Provide an API for the Assistant

pasta

Many code editors have LLM chat functionality either built in or as an extension, providing an API would allow one to chat without ever leaving code editor.

Jimjiminyjim

Thibaultmol I came into using LLM’s via Kagi. I don’t know how many tokens I use. What I do know is that I can’t get my todos into it to help my adhd mind make a plan and that if I keep a chat going for days on end, thinking it will remember my projects, goals, todos, routines, and approaches to prioritizing and scheduling, it doesn’t. It doesn’t appear to be aware of the vast majority of the middle portion of the text of two weeks of planning. So if I want to support kagi, and use an LLM in this way, I do have to pay two subscriptions. I don’t know how I would tie into an API, but maybe kagi would do something in the browser? I am aware of the up to 5000 words you can put in the settings as instructions, which I have been using, but it’s not the same as including thoughts and approaches distinct to each project/goal/routine/schedule.

blobmartin

Jimjiminyjim you have to tell it to save your chat if you want to save a chat. Part of it's features are that it's private by default and part of that is it will delete your chats after a day, unless you tell it not to.

poli0iq

Since Kagi now has price-based limits for LLM usage, abuse of Ultimate accounts shouldn't be a problem anymore. It would be nice to have an API that would use the primary account credits.

strategic040

I would love to build my own tty application while having the search capabilities / model selection kagi provides implementing all the features that kagi assistant currently lacks. Also using an ollama bridge to use it in IDE.

youngji22

As an aside, it would also be nice to be able to define your own functions for tool calling so that you can allow the assistant to access certain knowledge bases, for instance.

oxlvlnle

Really want this feature.

eSko

+1 for this functionality. Like many others, I cancelled my other LLM subscriptions after subscribing to Kagi, and I would love to utilize the allocated Ultimate tokens via integrations through various apps that support APIs.

lyc1

I'm about to downgrade my Kagi subscription from Ultimate to Pro to streamline all my AI subscriptions because of this missing feature.
I love Kagi search engine and the integrations you've done with many AI models, but this ecosystem is too closed for me.

oxlvlnle

lyc1 I wish this feature could be implemented too, but I guess there is a reason that Kagi doesn't want to provide this at the moment.

bhagwad

There's a real demand in the market for LLMs that can natively access the internet for real-time results without a tool or a plugin. As far as I know, right now only the Gemini models have that capability and even those are not easily accessible through the standard AI aggregator services like OpenRouter etc.

Kagi exposing an LLM API with native web access for all models would be a game changer, IMHO. I think it'll really drive an increase in subscriptions like a kind of Trojan horse for accessing the other features of Kagi like search etc.

blobmartin

bhagwad ChatGPT will search the internet. I agree though that the killer app for the Kagi API would be the fact that Web searching is done using Kagi's search technology which provides a higher quality answer IMO.

blobmartin

Thibaultmol I assume you are seeing this as "Abuse" because someone might opn lots of accounts?

blobmartin

RoxyRoxyRoxy Good thing there is a voting system that show where their users want their attention focused.

bhagwad

blobmartin

That's true, the web app of ChatGPT can search the Internet. I was referring to APIs, of which only Perplexity and Gemini have native Internet access. But Perplexity search is driven only by Llama, whereas Kagi allows us to use any model with search - a huge bonus IMHO. And Gemini grounding is, of course, only available with Google.

As far as I know, there exist no other APIs of LLMs that respond with search-powered results. Kagi can be unique in that. And for subscribers, there can be a free tier usage. In fact, right now I use Gemini exclusively for search, which means I'm not really using my Kagi subscription. An API will help me get more value out of the monthly fee.

id

With recently added fair-use enforcement, and detailed billing for consumed AI tokens, there is absolutely no reason to not let users with Ultimate subscription access LLM models via OpenAI-compatible API endpoints.

seany

I'm looking for this too. The underlying API behind the existing web implementation doesn't look too complex after examining a few recorded HARs...

johnnyasantoss

Hello,

I'm writing based on the conversation I had on Discord about this feature (from this thread https://discord.com/channels/1256077108111868035/1407741204870729859)

My core idea here is pretty straightforward. Kagi Assistant has unique capabilities like personalized search and Lens. If Kagi were to expose these models via an API that's compatible with OpenAI's standards, it would open up a world of possibilities.
From my perspective, and I'm sure many of you share this, the benefits are huge:

Seamless Integration with Self-Hosted Tools: I use self-hosted platforms like OpenWebUI and AnythingLLM for local/private RAG (Retrieval-Augmented Generation) setups. These tools already support OpenAI-compatible endpoints. An API from Kagi would make it a true "plug-and-play" solution, allowing us to leverage Kagi's excellent search grounding with our private documents.
Privacy & Control: Keeping chats and AI interactions in a single, controlled environment is paramount. An API would allow for private, centralized storage of our AI conversations, mitigating concerns about data privacy and user data retention nightmares.
Consolidated AI Subscriptions & Credit Usage: Like many of you, I'm trying to optimize my AI tool subscriptions. If Kagi provided an API, I could use my existing Kagi credits directly instead of paying for multiple platforms. This would greatly reduce the number of separate AI subscriptions I need.

I see this as a potentially low-lift, high-gain feature for Kagi. While I understand there might be concerns about token usage or abuse, Kagi already has robust pricing models and usage limits in place that could easily be extended to the API. Extra AI Credits bundles could be deducted from the API Credits, just as searches do.

This feature would also allow users to use their custom assistants as models and have other external models natively be able to search, summarize, and ground facts with the other Kagi APIs.

nichu42

What would I like most about this: Kagi search powered LLM in my mail apps (Thunderbird and Fairemail). I currently use OpenRouter. But I am hoping to be able to use Kagi one day, with access to search + custom assistants w/ permanent storage.

« Previous Page Next Page »