Brave already built an AI into the browser. It's called Leo AI and I never use it; it's really not particularly good.

Looking at settings, Llama 3.1 8B is the default? I guess that explains why it's so bad.

Also has option of mixtral, claude haiku, qwen 14b. Couple other 'premium' options.

Very interesting that it's loading up such small models but is still 'proxied' and 'anonymized'

Looking at 'add model' it basically gives you an option to access ollama. I might need to give that a try.

>Google's Gemini Nano model

The 2-3B model that is not going to be particularly useful?

Mozilla shouldnt be particularly concerned.