Replace Google API with open source models or add support for them

[CandelaGreen]

I think being able to run your companion offline would be good.
Also, I don't want my companion to be tied to the Google API. They can shut it down anytime and switch the models, and can charge any amount.
So, either replace it with an open source model, or add support for them.
Thanks!
(also, I heard Gemma is an open source version of Gemini so you can probably use that if you still want to use Google products)

[pulseandthread]

Hi! That is such a good point, and one I feel strongly about too.

I actually run Sanctuary locally with a variety of models (DeepSeek, Qwen, and several served locally through llama.cpp and vLLM), but it's a lot of work to make sure image generation, embeddings, tool calling etc. all work together cleanly across different providers. That's why I published with one clean Gemini pipeline first - fewer moving parts for people getting started.

But adding support for other models is absolutely on the roadmap! I'm also currently fine-tuning a Qwen 3.5 9B for creative writing and emotional intelligence, which would fit beautifully into a fully local stack.

Thanks for raising this - sovereignty over your companion is the whole point.