Fenix Deorum

Teams shipping customer-facing assistants give them a persona through a system prompt. Persona Audit checks what else that persona changes.

You supply a persona prompt and choose a model. The audit runs a fixed set of prompts twice, once with the model's default behavior and once under your persona. An AI judge scores every response on each trait. The result is a trait-by-trait side-effect matrix, and every cell links to the conversations behind it.

What it can't tell you: a persona prompt is not the same as changing the model's internals; the judge adds noise of its own; the prompt set is small and fixed; and results hold for one model on one date.