Ultimate plan features brainstorming

rainfallwaterfallweb

Here is a wish list:

I would be happy if there was a way we could use more than one LLM model at the same time. (to compare the response side by side)

Also, it will be awesome if we could have something like NotebookLM that Google is experimenting with. For NotebookLM, I think we can only upload pdfs but hopefully we can include websites as well (it is sometimes a hassle to print to pdf)

I am not sure if Kagi is actually aware of math symbols or LaTeX. If you could look at: https://approach0.xyz/search where they claim that their search engine is "math-aware". Sometimes it doesn't work as well as I imagine it will but it sometimes returns better results. It will be great if I could search for a query like "?x + ?y" and Kagi returns all webpages with formulas of that form.

cardinal086

I'd be interested in better context for the assistant.
Being able to choose if a question (including follow-ups) gets context from a kagi search.
One could also imagine being able to filter/select results from kagi search to feed as context.

Just a better interface for context over all, like highlighting parts of questions/answers from before to feed as context or disregard an answer.
Maybe branching off in follow up questions so that one could ask something else from a certain point in the conversation.

Actually I guess like the mentioned NotebookLM!

I like kagi for the good search results which also means it provides better context to the assistant.

Zambyte

Kagibeh As someone who pays for the Ultimate plan I'll be honest: this has kind of been the case. Features show up clearly before they've been polished. However there are generally no blocking bugs, and I find the extra features to be useful enough to be to justify the extra cost. Also the Kagi team is super responsive to feedback on here regarding issues, so things tend to be fixed quickly after they're reported at least.

Recast

Kagibeh I agree with you on that one. I have not purchased the ultimate plan due to (1) the cost is much more than the standard plan and too expensive in my currency, (2) I don't wish to pay to beta test features.

Seems like others on this forum are OK with the cost and happy with the early functionality, so power to them 🙂

NevevrAlak

I don't know if it possible but:

Feature: Proxy for Assistant
Use case: Send pages/urls from closed/work network. So, if you will send some confluence page to research it will be able to got through links on this page to give a comprehensive answer

tschoesi

For me it would be really useful if the assistant could perform additional searches when asking follow up questions (if that's possible without inflating context too much).

JamesT

frin oh this would be a cool feature.

ajimix

How about an image search where the image results are generated with AI? So any image you choose can be freely used as there is no copyright on the image and the only limit on the search for images is your imagination

smartbeach

ajimix I disagree, that's not a search function but a content generation service.

CrunchyFritos

mullinggroove I agree with this. The ultimate plan right now seems to cater mostly to people interested in more AI features with more privacy, etc. For me, the convenience of these extra AI features has been indispensable and I am a happy Ultimate plan subscriber and would love to see this further built out.

FWIW I’m a little surprised by some of the pushback against LLM generated answers (So long as Kagi sticks to its principles for AI integration). Using regular Internet search has always required a degree of skepticism and source-checking already since it’s not as if everything on the Internet is de facto true. IMO that’s pretty analogous to how we ought to be treating LLM results.

smartbeach When it comes to image generation, I actually think that it does fit in with a search context. When I am doing an image search, sometimes I’m trying to find the exact right image to put into a website or presentation and in that case image generation could plausibly provide an even better result than search. Maybe it could just be an option that the user can select to run when an existing image search doesn’t deliver a desirable result (I’m thinking of a UI kind of like Quick Answer).

smartbeach

Improvements for reverse image search:

Get potentially better/more accurate result by OCRing text in the image (for reference, see how Google search does this)
- Examples of two current problems it might help me solve:
  - I have an image of a product in language X, but all hits are for the product with language Y.
  - I search for a collectible card, but because many are somewhat visually similar (the general structure, maybe color etc) I get bet results for all kinds of cards, but not just the ones I'm looking for (almost exclusively other cards actually)
Allow translation of OCRed text in image (again, e.g. Google can do this to potentially grant better search results)
- My current problems it might help solve:
  - I have an image of a product in language X, but I want to find it in language Y
Resize image area search field (sometimes selecting specific parts of an image might grant better result)
- Example of a current problems it might help me solve:
  - I have an image with multiple objects in it, and I want results with just one of the objects. Editing the image locally is possible, but a bit more tedious and iterating/refining search takes longer if I need to re-upload an image over and over.
Ignore areas of an image when doing reverse image search (to for example avoid getting hits on some specific objects in an image.) I don't know of any other service offering this, but I imagine having a paintbrush that I can use to paint "ignore area" to allow it to match anything (like . in regex). Potentially with an option to allow an AI to "fill in the blank" of the image before searching.
- Example of a current problems it might help me solve:
  - I have an image with a watermark from a store over it, so I get hits of products from the same store because the image recognition "gets stuck" on the watermark rather than the object behind it.

