Design AI Voice Conversation Flows With Visual Builder

VoiceInfra’s workflow builder transforms how you design AI voice conversations. Instead of describing a single behavior in a system prompt, you map out your full conversation logic on a visual canvas — connecting nodes that handle dialogue, execute API calls, search your knowledge base, and transfer callers to humans. Use the workflow builder when your calls follow structured paths, branch based on caller responses, or need to trigger real-time actions at specific points in the conversation. For simpler single-purpose agents, the no-code builder is the faster starting point.

Node Types

Every workflow is composed of six node types. Each node handles a specific function, and you connect them with edges to define the conversation path.

Node	Role	Key behavior
Start	Entry point	One per workflow; sets the opening message and context
Conversation	AI dialogue	Handles natural-language turns with the caller; extracts variables
Tool	Action execution	Calls a REST API, runs a custom function, or invokes an MCP server action
Transfer	Human handoff	Passes the call to a human agent via cold, warm, or attended warm transfer
Knowledge Base	Document search	Queries attached knowledge sources and returns relevant content to the caller
End	Termination	Closes the call gracefully; triggers post-call actions like email summaries

Routing Types

Nodes connect through edges, and each edge carries a transition condition that determines when the workflow moves to the next node.

AI Transitions — write the condition in plain language: “When the caller confirms their appointment” or “If the caller asks about pricing.” The AI evaluates the conversation in real time and follows the matching path.
Equation Transitions — define variable-based logic: caller_intent == "cancel" or account_balance < 0. These run as deterministic checks against variables extracted during the conversation.
Combined — apply both types on the same node. AI handles the nuanced language understanding; equations handle the hard rules. The first matching condition wins.

Build a Workflow

Create or Pick a Template

Start from a blank canvas or choose from 12+ pre-built workflow templates. Templates cover the most common scenarios and include pre-configured nodes, routing logic, and sample prompts ready to customize.

Customer support escalation
Lead qualification and routing
Appointment booking and confirmation
Technical troubleshooting triage
After-hours call handling
And more

Name your workflow, add a short description, and you’re on the canvas within seconds.

Design on the Canvas

Drag nodes from the panel onto the canvas and connect them with edges. Click any node to open its configuration panel on the right — set the prompt, attach tools, define transition conditions, and configure extraction fields.

Auto-layout — rearrange messy flows with one click
Zoom and pan — navigate large workflows without losing context
Minimap — see the full workflow structure at a glance
Fullscreen mode — maximize your editing space for complex flows
Validation — the builder flags broken edges and missing configurations before you can save

Configure Model, Voice & Settings

Open the workflow settings panel to select the AI provider and voice for this workflow, then tune the advanced controls that apply across all nodes.

AI provider — Anthropic, OpenAI, Gemini, Groq, or Mistral
Voice — select from 40+ premium voices; per-node voice overrides are also supported
Interruption handling — decide whether the agent stops speaking when the caller talks
Call limits — set maximum call duration
Timezone — used for date/time variable resolution
Post-call emails — send AI-generated call summaries to a configured address

Test With the Debug Panel

Trigger a test call directly from the builder. The debug panel shows you exactly what happens at each step in real time.

Which node is currently active
Variables extracted from caller responses
Tool execution requests and API responses
Transition conditions evaluated and which path was taken
Errors or fallback triggers

Fix issues on the spot and re-test without leaving the canvas.

Deploy

Save the workflow and assign it to one or more agents. Changes apply to new calls immediately — no downtime, no redeployment process. Calls that are already in progress continue on the previous version of the workflow until they end naturally.

AI Builder: If you’d rather describe your workflow in plain English than build it node by node, use the AI Builder. Type a description of what the workflow should do — for example, “Qualify inbound leads by asking for company size, budget, and timeline, then transfer to sales if all three criteria are met” — and VoiceInfra auto-generates the full node structure, routing conditions, and prompts for you. You can refine the result on the canvas afterward.

Reusability

Build a workflow once and assign it to as many agents as you need. Every agent that uses the same workflow always runs the latest saved version — update the workflow once and all agents pick up the change automatically. You can also export any workflow as a JSON file to back it up, share it with your team, or import it into another VoiceInfra account.

Example Workflow JSON Export

{
  "name": "Lead Qualification Flow",
  "version": "1.0",
  "nodes": [
    {
      "id": "start-1",
      "type": "start",
      "firstMessage": "Hi, thanks for calling. I have a few quick questions to connect you with the right person.",
      "next": "conv-1"
    },
    {
      "id": "conv-1",
      "type": "conversation",
      "prompt": "Ask the caller for their company size, monthly budget, and timeline to purchase.",
      "extractVariables": ["company_size", "budget", "timeline"],
      "transitions": [
        {
          "type": "equation",
          "condition": "budget >= 1000 && timeline <= 90",
          "next": "transfer-1"
        },
        {
          "type": "ai",
          "condition": "Caller is not ready to buy or has no budget",
          "next": "end-1"
        }
      ]
    },
    {
      "id": "transfer-1",
      "type": "transfer",
      "mode": "warm",
      "destination": "+15551234567",
      "next": "end-1"
    },
    {
      "id": "end-1",
      "type": "end",
      "closingMessage": "Thanks for your time. Have a great day!"
    }
  ]
}

Frequently Asked Questions

Can I use external APIs inside a workflow?

Yes. Add a Tool node at any point in your workflow and configure it with your API endpoint, HTTP method, authentication headers, and request body. The agent calls the API in real time during the conversation and can use the response to decide what to say next or which path to follow. You can add custom headers for API key authentication and reference extracted caller variables in the request body.

How do I handle unexpected caller input?

Each Conversation node has its own AI model that handles natural language freely within the scope of that node’s prompt. If a caller goes off-script — asks an unexpected question, gives an unusual answer, or changes the subject — the AI engages naturally, answers if it can, and then guides the conversation back to the intended path. The workflow’s overall structure remains intact even when individual turns are unpredictable.

Are changes live immediately?

Yes for new calls. The moment you save a workflow, all new calls routed to agents using that workflow run the updated version. Calls already in progress at the time of the save continue on the previous version and complete normally. There is no downtime and no need to reassign agents after an update.

​Node Types

​Routing Types

​Build a Workflow

​Reusability

​Example Workflow JSON Export

​Frequently Asked Questions

Node Types

Routing Types

Build a Workflow

Reusability

Example Workflow JSON Export

Frequently Asked Questions