6 ms·

The thing still giving me pause is a lack of "bring your own model connection" between saas uh services. If I already pay for Gemini Advanced (or OpenAI/ChagtG

by basch 1y ago

The thing still giving me pause is a lack of "bring your own model connection" between saas uh services. If I already pay for Gemini Advanced (or OpenAI/ChagtGPT Pro), I then have to pay for it again at every service that offers a Pro 2.5 (or 4.1/4o) tier. I should be able to connect my Gemini Advanced access to any service that offers Flash and be able to upgrade. Signing up for a bunch of services is starting to feel like being triple, quadruple or more dipped. Similar to how I am annoyed seeing media content cross licensed to three streaming services and not getting a bill reduction when subscribed to multiple services with the same content.

everforward 1y ago
I doubt that fits in the usage model for those pricing plans. They're priced assuming it's inconvenient and less used because of it. I keep hearing they're losing money anyways, so despite being triple or quadruple dipped you're still paying less than it costs to run the models.
- yellowapple 1y ago
  > I keep hearing they're losing money anyways Hearing from whom?
- basch 1y ago
  How would it cost Kagi anything, if I bring my own keys and all the costs are offloaded to my personal Google/ChatGPT account? I pay them the $10/month and get the Unlimited Pro access (but only to models I subsidize the cost of.) If anything it would save them money to offload my usage to my account vs using up some of their Flash tier quota. It would be ideal for them for LLM access portability to let them offer higher end models and have the end consumer pay directly for the usage. Id much rather pay for one AI License, and a small fee to each service I use, rather than paying a high tier AI Bonus price at every single service.
  - macrocosmos 1y ago
    I think they might be talking about the cost to the model providers.
    - basch 1y ago
      Then Google should offer Advanced Portable for $30 that includes being able to plug it into third party services. With some kind of policing to prevent it from being used as the entire backend to a service (for all users) itself.
      - everforward 1y ago
        Third party usage is too disparate for a single flat rate to make sense. Perplexity probably uses orders of magnitude more tokens than Kagi does (the former uses tokens on every search, the latter only does if you manually invoke the Assistant). Then there's the disparity in how many third party services each user uses. Some will use one, some will use 50. Google could charge per integration, but that's basically the status quo.
        basch 1y ago
        Why would the number of integrations be the price point and not token tiers? I might want to connect my account to 100 services to test them, but only perform a couple uses each before forgetting about it.
        everforward 1y ago
        I agree that makes sense, but it makes potential subscribers to the third parties nervous. I like Kagi Assistant because my payment is capped. I get nervous plugging my pay-as-you-go API keys into random software because of the risk they rack up a $1000 bill doing something I wouldn’t have paid $20 for them to do. The other three things are that economies of scale make it cheaper for Kagi to buy a bajillion tokens, Google et al don’t want to manage the customer side of things (what service ate how many tokens?), and service providers don’t want you seeing their “magic” in your console. Seems like there’s a lot of power in the system instruction side of things, and Perplexity probably doesn’t want you seeing their prompt.
eightysixfour 1y ago
I really feel like the software model is going to start to break in a couple of ways in the near future. Instead of vertical software slices, many people are going to want their single “horizontal” agent that they pay for (e.g Gemini Advanced, Claude) and connectivity to all of their other services. MCP (which I personally think is a mediocre protocol, so it will probably win) as the glue for a bunch of services we OAUTH against and choose when our agents do/do not have access to certain tools. The idea of a “GPT” App Store from OpenAI was sort of right, but just wrong enough. We are going to have an App Store inside of our preferred AI platform and subscribe/connect our other services from there.
- basch 1y ago
  I prefer identity as the top layer. Give me an identity portal that acts as an account manager, access to api layer, let me perform password resets, and let me avoid 2fa if I’m logged into my identity manager account. It should allow bidirectional control: both what services have access to my identity, and what services anything in my identity are granted access to access. Things like LLM, cloud storage etc should be obscured away from services so they can’t tell my storage providers from one another. All access between services should cascade down from my identity portal.
  - orand 1y ago
    Who is going to pay for identity as a product and not just expect it as a free feature of another product suite?
    - basch 1y ago
      Why not both? Use Apple or Microsoft or a third party identity provider.
  - jcgl 1y ago
    I mean, that sounds basically like Oauth provided by the LLM provider, right? The provider just needs to build out the features.
- 1y ago

