Launch Week 3: Five days of launches
Confident AI
Docs
Pricing
Products
LLM Evaluation
LLM Evaluation

Benchmark LLM systems with research-backed metrics.

LLM Observability
LLM Observability

Trace, monitor, and alert on production LLM systems.

AI Red Teaming
AI Red Teaming

Stress-test LLM apps against adversarial attacks.

AI Governance
AI Governance

Enforce AI standards and controls across teams.

Book a DemoSign Up
Book a DemoSign Up

Personas

Reusable simulated users for multi-turn goldens, including how they talk on voice calls.

Why Use Personas?

A persona is a saved description of a user — how they behave, what they know, and, on voice calls, how they talk — that multi-turn goldens can simulate the user as. When a golden has a persona, the persona's characteristics take precedence over the golden's User Description. That lets you:

  • Run the same task with different users. Keep one scenario and expected outcome, and pair it with several personas — a frustrated repeat caller, a first-time buyer, someone who prefers simple English — to see whether your agent copes with each, without rewriting the scenario's user description every time.
  • Write a user once and reuse it everywhere. Pick the same persona on goldens across datasets; editing it updates every golden that uses it.
  • Give the simulator facts to work with. A persona's Metadata (an account, an order, a resume) is available to the simulated user and to your AI connection payload.
  • Make voice evaluations realistic. Set how often the persona talks over your voice agent and what happens when both sides talk, and play background noise under its speech.
Personas available to multi-turn datasets

Create a persona

Open Personas from the project sidebar (under More in Build), then click Add persona.

Add a persona
  • Name — a recognizable name, such as Frustrated support caller.
  • Characteristics — how this persona behaves during a multi-turn conversation: their mood, how much they say, what they care about. Be specific; this is what the simulated user acts on.
  • Metadata — optional structured facts the persona can draw on, as JSON (for example {"accountId": "acct_48213", "orderId": "ord_99127"}).
  • Interruptions — for voice calls: the Frequency this persona talks over the agent (Off, Rare, Normal, or Frequent), and When both sides talk, whether it Backs off, is Adaptive, or Insists.
  • Background noise — for voice calls: upload a .wav or .mp3 file to play underneath the persona's speech, and set its Volume.

Click Create. Each persona's row shows its characteristics, interruptions, metadata keys, and background noise; use the pencil to edit it or the trash icon to delete it.

Use a persona in a golden

Open a golden in a multi-turn dataset and pick a saved persona under Persona. When the conversation is simulated, the persona's characteristics replace the golden's User Description, and its metadata and voice settings come with it.

To run one scenario with several personas, give each persona its own copy of the golden: the scenario and expected outcome stay the same, and only the user changes.

In an AI connection's payload, reference the persona with conversationalGolden.persona (or .name, .characteristics, and .metadata), and a single metadata value with conversationalGolden.persona.metadata.<key>.

Next Steps

Multi-Turn Evals

Simulate conversations from multi-turn goldens and evaluate them.

Voice

Call your voice agent and evaluate it with simulated conversations.

Scaling beyond prototype?For teams evaluating Confident AI in productionTalk to us

Last updated on

Built byConfident AI