IntelliDesk's AI helps at three moments: when a ticket arrives (triage), when you read it (summary and warnings), and when you reply (drafts and suggested templates). Everything it suggests is shown to you first, and nothing is sent to a customer until you press Send.
When a ticket arrives: AI triage
Switch on AI triage & assist for a desk under Settings → Help Desks → your desk. New tickets on that desk are then checked within a second or two, and the ticket shows what the AI decided:
- Priority: set automatically when the AI is confident, marked Set by AI · 99% confident under the priority. It never overrides a priority a person (or an API caller) chose.
- Type: Question, Incident, Problem or Task, when confident.
- Frustrated customer: a flag in the side panel, so an angry customer gets a careful reply.
- Automated mail and spam: out-of-office replies and junk are recognised, so they don't sit in the queue as real work.
- Possible duplicate: a banner links to an earlier ticket that looks like the same issue, with Review & merge.
- Suggested tags: one click adds them.
Triage uses a small decision model and needs no key of your own. See Turning on AI triage for a help desk.
Reading a ticket: Summarize with AI
On a long thread, click Summarize with AI under the subject. You get a few sentences: what the customer asked, what has happened, and where it stands. Re-summarize refreshes it after new messages.
Replying
Suggested replies
When a desk has triage switched on, the reply box offers Suggested replies: the team templates (canned responses) that best fit the customer's latest message. Click one to insert it with its placeholders filled in. See Canned responses: reusable templates with macros.
Draft with AI
Draft with AI in the reply toolbar writes a first draft from the conversation (public messages only; internal notes are never used), your published help articles and, if you've connected a knowledge search, your documents. The help articles are found with the same typo-tolerant, meaning-based search the live chat AI uses, so a customer's misspellings or different wording don't stop it finding the right article. The draft:
- uses only what's in the conversation and your knowledge base, and is told not to invent facts,
- is signed with your name,
- appears in the reply box marked AI draft, with how many help articles (KB sources) it used.
If you've already written something, you're asked before it's replaced. The automatic "Hi ," greeting doesn't count.
Refine with AI
The AI button in the toolbar rewrites what you've written: Polish (clarity and flow), More professional, More friendly, Shorten, Expand, or Fix grammar (spelling and punctuation only).
When AI drafts shine, and when they don't
Good for: long threads where the answer is buried, questions you've answered before, and messages that are hard to read.
Be careful with: judgement calls (refunds, escalations, anything with policy nuance), brand-new problems with nothing in the knowledge base, and emotional complaints, which need a human voice.
Before you send
- Facts: check every claim about prices, policies and dates. If your knowledge base is out of date, the draft will be too.
- Tone: a two-line question deserves a short answer.
- Signature: it should be your name. Anything else is a bug; please report it.
- Does it answer the question? "We'll look into it" is rarely the right reply.
On desks with AI triage & assist on, Jev also checks your reply when you press Send. See Reply checks: what the "Before you send" warnings mean.
Providers and privacy
Admins set the AI provider in Settings → AI: connect your own Anthropic or OpenRouter key and choose the model, and Test it. Without your own key, IntelliDesk's platform account is used (GPT-5.6 Luna through OpenRouter by default). Ticket text is sent to the provider only when you use a feature (or, for triage, when a ticket arrives on a desk that has it switched on). See What data does the AI see? AI and your privacy.