Configure a Smart Assistant

AI Agents2 min read

Choose Smart Assistant for model-generated answers that can use RAPTIX knowledge.

  1. Select an active Generator Model configured in AI authentication.
  2. Leave Document processing on Automatic unless a prepared preset is required for files uploaded in chat.
  3. Set Knowledge scope. An empty scope means all knowledge the assistant and requesting user are otherwise allowed to access.
  4. Set Temperature from 0 to 1.
  5. Review the English and Arabic System Prompt fields. Reset restores the supplied prompt for that language.
  6. Configure visibility, save, and test with representative questions.

Knowledge scope controls where the assistant searches; visibility controls who can find and chat with it. Both must be configured correctly.

Users see a Smart Assistant as Document in the hub and can receive citations, attach a file, and use owner-only per-session settings.

What a Smart Assistant can do (agentic tools)

A Smart Assistant does not only retrieve text and quote it — it works agentically, and this behaviour is always on. For each question it can decide to call built-in tools and act on the results before it answers:

  • Documents — find, list, count, and open documents, read a document's details, and search inside document content. This is why a Smart Assistant can answer meta-questions such as "how many documents do you have?" or "list the contracts you can see," not just quote passages.
  • Workflows — look up workflow and process information available to the assistant.
  • Web — retrieve current information from the web when the answer is not in your knowledge.

Everything the assistant reaches still respects the assistant's Knowledge scope and the requesting user's own permissions — the tools widen what it can do, not what it is allowed to see.

When an answer relied on a tool, the transcript shows a small tool-trace chip under the reply — for example search_document_content · found 3 results — so the reader can see which tool ran and what it returned. A chip with a warning icon means that tool call did not complete. The chips are informational; the answer, with any citations, follows below them.

Still stuck? See Get Help with Support or reach your workspace administrator.