete it without that permission.
  • When Squid cannot check whether you may review the ticket, scoring is off and "Retry eligibility check" tries again.

The "AI QA" tab is about the AI only. It can be opened on tickets with at least one AI message, by people with 'View QA Reviews'. A saved FAQ answer of the chat widget does not count as an AI message.

  • "Rate the quality of AI response": pick "Good" or "Bad". For "Bad", "Which errors did AI make?" is required: pick one or more from your organization's shared list of errors, or create a new one, which adds it to that list for everyone. "What did AI do wrong?" is an optional comment; for "Good" the comment box is called "Notes". "Submit AI QA" saves it; saving and creating an error need 'Submit QA Reviews'. A ticket has one AI QA rating, whoever gave it; "Edit" and "Update AI QA" overwrite it.
  • "Ticket score" lists every rating the ticket got, with who gave it: "Manual takeover" (given when someone took the ticket over from the AI), "Escalation" (given in "Rate AI before replying") and "AI QA". The "Final score" is the worst of them. "Don't know" counts only when there is no other rating.
  • The ratings and comments are collected under AI → Feedback.

The green dot on the "AI QA" tab shows that the ticket already has an AI QA rating.

Custom fields and required fields

"Custom Fields" in the details panel shows your organization's active ticket fields for this conversation. Fields are created under Settings → Custom Fields.

  • A text or number field is saved when you press Enter, click outside the field or click the "Save" tick that appears beside it.
  • A date, a dropdown, a yes/no box and a multiple-choice list are saved as soon as you change them.
  • A yes/no field shows "Not answered" until someone answers it. On a required yes/no field, "Save No" stores the answer "No".
  • A value you typed but did not save is saved when you switch to another conversation or leave the Inbox.

Seeing the fields needs the 'View Custom Fields' permission; without it the block is not shown. Changing them needs 'Edit Custom Fields'; without it the fields are shown greyed out.

A field marked as required must have a saved answer before a person closes or resolves the ticket, also with "Send & Close", "Send & Resolve" and when approving an AI draft with a closing option. Squid refuses the close and names the fields that are missing. "No" and 0 count as answers; an empty multiple-choice list does not. The AI closing a conversation is not held back by required fields.

Required fields carry a red asterisk in "Custom Fields". A person who can close conversations but lacks 'View Custom Fields' or 'Edit Custom Fields' gets a separate block, "Required ticket fields", with "Complete these fields before closing this ticket." and only the required fields, so they can still close the ticket.

Orders and order actions

Needs the 'View Integrations' permission.

"Shopify" in the details panel shows what Shopify knows about the customer in the store of this conversation. On a mailbox shared by several stores it reads the store the AI matched the ticket to. Squid looks the customer up in Shopify by email address. Seeing it needs the 'View Integrations' permission; without it the block says "Order history isn't available with your role."

  • Without a connected store the block is called "Shopify Data" and says "Connect Shopify to view order history". People with 'Edit Integrations' get "Connect Store"; others are told to ask an administrator.
  • When the connection is broken it shows the error. "Verify Permissions" is offered when the Squid app lacks a permission in Shopify and "Reconnect Store" when Shopify no longer accepts the connection, both to people with 'Edit Integrations'. "Try Again" is offered for most other errors.
  • "No Shopify customer found" means the store has no customer with this email address. A customer without an email address, such as someone who only wrote on WhatsApp, always shows this.
  • Otherwise it shows "Lifetime Value" (the total the customer spent according to Shopify), the number of "Orders", the customer's tags in Shopify, the "Last Order" and the "Order History".

Only the "Last Order" can be opened: click it to see its first three items, "Tracking" with the carrier and a link to follow the parcel, and the "Shipping Address". The label beside the order number is the fulfilment state: "Fulfilled", "Partial" or "Unfulfilled". The orders under "Order History" are one line each, with the fulfilment state and the total. An order number is a link to that order in Shopify. The payment state of an order is not shown.

"Active dispute" appears under the last order when a payment dispute is open on the order of this ticket, or on the last order when the AI has not recorded an order for the ticket. It shows the payment provider, the order, the state, the reason, the amount and "Evidence due". "Show N other active disputes on this customer" lists open disputes on the customer's other orders, of which at most 20 are checked.

