Manual Setup

Starting from scratch gives you complete control. You’ll configure every detail yourself — model, voice, prompt, and behavior.

Step 1: Click Start From Scratch

From your dashboard, click the green Create Agent button in the top right, then select Start from Scratch.

Start From Scratch button in Create Agent modal — Click Start From Scratch in the Create Agent modal

Step 2: Select Single Prompt

Choose Single Prompt as your agent type.

Select Single Prompt agent type — Select Single Prompt as your agent type

Step 3: The Editor Opens

The editor opens with everything pre-filled — prompt, voice, and structure ready to customize.

Step 4: Configure Your Agent (Left Panel)

Before writing your prompt, set the basics in the left panel:

Field	What to choose
Agent Type	Single Prompt (already selected)
Call Direction	Inbound if customers call in, Outbound if the agent makes calls
Emotive Model	Toggle on for more expressive voice (Beta), or leave off
Voice	Pick a voice from the library — use the preview to listen
Knowledge Base	Optionally attach an existing KB so the agent can use your docs/FAQs

Step 5: Write Your Prompt

The right panel is your prompt editor. This is the heart of your agent — it tells the AI exactly how to behave on every call.

Write clear instructions covering:

Role & Objective — Who is this agent and what’s their goal?
Conversational Flow — What steps should the agent follow?
Dos, Don’ts & Fallbacks — How should the agent behave in tricky situations?
End Conditions — When should the call end?

First time? Start simple. Write a few sentences about who the agent is and what it should do. You can refine everything else as you go.

Step 6: Test Your Agent

Click Test Agent in the top-right to start a test call.

You can test your agent in three ways:

Web Call — talk to your agent through your browser microphone
Telephony Call — enter a phone number and get a call from your agent
Chat — text-based conversation with your agent

Test your agent via Web Call, Telephony, or Chat

Talk through a few scenarios — ask a normal question, ask something unexpected, and interrupt mid-response. Listen for clarity and that the agent follows your guidelines.