A Few Months with Kagi

Posted:

I've subscribed to Kagi for a few months now and wanted to collect some of my feelings towards it, particularly addressing whether I think it's worth paying $10/mo for a search engine. To summarize: yes, I do think Kagi is worth the price. Kagi performs as good or better than the competition for the majority of things I search (programming stuff and a healthy dose of Emacs) and offers a few useful features that add to the experience. The fact that it performs so well without abusing my personal data is quite the achievement.

The thing that got me into Kagi in the first place was their post about Kagi Small Web. It shouldn't really be surprising that I am passionate about RSS and self-hosted blogs, you're reading one. What Kagi is doing to help feature small blogs in their search results is commendable. It clearly demonstrates the niche that Kagi is trying to carve out as a search engine in a space owned by giants: the search experience provided by Google is plagued with incentives that make the web worse and the problems of such incentives are rooted in ads and SEO.

Whether or not you're willing to pay for Kagi essentially boils down to whether you think search is important enough to justify the bill, with a couple caveats. The first caveat is considering what your personal data is worth to you. The second is whether you think paying for a service subscription is a better business model than paying with your eyeballs on a free service that serves ads.

To help you figure out how Kagi performs, the rest of this post provides some of my takeaways for day-to-day features. To put things into perspective, I average about 1k searches a month and use Kagi across all of my devices: home pc, work pc, and phone.

Search results

Let's start with the search results. The vast majority of my searches are programming-related, either APIs in programming languages, APIs in third-party libraries, or documentation resources and higher-level ideas. In each of these categories I find Kagi provides really excellent results, particularly when compared to my previous search engine, DuckDuckGo (DDG).

My DDG use can be summarized by one symbol: !g. I had to constantly redirect my searches to Google to get good results on the first page. DDG works fine for general searches ("what's this word?", "who's in this movie?") but I had a really hard time using it when doing anything technical. With Kagi, I basically never use !g.

It's hard to judge "quality of search" without exhaustively comparing results for certain queries across different search engines, which I am not going to do. Generally speaking, I'm impressed by Kagi's highest-priority search results. There seem to be fewer blog-spam articles from sites like Medium or dev.to (which you can further exclude, see: personalized results) and more self-hosted blogs, Stack Overflow posts, and Github discussions/issues. I am often surprised by the quality of Github content in Kagi, as many of my queries have been answered by a comment in a Github discussion or an issue that is near the top of the results page.

By far my favorite aspect of Kagi search results is the prevalence of source code links. For example, searching "blank? rails" turns up this link to the Github source as the third result.

These source code links are so insanely useful. Normally I'd have to (a) search the library (b) go to the library website (c) try to find the Github link (d) use Github search to find the relevant file. Kagi bakes all of those steps into one. I often use Kagi to search for a file that I know exists in some Github repository, like "autorun minitest" and follow the Kagi result rather than going to Github and using its navigation features.

Personalized results

I mentioned this in the previous section, but you can "raise", "lower", or "block" certain domains to affect their results in your search output. It's not totally clear to me how significant this setting is, but I use it to prioritize gnu.org documentation for Emacs searches and lower Medium or dev.to results to avoid blogspam. I have noticed that lowered results often sit at the bottom of the page, so if you're a fan of not seeing any Medium articles or Quora questions in your top-most results it's a very nice feature.

Kagi also hosts a public leaderboard that shows you how other people use this tool.

Lenses

Lenses are a Kagi feature that I thought I would never use, but given the right application they're actually pretty helpful. A lens effectively limits search results to a particular domain. For example, a Hacker News lens would restrict results to sites with https://news.ycombinator.com.

I actually use that Hacker News lens frequently to track down articles that I remember seeing, but don't exactly remember the author or title. I can provide some vague search terms about the content in the article, filter by my Hacker News lens, and quickly look through a bunch of submissions that fit the criteria. I find that the Kagi lens performs better for this specific use-case than the Hacker News search on the website.

Things kagi does not get right

Image results are good for the first few columns, but very quickly devolve into complete nonsense. It's actually hilarious how off-the-rails image searches can get with Kagi. I've shared my results with friends quite a few times because the results were so outlandish when compared to the search term that it made for good comedy. This has improved significantly with a recent update but I still go to another source when searching images.

The universal summarizer is a neat idea but is untrustworthy in very subtle ways. I primarily use it when summarizing Wikipedia articles with the shorthand flow: first search Wikipedia ("!w dynamic programming"), then summarize the link ("!sum https://en.wikipedia.org/wiki/Dynamic_programming"). Like all LLMs, the summarizer sounds authoritative but frequently fabricates ideas or misrepresents certain pieces of information. Perhaps this is less of a critique of Kagi and more of LLMs in general, but I have barely touched this feature as a result.

Uptime for Kagi has been generally good, but last month there was a five-hour outage that was the source of much hand-wringing and consternation. If anything, this outage demonstrated to me how much I prefer Kagi to other search engines. The postmortem is worth a read.

Should you use Kagi?

Yes, give it a go! Kagi outlines why you might want to pay for search rather eloquently in its documentation. I echo the philosophical argument and add that the product itself is compelling.


Thanks for reading! Send your comments to blog@mgmarlow.com.