Three buttons under the opened last order act in the connected Shopify store. All three need the 'Edit Integrations' permission and access to that store, and none of them can be undone from Squid. Squid does not record them on the ticket or in the Refund Hub, and the panel keeps showing the order as it was until you open the conversation again.

  • "Create Discount" creates a discount code in Shopify: "Percentage off" (10% unless changed, at most 100%), "Fixed amount off" or "Free shipping", with an optional "Minimum order value", "Expires in" 1 to 365 days (30 unless changed), a "Usage limit" (1 unless changed) and "Once per customer" (on unless changed). Leaving "Expires in" or "Usage limit" empty still gives 30 days and one use. Squid makes up the code. It is not tied to this customer or this order: anyone who has the code can use it in that store, on any product, within the usage limit. Squid tries to copy the code to your clipboard so you can paste it into your reply. If copying fails, it displays the code for you to copy manually. The customer is not sent anything until you send that reply.
  • "Cancel Order" is offered while the order is not cancelled and nothing of it has been fulfilled. You choose a "Reason", "Restock inventory" (on by default) and "Email customer" (on by default; Shopify then emails the customer). Squid asks Shopify to cancel without a refund; use "Refund" to pay money back. Shopify cancels in the background. When it has not confirmed within a few seconds, Squid shows an error although the cancellation may still go through, so check the order in Shopify before trying again.
  • "Refund" is offered on an order that is paid, partly paid or partly refunded and not cancelled. "Custom amount" pays back the amount you type, through Shopify, to the payment the customer made, in the currency the customer paid in. It is refused when the order has no successful payment to refund against. "Refund items" starts with every item ticked; untick items or lower the quantities. Squid then sends Shopify those items and quantities, without an amount and without restocking them. "Estimated refund" is the item price times the quantity, before discounts, tax and shipping; check the refund in Shopify afterwards. Options: "Reason (optional)", which is stored as a note on the refund in Shopify, "Notify customer" (on by default; Shopify emails the customer) and "Refund shipping" (off by default), which also marks the shipping cost as refunded. "Create Refund" sends it to Shopify at once, without a further confirmation.

AI activity and reply sources

The "AI activity" tab in the details panel shows what the AI did on this ticket. It can be opened once the AI has worked on the ticket; before that the tab says "AI has not worked on this ticket".

  • A summary: the latest recorded "Category" with the AI's confidence; "Sentiment", "Urgency" and "Language" from the first recorded reply in the timeline; the "Config Set" that ran and the number of "AI Turns".
  • The escalation reason, when the AI handed the ticket to a person, with a "Suggested fix" where Squid has one.
  • "Retry AI" is shown beside the error when an AI run failed. It asks the AI to try again. When the failed run had handed the ticket to a person, retrying also gives the ticket back to the AI: the assignee and the team are removed and a note in the conversation names who did it. Needs the 'Assign Conversations' permission.
  • "Tool side-effects": what the AI started outside the conversation: "Refund queued", "Auto-refund", "Discount issued", "Supplier ticket" or "Cancellation ticket", each with "Open" where there is something to open. Supplier and cancellation tickets are listed only for people with the 'View Supplier Tickets' permission.
  • "Timeline": each step in order, grouped by day. Tools the AI called, how long they took and whether they succeeded; specialists it called; each "AI Response" (marked as a draft while it waits for approval), with the model's reasoning where it was recorded; "Draft approved", with "edited" when the text was changed, and "Draft rejected" with the reason; an acknowledgment sent to the customer or kept as a draft, or "Acknowledgment skipped" with the reason; "AI Response Blocked" with what happened next (rewritten, escalated to a person or sent unverified); and "Escalated to human" with the trigger, for example "Keyword match", "Categorizer flagged", "Max turns exceeded", "Tool requested", "Empty response", "Service timeout" or "AI hallucination".
  • "Replay in Playground", under an "AI Response": opens that run in the AI Playground to test it. It is shown to people with the 'View AI Config' permission.

"Sources used", under each "AI Response" in the timeline, lists what that reply was based on. The link "N sources" under an AI reply in the thread opens it here. The sources are grouped as "Knowledge", "Product facts", "Tools", "Specialist" and "Release". The block is marked "Team only": sources are never shown to the customer. A reply from before sources were recorded says so.

