this seems like a crazy idea as the cli client has nothing to do with how many tokens per second the api streams
The api is never the bottleneck but how fast the cli provides context. So just by using ripgrep it will be faster than using grep. On top of this concurrent code search compared to sync etc
A bit comparable to how pressure was put onto payment providers to restrict sites with extreme adult content.
Side note: I find it super interesting in the filing https://fingfx.thomsonreuters.com/gfx/legaldocs/xmpjezjawvr/... that Google was able to provide them with the exact number of automated requests broken down by company.