Post-processing: a way to verify the LLM is actually working?

[jarvis2k1]

Handy can't currently tell you whether your post-processing LLM is actually alive. Refreshing the model list only proves the endpoint answers /models — it never runs the model. When my local server died, post-processing silently fell back to raw transcripts and I didn't notice for hours.

Would a "Test" button in Post-Processing settings — one that sends a real completion and shows the model's actual reply — be something the community wants? I have a working implementation, but given the feature freeze I wanted to gauge interest before opening a PR.

Related: the silent fallback itself may deserve a fix regardless (a toast when post-processing fails), since that's a bug rather than a feature.

[cjpais]

I think test would be really good, you can submit this. It helps clarify things and can fit neatly in the app I think