  • For each source you can give a thumbs up or a thumbs down. Everyone who can open the conversation can rate. The number beside the thumbs down counts your organization's thumbs down on that source in the last 14 days.
  • A thumbs down asks for a reason: "Outdated", "Wrong for this case", "Not relevant" or "Missing info", or a comment. For most people the thumbs down is recorded at once and the reason is optional. For people with the 'Edit AI Config' permission it is recorded only once a reason or a comment is given.
  • After that, on a knowledge source, people with 'Edit AI Config' are offered "Fix it now": Squid Agent drafts a change to that text as a proposal, which a person approves before customers see it. "Just record it" only keeps the feedback.
  • "Missing a source?" records which knowledge the AI should have used. People with 'Edit AI Config' can also use "Create knowledge", which opens a new draft in the AI Builder.

Live chat

Chat conversations come from the chat widget on your store.

  • Presence: the dot on the row in the list and the label in the header show whether the visitor's chat widget is still connected: "Online", or "Last seen" with how long ago they left. Only live chat has this; a WhatsApp or email conversation never shows it.
  • Replies are plain text and appear in the visitor's widget at once. Macros open with a semicolon or two slashes.
  • No snooze: a live chat cannot be snoozed.
  • Follow-up tickets: Squid can continue a chat as an email ticket, when it closes a chat because the visitor left and the store is set up for that, or when the AI turns a chat into a ticket. A chat gets at most one follow-up ticket. The chat then shows "Follow-up ticket #…" and the email ticket shows "Follow-up of #…". Each links to the other.
  • "Saved FAQ" answers in the thread are the fixed answers of the widget's FAQ buttons.

Whether chats close when the visitor leaves, the follow-up ticket and the widget's texts are set under Settings → Chat Widget.

WhatsApp

WhatsApp conversations show a delivery state under each reply: "Sending", "Sent", "Delivered", "Read", "Delivery uncertain", "Not delivered", "Expired before it could be sent" or "Cancelled", with WhatsApp's reason where it gave one. The row in the list gets a red triangle when a WhatsApp delivery failed.

"Retry" is offered on a reply that is "Delivery uncertain", "Not delivered" or "Expired before it could be sent". It queues the undelivered or uncertain parts again and needs the 'Send Messages' permission. Squid never resends a WhatsApp message by itself. A reply marked "Delivery uncertain" may have reached the customer after all, so a retry can deliver it twice.

WhatsApp only allows free-form replies for 24 hours after the customer's last message. Squid treats the final minute as closed to leave time for delivery:

  • Inside the window the composer shows "Free-form replies allowed until" and the time.
  • After it, the composer says the window closed and free-form replies cannot be sent. A reply from a person is refused, never held; the same goes for approving an AI draft. "Send re-engagement template" sends the store's approved template that invites the customer to reply; the window reopens when they do. It needs 'Send Messages'. A ticket can have one such template at a time: a second one is refused unless the first failed, expired or was cancelled.
  • "Waiting for the customer to reopen the 24-hour window" is shown under a reply of the AI, or a system acknowledgment, that could not go out because the window was closed. That happens when the AI answers after the 24 hours have passed, or when its reply was still waiting to be sent at the moment the window closed. Squid holds the message. For an AI reply it also sends the re-engagement template once and adds a note to the conversation. When the store has no usable template, the note says a person has to reach the customer another way and the conversation is flagged as a failed delivery.
  • When the customer writes again, the held messages are sent in their original order. A message that is still held after 7 days is given up and shows "Expired before it could be sent", with a note in the conversation.
  • Before the customer has written at all, only the re-engagement template can be sent.
  • A store without a WhatsApp connection cannot send; the composer links to Settings → WhatsApp.

Internal notes work in all of these cases. WhatsApp conversations can be snoozed like email.

Failed emails

When an email to a customer cannot be delivered, the conversation gets a red envelope in the list, appears in the "Failed Emails" view and shows a red banner: "Email bounced", "Email dropped", "Email blocked" or "Email delivery failed", with the address, "Hard bounce" or "Soft bounce" where the mail server said which, the mail server's answer and its status code. A hard bounce records a permanent delivery failure; it can reflect a policy or content refusal as well as a bad address. A soft bounce records a temporary failure.

