This is probably similar to the Wolfram|Alpha + Siri integration (which AFAIK no longer exists):

- Kagi will rely on Wolfram's okay-ish stack to immediately add features to their engine

- Kagi will, over time, 1-by-1 remove the reliance on the (generally slow and expensive) Wolfram API with built-in features

- Once the meat of the features have been rewritten internally by Kagi in more performant and self-controlled ways, remove the dependence on the slow and expensive Wolfram API

Not saying that's a bad thing, but it's quite likely this is how it plays out. We've seen it before.