I desperately hope kagi adds a vision llm in chat mode.
76
Ultimate plan features brainstorming
5 days later
Vlad I am currently using it. It worked, but would be more efficient to polish it a bit more? for example, the vision llm should allow prompt and picture to be uploaded at the same time. I always need to wait for the research assistant to output something I probably don't want before I type in additional prompt. And there is no way to add a picture&document in the middle of a research. The previous conversion will reset once a pic is uploaded.
So for Ultimate plan features, probably adding a working MLLM would be the best? No need for choosing, no need for a lot of conversion, just one place for a whole research project (and that's why storing previous conversation is quite crucial).
In my personal opinion it would be best to enable ultimate plan user to have several custom assistant too. Currently only one is included?
OldMan973471 wait, isn't the AI credited unlimited currently for ultimate plan?
Vlad unstickied the discussion .
oxlvlnle It is, but the reason why I pay for Unlimited is so that I'm not a financial hit to Kagi, I want them to benefit. So if I have an especially heavy research weekend and some extra cash, I'd love a way to throw it Kagi's way. 
11 days later
Stealing this idea - https://kagifeedback.org/d/4050-add-the-ability-to-sponsor-a-bug-reportfeature-suggestion
Maybe Ultimate users could get a few credits each month/year to vote on features/bugs they really care about.
Text to speech - ultimate could have the ability to play an audio of the printed text. There is an OpenAI API that does a good job, and since it costs money, it would make sense to include it with Ultimate only: https://platform.openai.com/docs/guides/text-to-speech
User could have the option to choose the voice and ability to download the recording as mp3/aac.
There are 2 models, regular TTS (costs $15/million characters) and TTS-HD ($30/million characters). I think the regular is pretty sufficient.
silvenga I like this idea. Or at least some way to explicitly indicate that Ultimate users get some preference in terms of helping to set Kagi priorities/roadmap.
16 days later
frin Being able to “talk” to Kagi, especially the fast/expert mode with follow-up questions would be a cool ultimate plan feature.
3 months later
yes please
2 months later
Vlad stickied the discussion .
Maybe a niche usecase, but I've always wanted to use search/assistant from the command line.
I'm a (n)vim user and dislike having to switch to the browser to search or use assistant.
I realize that the api limitations are in place to prevent abuse and overuse, which is probably the biggest challenge in exposing this to a cli tool, so I would be perfectly ok with having a lot of throttling restrictions on this as the operations I want to do would be 100% the same as I do on the web, I just want to stay in nvim.
Elifino I currently use Shell GPT: https://github.com/TheR1D/shell_gpt
it connects to openai but once an assistant api is available, you could use that as well instead of openai.
I use it a lot but mainly so it generates commands for me. for example I can
sgpt -s "extract all even pages from document.pdf and put them in a new folder 'newfolder'"
(-s means shell command)
and it'll give me the command and ask if i want to run the command
6 days later
Thibaultmol Thanks for sharing, I'm using gen.nvim connected to local ollama (I don't trust openai from a privacy perspective), but web search functionality is what I lack. Having this integrate with kagi would be superb 
Elifino consider upvoting this: https://kagifeedback.org/d/5115-provide-an-api-for-the-assistant
What about using kagi to get local context of the page? Like pages behind a login. Kind of like brave leo does but in a much better and accurate way??
5 days later
I feel like priority support would be a great addition to the ultimate plan