The conversation is flagged for a hard bounce, for an email that was dropped or blocked, and when Squid could not send the email at all. A soft bounce alone does not flag it. The teammate who sent the email also gets a notification, "Email delivery failed", or "Email delivery delayed" when the mail server is still trying.

Delivery failed for the recipient named in the warning. Other recipients of the same email may still have received it. Check the address and failure reason, then answer again or reach that recipient another way. The warning goes away by itself when a later email on the conversation is sent and nothing has failed since. The cross on the banner dismisses the warning and takes the conversation out of "Failed Emails"; it needs the 'Edit Subject and Dismiss Bounces' permission. Dismissing does not resend anything.

Who is viewing and reply collisions

The Inbox updates live: new conversations and messages, status and assignee changes, AI drafts and delivery failures appear without reloading.

  • Who is viewing: names at the top of a conversation, with "viewing", and the eye chip in the list show teammates who have it open. A banner above the thread says who is also viewing the ticket; it can be dismissed.
  • Who is replying: when a teammate has started a reply, the chip turns red and says "replying", and the banner says they are already writing a reply to this ticket. "X is typing..." shows under the thread.
  • What the customer sees of your typing: a visitor in the chat widget does not see that a teammate is typing. The typing dots in the widget are only shown for replies of the AI. On WhatsApp the customer does see it: while you type, their last message is marked as read and a typing indicator is shown, at most once every 20 seconds. This only happens when "Simulate typing" is on for the store under AI → Settings.
  • Before you send while someone else has an unsent reply open, Squid asks "Someone else is replying to this ticket" with "Send Anyway".
  • If a reply was sent while you were writing, your send is stopped: "Someone already replied", or "You already replied" when it was you, for example in another tab. That reply is shown above; "Send Anyway" sends yours too, "Keep My Draft" keeps your text without sending. Internal notes are never stopped this way.

These warnings are there so a customer does not get two answers. They do not lock the conversation for anyone.

Keyboard shortcuts

With a conversation list in front of you and the cursor not in a text field:

  • J and K: next and previous conversation in the list.
  • R: reply. N: internal note.
  • E: close the open conversation. It closes, it does not resolve.
  • S: open the snooze menu. It does nothing on a live chat or on a conversation that is not open.
  • A: open the assign menu. It does nothing while the conversation waits for "Take over" or for the AI rating of an escalation.
  • Esc: leave the reply box, so the keys above work again.
  • Cmd or Ctrl + Enter in the reply box: send.

A shortcut does the same as its button and needs the same permission. The keys are off while a dialog is open and while the inbox is locked. The question mark key shows the full list of shortcuts for the app.

Banners and the locked inbox

Banners at the top of the Inbox point at things that stop customers from being answered:

  • Stores without AI answers: the AI is switched off, or setup is not finished, for the named stores. Messages for those stores wait for a person. With a conversation open, the banner speaks for that conversation's store only. For one store it offers "Turn the AI on" to people with the 'View AI Config' permission, or "Finish setting up" with the store's name to people with 'Edit Brands'; for several stores it offers "Review these stores". Others are told to ask an administrator.
  • Stores that cannot send: a store cannot send email while customers are waiting in open email conversations, with the number waiting. "Set up sending" is offered to people with the 'Edit Email Config' permission.
  • An incident's tickets: when you arrive from an incident in Squid Agent, the list shows only that incident's tickets, in every status, with "Back to the incident" and "Show all tickets".

The first two banners can be dismissed. They stay away until you close the browser tab, and come back sooner when the stores they name change.

"Inbox Locked" covers the page when the subscription is not active: past due, unpaid, cancelled or never completed. Messages from customers are still received, but the team cannot work in the inbox and replies are refused. "Go to Billing" is shown to people with the 'Billing' permission; others are told to ask an administrator.

The lock screen says messages are kept for at most 30 days. For an unpaid subscription, incoming messages are accepted during a 30-day grace period. After that, intake is refused. Cleanup can run in the background or when another message arrives; it deletes conversations created or last contacted by the customer since the unpaid period began. Older conversations are outside that unpaid-period cleanup. A cancelled subscription also has a separate cleanup that deletes all conversations and customer records after the cancellation grace period, which is 30 days unless changed. These deletions cannot be undone from Squid.