Introducing the AI Inbox

"Thanks for reaching out!" is the most expensive sentence in customer support — not for the words, but for everything around them: opening the thread, recalling what your help center says, matching the tone your team agreed on, and deciding whether this message needs a human at all. Multiply that by every conversation that lands in the inbox, and you can see exactly where the response time goes.

The AI inbox does that first pass for you. From today, every new customer message — email, portal message, or API submission — gets a reply drafted automatically, grounded in your project's help center articles, canned responses, and the instructions you write for your agents. It isn't a chatbot answering in your place. It's a first draft, waiting in the thread before you've finished reading the ticket.

Grounded in your help center, not in thin air

A generic model can't answer your questions — it doesn't know your product. The AI inbox doesn't guess. For each inbound message it retrieves the most relevant help center articles, pulls in your canned responses, and shapes the reply with the voice and policy notes you write in Project Settings → AI. The same grounded pipeline powers the manual "Draft AI Reply" button on any thread, so a draft you ask for and one the inbox produced on its own come from the same process.

Approval-first: nothing sends without you

Every draft appears in the thread as a reviewable card. Approve it, edit it, send it, dismiss it, or regenerate it until it's right. No message goes out without a member's approval — unless you turn that off on purpose.

Auto-send: the opt-in that answers for you

For teams that want to close the loop instantly, auto-send can ship a draft the moment it's written. It's off by default and enabled per project, and it's guarded by a risk classifier that reads the conversation and the draft. Threads about refunds, billing, legal, complaints, or escalations — or carrying links or attachments — are never auto-sent. Auto-send fires at most once per thread per day, and the team is notified the moment a reply leaves, so nothing slips past unnoticed.

Where it works

Email is the obvious channel — every reply comes back through your reply+<token>@ address. Portal messages and API submissions run through the same pipeline. Live chat is deliberately left out: a chat reply belongs in the chat window, not in someone's inbox.

Metered on its own allowance

A grounded draft is a whole pipeline — retrieval, grounding, model call, risk classification — not a single API call, so it gets an allowance of its own. Pro includes 500 drafts a month; Team includes 2,000. Bring-your-own-key drafts are metered the same way, because the pipeline runs identically whichever key pays for the tokens. Overage is $0.05 per draft, and drafts that already exist are never taken away when you reach the cap.

Get started

1. Open Workspace Settings → AI and enable AI.
2. Write your agent instructions — your voice and reply policy, in plain language.
3. In Project Settings → AI, turn on Auto-reply.
4. Optional: once the drafts look right, enable Auto-send per project.

The AI inbox is included with Pro and Team plans. Read the AI features docs for the full walkthrough, and get started at supdesk.app.