This section provides a high-level overview of how to create and configure a voice agent on Hooman Labs to handle phone calls. Agents can be set up to follow a simple conversational prompt or a more advanced flow with multiple logic-based steps.

Creating a new agent

To get started:

1. Navigate to Agents → Create Agent.

2. Enter the Agent Name.

3. Select the Language for interaction.

4. Choose a Use Case that the agent will handle (e.g., appointment booking, lead qualification, collections).

5. Select the Prompt Type — either Single Prompt or Flow. Note: This choice is fixed and cannot be changed later.

6. Click Create to begin configuring the agent.

You can also set the agent's timezone by going to Agent → Edit → Metadata.

Prompt type

a. Single Prompt (Simple)

A minimal setup where the agent uses one core instruction or prompt to handle the entire conversation. Ideal for linear conversations.

b. Multi-Prompt (Flow Type)

A flexible flow system where each step (node) can have its own prompt, tool, or transition logic. Ideal for branching conversations with multiple stages.

Flow-based agents support multiple node types:

• Prompt Node: Standard language model instruction.

• Fixed Node: Sends fixed messages to the user.

• Condition Node: Enables branching based on previous values or inputs.

• API Node: Makes an external API call.

• Reasoning Node: Executes reasoning steps without directly speaking to the user.

Configuration

Once an agent is created, configure it via the following sections:

3.1  Think Settings

Define how your agent thinks and responds:

• LLM model selection

• Prompt or Flow: Configure based on selected mode.

• Tools: Enable tools like Calculator, Memory, API connectors.

• Library: Attach reference documents the agent can use (for RAG).

3.2  Transcription Settings

Control how speech is converted to text:

• STT model selection

• Interruption and endpoint logic

3.3  Speech Settings

Manage how the agent speaks:

• TTS model and voice selection (male/female, accent)

• Speed and pitch tuning

3.4  Call End

Configure how and when a call should end:

• Define end prompt and message

3.5  Call Transfer

If human intervention is needed:

• Set rules for when and how to transfer to a human agent

3.6  Analysis

Enable post-call analytics:

• Summary

• Outcome

• Information extraction

3.7  Pre-Call API

Fetch customer-specific context before a call begins:

• Pull from CRM or internal system

• Populate variables for use in agent prompts, APIs, or logging

3.8  Post-Call API

Send call summary, outcomes, recordings, or structured data to your backend:

• Push to CRM, trigger workflows

• Fully customizable payload with ${variable} syntax

3.9  Advanced Settings

Control operational behaviors:

• Idle Settings: Define inactivity timeout behavior

• Voicemail Detection: Enable/disable auto voicemail tagging

• Max Call Duration: Set cap on total call length

Publish

Once all configurations are set, you have two save options:

• Save as Draft: Recommended for testing and iteration

• Publish: Applies live changes to the agent