This is probably an obvious suggestion for when it is financially viable, but I couldn't find it here, so figured I'd mention it. But I think it would be good to see Kagi eventually hosting its own instances of open models like Deepseek and GLM, rather than relying on outside providers.
Ultimately Kagi has to trust the providers aren't lying about their privacy agreements and if there's one thing we know about big tech in general and the AI space even more so, lying and stealing are extremely popular and generally have very low penalties compared to potential profit. Having Kagi host some models themselves removes one layer of potential dishonesty. I understand that mean users still need to trust Kagi themselves, but that way we are trusting one specific company based on its reputation, rather than having to trust a chain of companies (like Fireworks, Nebius, etc.)
It could potentially even be a higher priced option, as it can obviously be expensive to establish and run. So for example you could have the option "Deepseek" or "Kagi-Deepseek" for better privacy guarantees.
I think this will be particularly important if Kagi introduces features such as Mail with AI integration, or the ability to reference past conversations in the Assistant. For example I may be fine to use Deepseek with some random provider for most matters, but if I'm going to give it access to the contents of my emails, I'd prefer everything stay within Kagi.
