Free API Allotment for Subscribers

gausby

I bought the Professional Plan because I liked the sound of unlimited questions I could post to the FastGPT assistant. I tried it out on the website, and I found the results were good. Better than my experience on the common search engines—also, I was excited about all the integrations with my code editor of choice. Why leave my editor if I could get all my questions answered in right in my editor?

But I hadn't noticed that the unlimited questions only applies to the "web experience."—I had a chat with support where I was informed that the API require me to pay on top of the ~$100 I already payed.

So, I suggest the following—and I know I am no businessman, but hear me out: How about allowing "unlimted" questions via the API but with throttling. Say, a couple of questions per minute; and perhaps a dosen request pr hour. That way the "unlimited" API could be exposed but not abused.

Currently, I could automate a browser, and post from my editor via it and get answers to my questions, but that is just silly.

Just like any other API request, but limited to a handful of requests per minute, and a reasonable amount of requests pr hour.

Vlad

gausby Mostly because any sort of automated use would probably propel the costs for us to the skies, and we are already on razor thin margins. So this is why we ask users to pay for additional scripted usage via the API.

CrunchyFritos

This seems potentially related to this post: https://kagifeedback.org/d/1624-free-api-allotment-for-subscribers

oscarvarto

Hi! I am an Ultimate Subscriber, and was surprised that I needed to pay additional for calls made to the !code bang outside the Orion Browser, for example, in Emacs, with FLOSS gptel (https://github.com/karthink/gptel)

It certainly makes more costly to use the AI for coding in the Code Editor of choice (pressumably, that's actually what most of us coders would want to do with AI for coding assistance!).

Why is it "free" to use it from the browser (Orion, for example), but it's required for more payments for code assistance?

If a user is already paying for !code assistance, it would be fantastic to get the same help in the code editor of choice!

arcfour

It would be really nice to have a free allowance of ₂-5$ in credits as ultimate user to power some smaller private projects and perform some tests.

martijn

Some sort of API access should be included with Professional and Ultimate. I think the Perplexity people do this the right way: Professional subscribers get a monthly $5 credit for API access. Need more? Buy more.

In Kagi's case, limiting this basic API access to Search will be appealing enough for most users.

goto10

Seconded - I'd hope to get at least some credits at the professional tier. I've been working on a Raycast extension (https://www.raycast.com/fearoffish/kagi-search) and I'd really like to use it more, but paying per search over my "unlimited search" plan is a real disincentive. I understand the fear of abuse, but a quota and a rate limit should do just that.

PeterDaveHello

I really support the idea of adding monthly API credits for paid users. These days, not everyone is searching through a browser all the time. Many of us use different AI assistants or set up automated workflows, and being able to integrate Kagi into those tools would make the service even more valuable.

A fair use approach seems like the perfect solution. It allows users to get more out of their subscription in various ways while still keeping things sustainable and preventing abuse through reasonable limits.

Hope this can be rolled out soon!

Nevon

I'm on the professional plan and was actually very surprised that API access was not included. In my book, whether I call the API from the web browser or phone app or from an integration with my desktop OS shouldn't make a difference.

Of course I understand the worry that a poorly implemented or even malicious integration could eat into the margins, but that sounds preventable with some exponential backoff or even just basic rate limiting.

prskr

I can only second what was already mentioned.
Completely understandable that there are concerns regarding margins and everything, personally, I wouldn't even mind to upgrade my plan from pro to ultimate if that would include Search API credits.

As pointed out in another post: especially for search - not talking about anything AI related! - it should not matter whether I use the API to kick of some search or open the browser to do it, apart from breaking my flow by switching context ( @goto10 thank you so much for the Raycast extension! That is exactly what I was looking for to stay in a "tight loop" 😍).

When talking about AI, I would even expect that when the Kagi Search API is easier to use, some of us might rather integrate it into already existing flows (Raycast, OpenCode, Kilo, Claude Code, ... ), using our own LLM credits, than burning tokens in the Assistant.

Independently of that, rigorous rate limits and quotas should be absolutely fine, I don't think anybody would be able to trigger more than 1 search/s or 10 searches/30s (for instance when using an MCP to do a few queries and rank results afterwards). I'm certainly not a representative user (I don't search as much as I used to) but when looking at my usage details, I'm doing ₄₀ searches/day on peek times. Maybe something like 10-20 free searches / day via API could be a good starting point and if costs are not covered re-evaluate?

Currently, I'm working in multiple phases:

Researching stuff in Kagi Assistant
Copying parts / whole conversation either into another thread or to another tool (like OpenCode)
Implement whatever I am working on

So, if I could interact with the Search API directly in OpenCode during the planning phase, I would most likely rather do that, because it spares me copying around results from the research (alternatively, having an API to fetch some "research results" in Kagi Assistant would also be nice, but that's probably something for another thread 😅).

Apart from the AI stuff, it is - at least for me - really only about avoid context switches. If I can search for a certain documentation page directly from Raycast or any other tool rather than having to switch to a browser (or PWA, doesn't matter), it means I'm not losing context and I am less likely to be distracted by something else.

« Previous Page