Agent Set-Up

Parameters

Customize voice, call behavior, AI model, recording, variable extraction, and background audio.

Parameters

In the Parameters tab, you control all technical settings of your agent — from voice settings to call limits to background sounds.

Voice Settings

Here you configure how your agent's voice sounds.

FieldDescription
SpeedSpeaking speed as a slider (e.g., 1.35). Default is approximately 1.0 — higher values = faster speech.
EmotionSelect the emotional tone of the voice from a dropdown (e.g., Neutral). Affects how friendly, factual, or dynamic the agent sounds.

Tip

Test different combinations of speed and emotion to find the right effect for your use case. For support agents, a calmer speed (0.9–1.1) with a neutral tone is recommended.

Call Settings

Control how long calls may last and when the agent hangs up.

FieldDescription
Max. Call Duration (Min.)Maximum call duration in minutes (e.g., 20). The call is automatically ended after this time.
Max. Wait Time (Sec.)How long the agent waits during silence before ending the call (e.g., 60 seconds).

Mind the Time Limits

Don't set the maximum call duration too low — complex queries need time. 20 minutes is a good default. The wait time should be at least 30 seconds so callers have time to respond.

AI Model

Influence the behavior of the underlying language model.

FieldDescription
TemperatureSlider from 0 to 1 (e.g., 0.50). Lower values = more predictable, factual responses. Higher values = more creative, varied responses.

Recommendation

For most use cases, a temperature between 0.3 and 0.6 is ideal. Higher values are only worthwhile for creative scenarios (e.g., role-playing or brainstorming).

Recording & Privacy

Define whether calls are recorded and how data is handled.

FieldDescription
Record AudioEnables recording of the call audio. The recording is then available in the call overview.
Automatic DeletionWhen enabled, recordings are automatically deleted after a defined period.

Privacy Notice

Inform callers about the recording — ideally right in the conversation opener. In many countries, active consent is required.

Variable Extraction

After the conversation, ScaleTalk automatically extracts structured data from the transcript. Here you define which variables should be captured.

Each variable consists of:

  • Name — The key name (e.g., Name, Category, Urgency)
  • Description — What the variable should contain, so the model extracts the correct value
ComplexityVariables
MinimalName, Concern, Phone, Category, Next_Steps
StandardName, Company, Concern, Category, Phone, Email, Urgency, Next_Steps
ExtendedName, Company, CustomerNumber, Concern, Category, Subcategory, Phone, Email, Urgency, Next_Steps, AdditionalInfo

Example Configuration

VariableDescription
NameFull name of the caller
CompanyCompany name of the caller, if mentioned
ConcernSummary of the concern in 1-2 sentences
CategoryClassification: Sales, Support, Accounting, Other
PhoneCallback number of the caller
UrgencyLow, Medium, or High
Next_StepsWhat should happen next

Tip: Restrict Values

Restrict variables like Category or Urgency to fixed values. Instead of "Category of the call," write: "Classification: Sales, Support, Accounting, Other." This gives you consistent, filterable values.

Using Extracted Variables

The extracted variables are used in several places:

  • In the email subject of follow-up emails
  • In the webhook payload sent to external systems
  • In the call overview for filtering and analysis

Background Audio

Make conversations more natural with ambient sounds and thinking sounds.

FieldOptionsDescription
Ambient SoundsOfficeSubtle office background noise. Creates the impression that the agent is sitting in a real office environment.
Thinking SoundsKeyboard typing, Keyboard typing shortTyping sounds that play while the agent prepares a response. Bridges wait times naturally — as if someone is looking something up.

More Options Coming Soon

Currently, only the Office option is available for ambient sounds. Additional options like cafe, reception, etc. will be added soon.

Tip

Ambient sounds and thinking sounds can make a big difference. A subtle office sound combined with keyboard typing makes the agent appear significantly more human.

On this page