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