Live chat mode

gladiator2339

Live chat mode, just like Google's Gemini or OpenAI's GPT can really make the search experience better, especially with mobile devices. Obviously this may be quite costly so it can only happen with Ultimate plan and honestly, that is the only feature that Kagi desperately needs and yet missing.

Kagi's mobile app can have Live chat & lens feature that will enable users to interactively chat and search information with voice or from an image.

laiz

Could you share further details such as images/videos of how other services do this, and what you'd like to see Kagi do? What would be the use cases for this feature?

gladiator2339

Sorry for not being clear. I thought that live chat was something that is widely recognized thing, turns out it was just what Google is calling their voice-enabled conversational LLM feature.

Basically, this. But also with the capability of searching through the web, just like current assistant can. If Kagi's LLM offering is really going to make it as an assistant, I think this sort of live conversational interaction is a must.

Nashwan

I was under the impression that kind of freeflowing conversation is only possible in ChatGPT 4o because it is fundamentally a multimodal model. By the looks of it, it's been less than a week since they enabled it for developers via the API?
https://openai.com/index/introducing-the-realtime-api/

Just from a privacy perspective I don't see myself wanting to send a live audio feed through to OpenAI, but we all make our own risk decisions on this kind of thing! I saw a cool demo the other day of Whisper-turbo running locally on-device in a browser for speech-to-text, could be a way to allow people to talk to Assistant if their device doesn't natively support STT.

gladiator2339

Nashwan OpenAI's API wasn't exactly what I was thinking. I was thinking more like a STT & TTS combination so that it actually would use Kagi's assistant features, not just a wrapper for another model. Even though with some added latency, I think that's more in-line with Kagi's mission.

And that new API is crazy expensive(~$200/1M). I just cannot see how that's going to work, at least for the time being with that price.

eirk

gladiator2339 BTW I think its better to compare using time, as speech/text is consumed by people much slower than text. The cost for output is equivalent to $14.4/hour, which is pretty reasonable to me