AI Search

Ask a question about your workspace in ordinary English and get an answer built from your real work items — with the items it used listed underneath, so you can check its working.

Plan: Pro and above

Where it lives now

AI Search is part of the AI Agent. It used to be a separate sidebar entry; it isn't any more. The agent recognises a "find out about…" question and runs exactly this grounded search behind the scenes, so there's one place to ask things rather than two.

The dedicated page still exists at /<workspace>/ai-search for bookmarks and deep links. If you liked the stripped-back, one-question-one-answer format, keep using it. For anything else — follow-up questions, or asking the AI to actually do something — use the AI Agent on the app rail.

What it does

  • Answers natural-language questions about the work in your workspace.
  • Answers only from your data — it retrieves the most relevant work items and is instructed to use nothing else.
  • Cites its sources. Every item it used is listed below the answer with its reference, project, priority and whether it's overdue.
  • Says "I couldn't find anything" when it can't, instead of making something up.

Using it

  1. Go to /<workspace>/ai-search.
  2. Type a question. Enter asks; Shift+Enter starts a new line.
  3. Read the answer, then scan the Sources list to see which work items it was based on.

If you don't know where to start, the page offers a few suggested questions:

  • What's overdue across all my projects?
  • Summarize the highest-priority open work items.
  • Which projects have the most unfinished work?
  • What was created this week and what got completed?

📷 Screenshot: the AI Search page with a question asked, a markdown answer, and the Sources list of work items below it.

How it finds things

It runs a full-text relevance search over work-item titles and descriptions across the workspace's projects, takes the best matches, and hands them to the AI as the only material it's allowed to use. That's why the answers cite real references and why it can't invent an item that doesn't exist.

It's a good fit for open-ended questions — "what do we know about the payments bug?", "anything about onboarding?" — where you don't know the exact wording to search for.

Limits & good to know

  • It reads; it doesn't act. There is no way to create or change anything from this page. Use the AI Agent for that.
  • It searches work items only — not pages, comments, attachments or project descriptions.
  • It has no memory. Each question is answered on its own; there's no follow-up thread. The AI Agent does have conversation history.
  • It answers from a bounded set of the most relevant items, so a question that would need to weigh hundreds of items ("summarise everything") will be answered from a sample. Narrow the question for precision.
  • The Sources list is trimmed to the most relevant items — the answer may draw on a few more than are displayed.
  • It needs an AI provider configured for the instance (admin console → Artificial Intelligence).
  • Verify before acting. The page says so itself: AI-generated from your workspace data.