3 ms·
I've been subscribed to Kagi since trying it out early on in 2023 and have been paying the $10/mo for unlimited usage since they introduced that tier (if you ar
by lhl 2y ago
I've been subscribed to Kagi since trying it out early on in 2023 and have been paying the $10/mo for unlimited usage since they introduced that tier (if you are a heavy search user, I'd agree the $5/tier "limited" tier doesn't make so much sense). It takes a bit more clicking, but I found 6mo usage view https://kagi.com/settings?p=consumption&range=1 https://kagi.com/settings?p=consumption&range=1 and its nice to check on it after having "lived with it" a while. My lowest months are 500 searches, and a few higher, but one month goes up to 1300 searches. The "killer app", and what made me subscribe when I started was being able to up/down-rank site results myself. Being able to just not see content farms is the starkest difference now vs when I use Google Search (it's striking and seems to make search downright useless).
I've used 0 "AI Tokens" for the entire period. (I've never found much use for their FastGPT or Universal summarizer, but I am already a heavy Claude, ChatGPT, and w/ Gemini 2.0 models, AI Studio user now). I was also subscribed to Perplexity Pro at a $10/mo introductory rate this past year, and while I enjoyed it when I used it, I found its utility somewhat situational and didn't renew at $200/y - I use Claude, ChatGPT, and AI Studio now w/ the Gemini 2.0 models extensively, and when I want search result analysis, I think both of the latter now fill that niche pretty well. I just saw thanks to this thread that Kagi's "Ultimate" tier has access to their new "Assistant" product, which looks pretty well thought out, so I'll be giving it a spin (they should probably give like once a day trial access or something if they're looking to grow it, most people probably wouldn't put down $25 sight unseen): https://help.kagi.com/kagi/ai/assistant.html https://help.kagi.com/kagi/ai/assistant.html
- entropyie 2y agoBeen using it for a year or two, but recently I've been making heavy use of LLM results using the ? character. I actually discovered this by accident while searching. Even though I have a ChatGPT subscription, the Kagi interface is faster to use for simple questions and I have web results underneath as a backup if the LLM answer isn't good enough. I agree with another poster that being able to test ultimate would be nice before I shell out more money.