Visual Builder Overview

The Visual Builder is where you design what your agent says and does. Instead of writing one long prompt and hoping, you build the conversation as a workflow of connected nodes that you can read, reorder and test.

Tap or click the screenshot to open it full size — the canvas has a lot of detail that's easier to read at full resolution, especially on a phone.

Why a workflow instead of one prompt

A single large prompt is hard to review, risky to change, and difficult to explain to the people who own the conversation. A workflow fixes all three:

• Reviewable — someone who does not write prompts can read the flow and tell you the third question is wrong.
• Changeable — you edit one node without disturbing the rest of the call.
• Testable — you can see which step a call went through when something goes wrong.

The building blocks

Each of these is its own guide:

• Business Context — what the agent knows, written once
• Start Call — how a conversation opens
• Conversation Nodes — the steps of the call
• End Call — how the call closes and what happens next
• Connections & Routing — how the agent takes different paths
• Extraction Variables — the structured data you get back
• Guardrails — the limits that apply across the whole call
• Human Handoff — when and how a call moves to a person

Test as you build

The builder is only half the loop. After each meaningful change, talk to the agent in your browser and listen to what actually happens.

→ Browser Test

A working order

1. Fill in business context first — everything else leans on it.
2. Add guardrails early, so you are never testing an unconstrained agent.
3. Build one straight path through the conversation, open to close.
4. Browser test it.
5. Add branches for the real answers you heard.
6. Define extraction variables for the fields you need.
7. Set the handoff rule and test that it fires.

Next

• Create Your First Voice Agent
• Browser Test
• Voice Agents Overview