Kagi’s bang autosuggest correctly shows available bangs when typing a base command like !amazon. However, when attempting to refine the search by typing a multi-word query such as !amazon india, the autosuggest fails to maintain context and instead defaults to a standard Amazon.com search for "india" . This behavior is not limited to Amazon—it occurs across all bangs where the intended target is a multi-word or region-specific variant.
For example, typing !wikipedia italian should ideally suggest or resolve to !wikipediait, but instead appears to fall back to a general Wikipedia search for "italian" on the primary domain. The system treats the input as a hybrid query rather than continuing to prioritize relevant bang matches.
Steps to reproduce:
- Go to Kagi Search (kagi.com).
- Click in the search bar and type a known bang prefix (e.g.,
!amazon).
- Observe that related bangs (e.g.,
!amazonin, !amazonde) appear in the autosuggest dropdown.
- Continue typing to form a multi-word query (e.g.,
!amazon india).
- Observe that the autosuggest reverts to a generic search on the default site (e.g., Amazon.com) for the keyword "india", losing the regional bang context.
This issue occurs consistently across all browsers and devices. Screenshots are attached showing the
correct initial autosuggest state and the incorrect fallback behavior.
The expected behavior is that after triggering a bang prefix (e.g., !amazon), typing additional keywords should continue to filter or prioritize relevant bangs rather than switching to a standard search. For instance, typing !amazon india should either:
- Highlight or suggest the
!amazonin bang
- Or allow partial matching so that "india" filters the list of available Amazon-related bangs
This would make the autosuggest system more intelligent and consistent with user intent when using multi-word or region-specific shortcuts.
Currently, the system breaks the flow by treating !x y as a search on site x for term y, even when x y could resolve to a known bang like !x-y or !xregion. The desired functionality would preserve the bang context during refinement, enabling faster and more accurate access to specialized search targets. This behavior should apply universally to all bangs with regional or multi-part variants.
