Hacker News Reader

Comment by denismi | Hacker News Reader

They've seemed very focused on their AI assistant recently, so I'm happy to see a useful new search feature.

Happy to see that custom bangs work (eg a discourse forum I visit), but eventually I'd like to specify how far along the path to "snap".

I'd like my @javadoc to hit `site:docs.oracle.com/en/java/javase/23/docs/api/` instead of the current `site:docs.oracle.com`.

View on HN

We do currently have a field to override the domain used for snaps (the `ad` field in the bangs repository[1]) which doesn't have much validation and is useful for situations like this. It's possible we'll expose this for custom bangs in the future, but for now I can add that as a bang to the main repository, as it seems like it would be generally useful.

[1]: https://github.com/kagisearch/bangs?tab=readme-ov-file#bang-...

https://github.com/kagisearch/bangs/commit/55e8612cd5f19b292...

I would normally agree. But Kagi's AI thing doesn't annoy me in ways it does in other products. End a query in a "?" to get a quick summary with links to the sources. Normal search results below.

Check the latest release notes for all search improvements that went out in this release:

https://kagi.com/changelog#4813

I have stopped to pay Kagi because their obsession to employ AI everywhere.

They are much more level headed about it, and they openly say that “AI is a tool and addition to search. Kagi is fine without it”. For their rationale, see “LLM Features” page in their help, which is linked at the bottom of Kagi Snaps page.

I don’t use any LLM features of Kagi, and it’s not hindered in any way.

Yeah, Kagi is fine without it, but if they would spend all the effort they pit into AI into normal Kagi search, Kagi could possibly be amazing without it. E.g. Kagi could spend their time on improving deep site search (Stackoverflow/Github/Reddit) which is one of the few areas where they still lag behind Google.

I think biggest obstacle in front of normal Kagi search is index size. Kagi is a multi-index search engine for now.

IMO, as they improve their (own/in-house) indexes, I think they’ll be able to build more features on top of standard Kagi search.

Isn't the problem that only google can crawl new reddit pages now?

That would be a very wrong characterization of Kagi does with AI.

We have a an AI integration philosophy that we have been following since day one.

https://help.kagi.com/kagi/why-kagi/ai-philosophy.html

The primary guideline of that philosophy is that any AI feature is opt-in and on-demand rather than being pushed down the user. So to characterize our efforts as you did, can at best be described as doing disservice to reality. Why the need for that?

I am leaving Google because the AI crap.

I mean: Companies seem to be using AI everywhere without a clear justification, just for the sake of using it, but that’s not the issue.

The problem is that Google ruined its reputation by showing completely inaccurate results when it started displaying AI-generated results.

Reply on news.ycombinator.com