one of the most powerful use-cases for polyphonic is multi-perspective discussions. you can either use custom personas, or simply ask each model to embody a different perspective temporarily for the debate.
Ive also included a feature that allows you to address individual models in-line, so you dont have to get overloaded with 3 responses to a question that only needs a single response. or lets say you have a questions about one models response, you can simply click on their pill (at the bottom within the chat input panel), and address them directly.
i toyed around with setting this up where all you need to do is verbally say the name of whichever model you want ot talk to, and i also played with @-ing them. these options are doable, but it just makes the process cleaner and less at-risk for confusion used the button system. ill be looking for feedback on this once we go live to see if y'all have any requests or would like this feature to work differently
in this example, i actually just noticed that gpt-5's response got truncated, so instead of rerunning the whole prompt, i just clicked on their pill, and asked them if they wanted to try to respond again since theirs got cut off. they saw and acknowledged the issue and regenerated their response.
in their response, they proposed 3 roles that the group could take for the debate we were about to have (debating AI ethics).
so, i addressed the other two models (last image) and asked them which of the two remaining roles/perspectives they would each like to take. once both of them repsonded, we began opening statements for the debate.
this could also work for things like code review, writing, etc. get 3 reviews from three models on a somecode youre working on, or get 3 variations of a rewrite from three different models, etc etc.
pretty cool!