Support tickets now arrive pre-sorted. The Hub has a built-in support flow backed by Zammad, with an LLM doing the intake work — categorising the ticket, asking the obvious clarifying questions, searching the docs, and (for clear how-to questions) drafting an auto-reply that closes the loop without a human round-trip.
Think of it as a triage nurse, not a doctor. Every ticket still lands in front of a person. What changes is that the person opens it already knowing the category, the relevant docs links, and — if the user attached screenshots — a summary of what's on screen. The slow part of support shifts from "what is this" to "what's the answer".
What's actually new
- Clarifying questions before the ticket is filed. If a description is ambiguous, the form asks one focused question rather than letting the ticket bounce.
- Auto-reply for how-to tickets. Confident matches against the docs get a public reply with the right links and a clear "if this didn't help, reply and we'll re-open" footer.
- Ticket deduplication and refinement. If a customer files a near- duplicate, the new submission attaches to the existing thread instead of opening a fresh one.
Screenshots flow through too. The capture button on the support widget forwards the image straight into Zammad as an attachment, so first response doesn't have to start with "could you send a screenshot?"
Under the hood, the LLM only ever reads the ticket plus the public docs — it has no access to your deployment, your environment, or your data. The goal is faster routing and accurate first-touch responses, not autonomous resolution.