Working the inbox

Everything on this page is available to both roles — working the inbox is the whole job of an agent. Configuring the workspace around it belongs to owners; see Team and roles.

Signing in drops you at the root of the app, which is the conversation list for every workspace you belong to. It updates itself: a new visitor message repaints the list and the counts without a reload, so you can leave it open on a second monitor.

Above the list are the last 30 days at a glance — conversations, the share the AI handled, how many you resolved, and average CSAT — plus a 90-day chart of conversation volume.

Pick a single workspace from the sidebar to narrow everything on the page, or stay on All workspaces. The status tabs above the list carry counts and filter it to one state:

Open

Live. Either the AI is answering or it's waiting on you.

Pending human

Waiting on a person — the visitor asked for one, the AI couldn't help, or a reply failed. These are the ones to look at first. Your first reply moves the thread back to Open, so this tab is always the unanswered queue.

Resolved

Closed by an operator. Reopen it from the thread — or just reply, which reopens it for you.

The list shows the newest 100 conversations by most recent message, with a preview of the last thing said.

Opening a conversation gives you the full history and a composer. Everything is labeled by who said it — the visitor, you, the assistant, or the system (handoffs and errors post as system notes so the record stays honest).

Visitor attachments and the selected routing group appear in the same thread, so operators do not need a second inbox to understand what was sent or where it belongs.

The header carries the visitor's name, a green ✓ if they're a verified logged-in user of your site, the workspace they wrote from, and the conversation's status. A context panel alongside the thread shows the visitor's email, when the conversation started and was last active, the routing group, who it's assigned to, and the CSAT rating once they've left one — on smaller screens it sits behind the Details button.

Replies send on Enter (Shift+Enter for a new line). A reply that fails to send stays in the thread marked Not delivered with its text intact — retry it as-is or pull it back into the composer to edit.

A sent reply has a Save as answer action. It turns that reply into a Knowledge source titled with the visitor's question, with no names attached, and closes any matching open gap under Knowledge → Gaps. The next visitor who asks gets your answer from the assistant.

When a conversation is handed to a person, Doorwick writes a handoff brief at that point in the thread: what the visitor wants, what the assistant already tried, why it handed off, the page the visitor was on, the sources it consulted, and whether the visitor is verified. The facts come from the conversation record; the two summary lines come from your AI provider and are simply absent when none is connected. Visitors never see the brief.

Draft reply in the composer asks the assistant for a first draft grounded in the same knowledge it uses with visitors. Nothing is sent until you press send, the draft cannot call tools or take actions, and the sources it drew from are listed under the composer so you can check them.

The Assign to teammate picker in the thread header records who owns the conversation, and the context panel shows the current owner. Assignment is bookkeeping for your team — it doesn't change who the visitor talks to, and clearing it back to Unassigned is always available.

The pill in the thread header shows who owns the conversation and toggles it. When it reads AI handling, each visitor message triggers an AI reply. Switch it to Human handling and the AI goes quiet immediately — useful when a thread is going sideways and you want to take it yourself. Handing a Pending human thread back to the AI also returns it to Open, so the AI actually resumes answering.

The AI also hands off on its own. request_human is always there; enabled Tools are additional. It's told to call request_human when the visitor asks for a person, sounds frustrated, or when the knowledge base doesn't cover the question. A handoff flips the conversation to Pending human, posts a system note with the reason, and emails everyone in the workspace.

For answers you type more than twice. Create them under Canned replies — a title you'll recognize in a dropdown ("Refund policy") and the body you want to send. Titles are unique per workspace.

In any thread, pick one from Insert canned reply… above the composer. It drops the text into the box rather than sending it, so you can adjust the details before it goes out.

Replies can carry files. The paperclip in the composer stages up to 5 files per message, 5 MB each — images (JPEG, PNG, GIF, WebP), PDFs, plain text and CSV, and Word or Excel documents. Staged files sit above the composer where you can review or remove them before anything goes out, and a reply can be files alone, with no text. If a file is refused — too large, or a type that isn't accepted — the composer says exactly why, and the rest of your reply is unaffected.

Attachments travel only with replies. Internal notes are text-only, so remove any staged files before switching the composer over to a note.

Resolve closes the conversation. The visitor is then asked to rate it from 1 to 5 with an optional comment; that score is what feeds average CSAT. Resolved threads are read-only for the visitor — they can start a fresh conversation from the widget, or you can bring the old one back.

To pick a closed thread back up, hit Reopen in the header — or simply send a reply, which reopens it with its whole history intact and lets the visitor answer again.