Keyword collections for AI as lens

Sabin_Stargem

The concept of lens for Kagi is currently all about focusing on a list websites in conjunction with a search phrase - but I think that AI could eventually intuit the intent of a search, based on what the user states, plus a chosen library of keywords that was paired with that search. An hypothetical example a decade from now:

I want to make a search concerning videogame source ports, and have made many such queries in the past. As the years went by, I have found many projects that pertain to my interests and saved certain phrases with Kagi. These phrases are part of a "Keyphrase Manifest". My current search is as follows...

"Kagi, I want to know about recent official source projects for 2D platformers available on KagiStation 9." Then I select my Manifest to form a lens. Within this manifest, the following phrases are found:

"Recent". Things that four years old or less, with the most recent first.
"Changelog". Only list features.
"GZDoom", an unofficial source project port of a PC game called DOOM (1993?). It is for PC.
"Ship of Harkanian", an unofficial source port of a console game called Ocarina of Time. It is for PC.
"AM2R" an unofficial remake of Metroid II for PC. Has the spirit of a source port.
"Super Mario 64-DS PC", an unofficial source port of the console game called Super Mario 64. It is for PC.
"Project 06", a formerly unofficial source port that been adopted by SEGA based on Sonic 2006. It is for PC and other platforms.

Kagi looks at this manifest, and tries to find commonalities with the search query. Assuming tech has advanced enough, Kagi can see that all these games have stylized aesthetics, originate from the 90's and 2000s, most are gamepad driven, and so on. Then it tries to find recent source code projects on consoles that feel similar to the manifest.

While a bit pie in the sky at the moment, I believe that AI will be able to someday pick out such themes. It is just that it is a pain to type out the key phrases for searches, especially if you do a particular type of search fairly often.

Vlad

Sabin_Stargem Can you summarize the MVP of the ask that we could implement right now as a simplest possible version?

Sabin_Stargem

Vlad The most simplified version? Hm. Assuming we use the current FastGPT page, you can add a "save this query." button, which adds the query to the search page as a shortcut. Clicking that button adds that text into FastGPT as the foremost search if there is no text already entered.

However, if there is text already in the prompt, clicking on the saved query will try to make the AI contextualize the main query.

Primary query: "Kosher food for a guest."
Saved query A: "Wendy's, Burger King, or Carl's Junior near [ZIPCODE] are my favorite fastfood."
Saved query B: "Discounts when available."

Based on these three, the AI will search the menus of the stated restaurants for grub that is kosher, then checks for a discount.

/Simplified version.

The original suggested manifest system is a much longer and indepth than the above example, because FastGPT tends to go after concepts and ideas that are unrelated to the query. By using a manifest, FastGPT can say "These are the most important parts of the topic, let's research this first".

Kinda like being given a stack of reference material, rather than a boss only saying "Look into that." and walking away.