Natural-Language Search

Filter your team's Issues board by typing a plain-English sentence instead of clicking through dropdown menus. Type what you're looking for, press Apply, and the board narrows to the matching issues — across your whole backlog, not just what's currently on screen.

Where to find it

Open any team's Issues page (iKanban → Issues in the sidebar). At the top, above the board, is the ✨ sparkle bar — that's the natural-language search box.

Your first search

1. Click the sparkle bar and type a request, for example:

   urgent issues tagged backend assigned to me

2. Press Enter (or click Apply).

3. The board updates to show only matching issues. A chip appears below the box showing your applied query.

4. To clear it, click the ✕ on the chip — the board returns to showing everything.

That's the whole loop: type → Apply → read the board → ✕ to clear.

Not sure what to type? Click one of the example chips under the box (urgent & unassigned, my in-progress work, tagged backend, stale > 14 days) to prefill it, then Apply.

What you can ask for

Everything below works, and you can combine any of them in one sentence — the parts are AND-ed together.

Examples that combine several at once

• my urgent bugs in backend idle more than 14 days
• unassigned features tagged Bug
• overdue high-priority work assigned to veronica
• done issues about billing

Fine-tuning with Refine and Sort

Some things read better as a control than a sentence. Right next to the sparkle bar:

• Refine — a dropdown for priority, type, assignee, project and tags, plus two extras:
  • Match any tag — when you've picked more than one tag, switch from "must have all of them" to "has any of them."
  • Has sub-issues — show only parent issues that have child issues.
• Sort — reorder the board: Newest, Oldest, Priority, Recently updated, or Due soon.

Natural-language search and the Refine dropdown share the same filter, so you can start with a sentence and then adjust it with the dropdown.

Choosing which AI reads your query

To the right of the search box are two small pickers:

• Provider — Auto (default) uses whatever AI your workspace has set up. You can also pick Claude, ChatGPT, Gemini, or GitHub Models for a single search.
• Model — appears once you pick a specific provider; leave it on Default unless you have a reason to change it.

For everyday use, leave both alone — Auto just works.

Good to know

• A search that shows nothing is still a valid answer. If a filter empties a column, there genuinely are no matching issues — for example, overdue shows nothing when every past-due issue is already Done.
• Free-text search looks at the description too. issues about login finds issues with "login" in the title or the body, so you may occasionally see one whose title looks unrelated but whose description mentions the word. Use a more specific phrase to narrow it.
• Results cover your entire backlog, not just the issues currently loaded on the board.
• Repeat searches are instant — a query you've run before comes back immediately, without re-asking the AI.
• Anything it can't map turns into a gentle hint. If part of your sentence can't become a filter, you'll see an amber "Couldn't map: …" note instead of a wrong guess. Rephrase that part, or use the Refine dropdown.

What it can't do (yet)

A few requests fall outside the filter model and will show up in the "Couldn't map" hint rather than being guessed at:

• Exact calendar months — opened in June. Use a relative window instead (created last week, created this month).
• Sorting by engagement — most commented.
• Relationship filters — blocked by IKA-42.

For any of these, rephrase your request or reach for the Refine dropdown and the Sort control.