Sure it can, easiest way is to get ollama, then `ollama run gemma3n` You can pair it with tools like simonw's LLM to pipe stuff to it.

This should run fine on most hardware - CPU inference of the E2B model on my Pixel 8 Pro gives me ~9tok/second of decode speed.