by basch 1y ago

everforward 1y ago

I doubt that fits in the usage model for those pricing plans. They're priced assuming it's inconvenient and less used because of it. I keep hearing they're losing money anyways, so despite being triple or quadruple dipped you're still paying less than it costs to run the models.

yellowapple 1y ago
> I keep hearing they're losing money anyways Hearing from whom?
basch 1y ago
How would it cost Kagi anything, if I bring my own keys and all the costs are offloaded to my personal Google/ChatGPT account? I pay them the $10/month and get the Unlimited Pro access (but only to models I subsidize the cost of.) If anything it would save them money to offload my usage to my account vs using up some of their Flash tier quota. It would be ideal for them for LLM access portability to let them offer higher end models and have the end consumer pay directly for the usage. Id much rather pay for one AI License, and a small fee to each service I use, rather than paying a high tier AI Bonus price at every single service.
- macrocosmos 1y ago
  I think they might be talking about the cost to the model providers.
  - basch 1y ago
    Then Google should offer Advanced Portable for $30 that includes being able to plug it into third party services. With some kind of policing to prevent it from being used as the entire backend to a service (for all users) itself.
    - everforward 1y ago
      Third party usage is too disparate for a single flat rate to make sense. Perplexity probably uses orders of magnitude more tokens than Kagi does (the former uses tokens on every search, the latter only does if you manually invoke the Assistant). Then there's the disparity in how many third party services each user uses. Some will use one, some will use 50. Google could charge per integration, but that's basically the status quo.
      - basch 1y ago
        Why would the number of integrations be the price point and not token tiers? I might want to connect my account to 100 services to test them, but only perform a couple uses each before forgetting about it.
        everforward 1y ago
        I agree that makes sense, but it makes potential subscribers to the third parties nervous. I like Kagi Assistant because my payment is capped. I get nervous plugging my pay-as-you-go API keys into random software because of the risk they rack up a $1000 bill doing something I wouldn’t have paid $20 for them to do. The other three things are that economies of scale make it cheaper for Kagi to buy a bajillion tokens, Google et al don’t want to manage the customer side of things (what service ate how many tokens?), and service providers don’t want you seeing their “magic” in your console. Seems like there’s a lot of power in the system instruction side of things, and Perplexity probably doesn’t want you seeing their prompt.

eightysixfour 1y ago

I really feel like the software model is going to start to break in a couple of ways in the near future. Instead of vertical software slices, many people are going to want their single “horizontal” agent that they pay for (e.g Gemini Advanced, Claude) and connectivity to all of their other services. MCP (which I personally think is a mediocre protocol, so it will probably win) as the glue for a bunch of services we OAUTH against and choose when our agents do/do not have access to certain tools. The idea of a “GPT” App Store from OpenAI was sort of right, but just wrong enough. We are going to have an App Store inside of our preferred AI platform and subscribe/connect our other services from there.

basch 1y ago
I prefer identity as the top layer. Give me an identity portal that acts as an account manager, access to api layer, let me perform password resets, and let me avoid 2fa if I’m logged into my identity manager account. It should allow bidirectional control: both what services have access to my identity, and what services anything in my identity are granted access to access. Things like LLM, cloud storage etc should be obscured away from services so they can’t tell my storage providers from one another. All access between services should cascade down from my identity portal.
- orand 1y ago
  Who is going to pay for identity as a product and not just expect it as a free feature of another product suite?
  - basch 1y ago
    Why not both? Use Apple or Microsoft or a third party identity provider.
- jcgl 1y ago
  I mean, that sounds basically like Oauth provided by the LLM provider, right? The provider just needs to build out the features.
1y ago