no bs answer, please: does Venice run it's inference Infra for oss models or not?

When reading about Venice’s privacy you put a lot of stress in not storing prompts. There are schematics of “distributed GPUs” [here](

https://venice.ai/blog/how-venice-handles-your-privacy) but you don’t really address the core of the question: when one uses GLM or Kimi models (which are open source); is it Venice running the infra? i.e. are the “distributed GPUs” run by Venice or run by, Z.ai for example where Venice acts as a wrapper around their inference?

closed source models, for sure, has to be a wrapper; but for the open source ones is not clear from any help centre article. It addresses the question partially and leaves a LOT of room for interpretation (and then pushes VVV tokens without answering the question).

would it be possible to document which models are provided end to end (inference included) by Venice and which aren’t? The copy in the tags at model selection is also pretty vague:
“This model is private and not prompt data is stored in any capacity” « in a generous interpretation this implies that Venice DOES run it’s inference for these models, but why choose vague language when you can be specific? The reverse can also be interpreted: Venice doesn’t store any data about the prompt, but nothing was said about the inference provider (if there is one).

more mixed signals: some oss models are pay per use (which hint at a wrapper model; GLM is not pay per use).

All I’m asking is to be transparent and clear about it. I cancelled my pro subscription because while testing GLM and Kimi across Venice and other services, Venice’s was the best; but it begged the question as to privacy; and the more I read the more obtuse the answers got.

Can we please please please be clear on that? I’m fully aware that you might not have a full fledged team to run inference for a bazzilion models; but knowing if Mistral or GLM models in Venice are fully ran by you or a provider is very important information to know.