It definitely can, it just depends on the complexity of the commands you want it to generate. Most of the time a small local model will do what you want. If you know of a small local model fine tuned to generate shell commands I'm all ears. I usually run it with a 9B qwen model.

Pairing this with dictation is fun though, and speeds things up a lot for terminal heavy work.