Hi,
It’s been a long time since I last visited this forum! 
I have a quick question about the AI features on Discourse, I saw that it’s using Qwen/Qwen3-VL-30B-A3B-Instruct (for thread summarizing for example).
Is the LLM running on a cloud? Is it a paid service? What is the business model behind it? Was it a by default choice to integrate AI on this website?
That’s not something I explicitly enabled. I suspect it came with a recent Discourse update, but I’ll double check.
I’ve disabled the summarization features for now and will review any other AI features that may be added on a case-by-case basis.
Thanks for the heads up and welcome back to the forum @josephh 
@sableraph Thank you for the quick answer! 
There’s still some features enabled when writing a new post.
Note that I’m not saying these features can’t be useful; indeed, they are if you are not an English native speaker (used it to correct this sentence) but then the question is where and by who this LLM is run and is it worth it.
According to this post on the Discourse board, the default LLM is hosted by CDCK (Civilized Discourse Construction Kit, Inc.), the company behind Discourse.
Based on this and looking at the admin dashboard, the model appears to indeed be the default open-weights model Qwen/Qwen3-VL-30B-A3B-Instruct, hosted by Discourse/CDCK rather than any third party provider.
Furthermore, the CDCK privacy policy states:
We do not use customer data to train AI models, nor do we share it with third parties for that purpose. Customer conversations and content belong to the communities that create them, and our role is to protect that trust. Customers always remain in control of their data.
That said, we will keep monitoring how Discourse’s AI features are rolled out and whether they align with our community values. If anything changes, we’ll update everyone here.
