Skip to main content
Personas define who calls your agent during a voice simulation. A persona bundles a synthetic voice with speech behavior—pace, clarity, timing, disfluencies—and the knowledge the caller can reference. Build your persona library once, then reuse personas across scenarios and evaluations.

Why Personas Matter

Real callers vary. Personas let you test against that variation systematically:

Personas Dashboard

Navigate to Library → Personas from the left navigation panel. The dashboard shows your personas as cards in a searchable grid: Personas dashboard grid
  • Search: Find personas by name, description, or tag
  • Create: Open the persona builder
  • Edit / Delete: Manage existing personas from each card

Creating a Persona

1

Pick an Avatar

Persona Avatar PickerChoose an avatar to make the persona easy to recognize across the dashboard.
2

Set Identity

Choose a behaviour preset—neutral, friendly, frustrated, confused, or custom—to shape how the persona reacts when the conversation gets difficult.
3

Configure Speech Traits

Tune how the persona actually talks:
Vague, rambling callers with disfluencies are the best stress test for STT accuracy. Save them for later-stage testing once basic flows pass.
4

Assign a Voice

Persona TTS SelectorChoose the TTS provider, model, and voice that will speak this persona’s lines. Available options depend on the provider you select—some models support accents and personality settings, others do not.
5

Add Knowledge

Provide the facts this persona knows (such as order numbers, dates, or account details) as key-value pairs. During a simulation, the persona can only reference this information—if your agent asks about something the persona does not know, it will say so.
6

Save

Click Create. The persona becomes selectable in the voice evaluation creation flow.

Using Personas

  • In evaluations: Each voice scenario requires exactly one persona; mix personas across scenarios in the same evaluation to cover different caller types
  • Across runs: Because a persona is fixed, changes in results between runs reflect your agent—not the caller

Best Practices

Building a Balanced Library

Cover the caller types your product actually serves:

Managing Personas

  • Name by behavior: “Fast rambling complainer” is more useful than “Test persona 3”
  • Keep knowledge minimal: Only give personas the facts the scenario needs
  • Reuse before you create: Check the library before adding a near-duplicate persona
Last modified on August 28, 2026