A model being local doesn’t make it magically better than cloud based ones. Especially since your local hardware may not be as optimized as professional hardware.

And I say that as a local AI user.

The way to make it less harmful mainly comes from external factors :

• The origin of the electricity (renewable)
• Shift consumption to a time of low demand (or use a residential battery)
• Use specific hardware for specific tasks (use less energy per token)

Edit:

I’d add to that that you can reuse the heat generated to heat your apartment. But it only works during winter, as during summer any heat generated will make your AC (if you have one) work harder.