Personas

Create synthetic users with distinct behavior, language, demographics, and voice conditions.

A persona defines how the synthetic user behaves during a NovaSynth run. Use personas to vary language, tone, interruption behavior, speaking speed, background conditions, and user context while keeping the scenario constant.

NovaSynth Personas page with Generate with AI and Create Persona actions plus a budget-conscious engineering lead showing goal, language, traits, edit, and delete controls — Use **Create Persona** for precise manual control or **Generate with AI** to propose several editable personas from the saved agent context.

Create a persona manually

Open NovaSynth → Personas and click Create Persona.

Enter a stable, descriptive Name. This is the only required field.

Add a Goal that states what the synthetic user should accomplish and a Description that provides any supporting context. Both fields are optional, but a clear goal makes the test easier to interpret.

Configure only the identity, behavior, language, voice, and context fields relevant to this test.

Click Create Persona, then confirm its languages and traits in the list.

Upper Create Persona dialog showing the required name, description, goal, tone preference, and gender fields — Start with the required name, then use the upper portion of the form to define only the user characteristics that should influence this test. Behavior sliders and background-noise controls follow further down the form.

The manual editor provides:

Group	Available fields
Identity	Goal, description, tone preference, gender, age, occupation, location, education level
Behavior	Patience from 0 to 1, speaking speed from 0.7 to 1.2, personality traits
Call conditions	Background noise type, level from 5 to 200, and mode, accent, and Max duration (s) from 60 to 600
Language and context	Primary languages, understood languages, knowledge-base topics, tags
Voice	ElevenLabs voice and TTS model
Custom context	Additional Config key-value attributes

Background noise mode chooses whether the noise is mixed under the caller's speech (Mix under speech) or plays for the whole call (Ambient for the whole call).

Max duration (s) caps calls for this persona. When the batch also sets Max call duration (s), a voice run stops at whichever cap is lower.

Scroll down to select a voice, TTS model, tags, and additional key-value context:

Lower Create Persona form showing understood languages, knowledge base, tags, and an ElevenLabs voice picker with language filters — Describe what the synthetic user understands, then filter the voice list by language before choosing a voice.

Pick a voice from the ElevenLabs list and filter it by language, or click Enter custom ID to paste any ElevenLabs voice ID, including cloned and generated voices.

Choose a TTS model

The TTS model picker checks each model against the persona's Primary language(s):

Automatic, the default, chooses a compatible model from the persona's spoken languages and shows the recommended model.
Catalog models list their supported languages and are flagged when they miss a spoken language, are unavailable in the voice runtime, or are deprecated.
Advanced: custom model accepts an exact provider:model identifier. NovaSynth cannot verify a custom model's language support.

Add the persona's spoken languages before choosing a model so the picker can check compatibility. A voice run fails before the call starts when the persona's TTS model does not support one of its spoken languages, when the model is unavailable in the voice runtime, or when Automatic finds no model for the languages. The run error names the model and the unsupported languages. The model catalog is also available from GET /api/v1/novasynth/voice-models, filtered with an optional language query parameter.

Use the same persona across several scenarios to determine whether failures are tied to a user's behavior rather than to the conversation flow.

Generate personas with AI

Generate with AI creates several personas asynchronously from an agent system prompt.

Save a detailed prompt and optional instructions under Agent Configuration.

Open Personas and click Generate with AI.

Review the Agent System Prompt, pre-filled from the saved configuration, and edit it for this generation job if needed.

Enter the Number of Personas as a whole number of 1 or more, add optional Additional Instructions, then click Generate.

Wait for the banner to move through pending or running to completed. Review every generated persona before using it in a run.

Generate Personas dialog pre-filled from the saved support-agent prompt, with persona count, additional testing instructions, and the Generate action — The saved system prompt and additional instructions are copied into the generation dialog. Review both, choose the number of drafts, and click **Generate** only when the requested variation is clear.

After submission, the dialog closes and the Personas page displays the active generation job:

NovaSynth Personas page with an AI is generating personas progress banner above a No personas yet empty state — The progress banner confirms that generation is asynchronous. This new project has no saved personas yet; review the generated rows after the banner reports completion.

AI-generated personas include a goal and behavior fields inferred from the prompt and your one-time instructions. The banner reports completed, failed, or cancelled generation and refreshes the list after success.

Generation proposes test records; it does not call the connected agent. Before running a generated persona, verify its goal, languages, speaking behavior, voice selection, background-noise settings, and custom context. Edit anything that is unsafe, unrealistic, duplicated, or outside the agent's intended audience.

Useful generation instructions describe the variation you want, for example:

Include one impatient customer, one first-time user, and one bilingual user. Cover both confident and uncertain communication styles.

Review and maintain personas

The list shows names, goals, languages, traits, and tags. Search it with Search persona..., click a row to open the full persona details, or use the pencil action to edit it. You can delete an unreferenced persona to remove it from future selections, but deletion is blocked after any synthetic run references that persona. This restriction preserves the source persona for existing run records. Create a replacement persona when future tests need different behavior.

Before a large batch:

remove duplicate personas;
make language and voice settings compatible;
keep demographic details relevant to the behavior being tested;
avoid real personal data;
use stable names so the run matrix remains readable.