Persona

What a persona is

A persona is the simulated caller on the other end of the line. It combines caller behavior instructions with a concrete voice identity, so the agent under test hears a realistic and — crucially — repeatable caller on every run.

A persona controls speaking style and intent, response timing and turn behavior, and the LLM settings used to drive the simulation. Pair it with a scenario to say why this caller is calling.

Why the voice mapping is pinned

Every persona is bound to a specific provider voice rather than "any female voice":

tts_provider
tts_voice_id
tts_voice_name
gender

Audio characteristics change acoustic and voice-quality scores. Pinning the voice means that when a score moves between runs, the cause is the change you made — a prompt revision, a model swap, an evaluator update — and not a different voice on the caller side. You can use built-in provider voices or custom catalog voices from your organization.

Configure a persona

What you configure

Persona profile and description
Voice provider and voice mapping
Simulation controls (temperature, delay, max turns, interruption behavior)

Recommended flow

Create the persona profile.
Select the provider voice.
Tune behavior controls for realism.
Reuse the persona across evaluation suites.

Reuse matters: a small, stable library of personas gives you a consistent baseline to compare against over months, instead of a new caller definition for every test.

See also: How It Works for how personas combine with scenarios into runs.