Workspace admin

What data does the AI see? AI and your privacy

Updated

IntelliDesk's AI features send ticket text to an AI provider to get an answer. This article explains what's sent, when, and where, so you can decide which features to use.

Nothing is sent unless a feature is on or used

Ticket text only goes to an AI provider when:

  • a ticket arrives on a desk with AI triage & assist switched on,
  • a new email ticket arrives on a desk with Write the acknowledgement with AI switched on,
  • a trigger has an AI check (Jev) condition that applies to the ticket,
  • an agent uses an AI feature, such as Draft with AI, Summarize with AI, the AI refine menu, or a "generate with AI" button,
  • an agent presses Send on a desk with AI triage & assist on (the reply check),
  • a visitor chats with the chat AI, which only runs when you've connected your own AI key, or
  • a manager runs Learn from past tickets, or clicks Draft an article on a Help centre gaps topic (both under Settings → Knowledge Base).

If none of those apply, no ticket text is sent to an AI provider. Your published help articles are handled separately, for the search index (see The help article search index below).

Internal notes stay out

Internal notes are never sent to the AI for triage, suggestions, summaries, drafts or Learn from past tickets. Only the public conversation is used.

What each feature sends

Feature What's sent
AI triage (new tickets) The ticket's subject and the customer's first message. Also your tag names and your saved replies' titles and opening lines, so Jev can suggest them.
Duplicate check The new ticket, plus the number, subject, short summary and status of up to 15 earlier tickets from the last 60 days.
Best match (AI) The ticket, plus the names of the agents on shift and what you wrote under What each agent handles.
AI checks in triggers The ticket and the statements you wrote in your rules.
Suggestions after a customer reply The last few public messages of the conversation.
Reply check The subject, the customer's latest message, and your reply.
Draft with AI / Summarize with AI The public conversation, plus any help articles (and Knowledge Search answer) that match.
Live chat AI The recent chat conversation, the matching published help articles, and the desk's Notes for the AI assistant.
AI acknowledgement (email tickets) The new email, the matching published help articles, and the desk's Notes for the AI assistant.
Learn from past tickets The customers' messages and your team's public replies on the desk's closed tickets in the period you choose, plus your published help articles to compare against. Uses your own AI key.
Help centre gaps Questions the chat AI or AI acknowledgements couldn't answer are grouped by meaning (via the search index below). Draft an article sends those questions and your team's replies on those tickets to the writing model.

The help article search index

To find the right help article even when a question is misspelt or worded differently, IntelliDesk keeps a meaning-based search index of your help articles, used by live chat answers, AI acknowledgements and Draft with AI.

  • To build the index, the text of your published help articles is sent to OpenRouter, which runs OpenAI's text-embedding-3-small model, using IntelliDesk's own account.
  • When the AI searches, the question being searched is sent the same way, to OpenRouter on IntelliDesk's account. The same index is used to group unanswered questions for Help centre gaps and topics for Learn from past tickets.
  • Before searching, the AI rewrites the question (fixing spelling and adding likely wording). That rewrite uses the AI key connected under Settings → AI.

Published articles are already public on your help centre. Drafts aren't included.

Where it goes

  • Jev (triage, suggestions, reply checks, best match, AI checks) is a decision model from TypeSafe. IntelliDesk reaches it either directly or through OpenRouter. If you've connected your own OpenRouter key, that key is used.
  • The writing model (drafts, summaries, refine, chat answers, AI acknowledgements, suggested articles) is the provider and model you chose under Settings → AI (Anthropic or OpenRouter). Without your own key, IntelliDesk's own AI account is used, except for live chat, AI acknowledgements and Learn from past tickets, which only ever use your key.
  • The search index uses OpenRouter (OpenAI text-embedding-3-small) on IntelliDesk's account, as described above.

With your own key, your provider's terms apply to what's sent. Check them if you need to know how your provider handles data.

Your key

Your AI key is stored encrypted, and only its last few characters are ever shown again. Only organisation admins can change or remove it.

Guard rails on customer-facing AI

  • The chat AI answers only from your published help articles (and the desk's Notes for the AI assistant), and every answer is labelled AI assistant, so visitors know they aren't talking to a person.
  • The chat AI is told never to ask for passwords, card numbers or other credentials.
  • The reply check warns agents before they send a password, card, bank or ID number to a customer.
  • Articles the AI suggests (from Learn from past tickets or Help centre gaps) are saved as drafts for a person to check. They're never published automatically.

Using less AI, or none

  • Turn off AI triage & assist per desk (Settings → Help Desks).
  • Turn off Write the acknowledgement with AI per desk (Settings → Help Desks → your desk → Settings tab, Auto-reply card).
  • Remove AI check (Jev) conditions from your triggers.
  • Click Remove under Settings → AI to stop the chat AI.
  • Agents can simply not use Draft, Summarize or refine: those only run when clicked.

For IntelliDesk's general privacy commitments, see our privacy policy at intellidesk.co/privacy.

Related: Setting up AI: your own key, the platform AI and what each feature needs · What is Jev? The AI in IntelliDesk explained