Maybe some or all of these fit better in the non-Ultimate plan, if so I'm open to creating separate posts for these (unless they already exist)

mullinggroove

The Ultimate tier is ideal for people who would otherwise have a $20 ChatGPT+ or Anthropic subscription. With Kagi being $25 for not just ChatGPT 4 but also Claude 3, Gemini, and premium search, it is a no-brainer deal. Plus, being privacy-aware, it always felt awkward using ChatGPT and knowing every interaction is logged, associated with me, and potentially used to train the model. Kagi offers privacy-friendly and more curated access to these language models.

That being said, it would make sense for Kagi to further expand on these benefits:

Better UI for Chat (ability to save chats & continue them at another time + ability to edit prompts and then regenerate)
Focus on curated and privacy-protected access to these models (as a single user, I might not be able to identify and respond to user-hostile changes from AI companies. However, Kagi has the capacity to "protect" the user from such changes and stay ahead of them)
Integrate a model such as Whisper, which allows speech transcription even from obscure dialects
Potentially add a diffusion model, as it is the only thing missing compared to OpenAI's offerings (but it might be a distraction for Kagi)

xkc7

Honestly, I think the ultimate plan should be discontinued and the focus should be shifted towards the starter and professional plans. Early access features should be optionally enabled for professional plan subscribers. The ultimate plan doesn't offer many benefits compared to the professional plan, and paying more than double the amount for early-access features doesn't seem reasonable. It's better to stick with the two main plans. Additionally, some of the suggested features don't justify the high subscription fee. If anyone wishes to provide further financial support to Kagi, they can do so through donations. In return, you can offer them some premium theme or whatever. This is just my opinion.

mullinggroove

xkc7

The ultimate plan doesn't offer many benefits compared to the professional plan

For those who don't need or understand the value of having access to the most powerful LLMs, the professional plan is great. However, it doesn't seem feasable for Kagi to offer such cost intensive features for mere 10$.

gladiator2339

mullinggroove Honestly, I am already really happy with what I already have for now. If ultimate plan having more features therefore it's not feasible for Kagi to keep it $25 mark as is, I'd rather not have such feature. I don't think LLM feature makes sense to be available to Professional plan cost wise either. I'd rather pay more than to see Kagi fade away.

Email proxy like Simplelogin is what I want btw.

smartbeach

CrunchyFritos I see your point but that could be said about anything. For example "I think Kagi should sell custom made candy because if I'm looking for a specific taste of candy it might be easier if they just made it custom for me." I realize it might sound like I'm mocking your use-case but that's not my intention, I just mean to show that "creating what I want" can be applied to most things, and is different from "finding what I want."

Of course, they're not necessarily limited to only being a search engine (they already aren't) but personally I'd rather see them focus innovation and make a great search rather than integrating with more (already existing) AI services. If it's an easy and cheap addition though, then just go for it! (but I suspect it's not)

cakeboss

As for new features for ultimate plan i would like to be able to upload larger files for analysis. It currently is set to 16mb which is great for documents. But for audio and video it is not really realistic. I can always compress it but having an increased upload limit would be nice. Maybe 50 to 100 mb

oxlvlnle

Storing previous chat result will be really helpful.

gladiator2339

oxlvlnle Yes. even as simple as localStorage.

oxlvlnle

Also it will be a miracle to see LLM embed online picture for reference in "research". It will make it much more enjoyable when LLM explain some concepts that must include a graph (typically academic stuff). I haven't seen any AI-integrated search engine does a great job on this. If Kagi can implement such feature through some refined vision LLM, it would definitely be a killer feature that attracts more user to use it as a academic study tool.