Skip to main content
A Neuron voice agent is the AI-powered caller at the center of every campaign. You give it a script, a language, a voice, and a set of rules for how to handle what customers say — and it executes that conversation thousands of times simultaneously, adapting to each customer in real time. This page walks you through everything you need to configure a production-ready voice agent.

​
What Is a Voice Agent?

In VInfer, a voice agent is a configured AI caller that:
  • Follows your conversation script while adapting dynamically to customer responses
  • Handles objections, interruptions, and topic switches across multiple turns
  • Tags every call with a structured disposition and logs it to your CRM
  • Escalates to a live agent when your rules require it
Voice agents are reusable — you configure one agent and attach it to as many campaigns as you need. Any change to the agent (script update, language change, new fallback rule) takes effect on the next campaign that uses it.
Script IDs are created and managed in the VInfer dashboard under Neuron → Scripts. You must create a script before you can configure a voice agent that references it.

​
Agent Configuration Fields

​
agent_name
string
required
A human-readable label for your agent. Used in the dashboard and in API responses. Example: "EMI Reminder Bot".
​
language
string
required
The BCP-47 language code for the primary language this agent should speak and understand. Example: "hi-IN". See the Languages page for the full list of supported codes.
​
voice
string
required
The voice profile for the agent. Accepted values: "male" or "female". Regional accent variants are configured via the dashboard under Neuron → Voice Profiles.
​
script_id
string
required
The ID of the conversation script to use. Scripts are created under Neuron → Scripts in the dashboard. Example: "script_emi_reminder_v2".
​
fallback
object
Escalation configuration. Defines when and where Neuron transfers the call to a live agent. See Fallback to Human Agents below for details.
​
dispositions
array of strings
required
The list of outcome tags this agent should use to classify call results. Example: ["paid", "callback", "dispute", "not_reachable"]. See Dispositions below.

​
Example Agent Configuration

Here is a complete agent configuration for an EMI reminder workflow: 
{
  "agent_name": "EMI Reminder Bot",
  "language": "hi-IN",
  "voice": "female",
  "script_id": "script_emi_reminder_v2",
  "fallback": {
    "trigger": "escalation_requested",
    "transfer_to": "+918001234567"
  },
  "dispositions": ["paid", "callback", "dispute", "not_reachable"]
}
You can create this agent via the dashboard (Neuron → Voice Agents → New Agent) or by posting the configuration object to the API endpoint /v1/agents.

​
Configuring Your Conversation Script

Scripts define the structure of the conversation — what the agent says, how it branches based on customer responses, and what it captures. You build scripts in the Neuron → Scripts editor, which supports:
  • Opening statement — the agent’s first utterance after the customer picks up
  • Response branches — distinct paths for key customer responses (e.g., interested, not interested, callback request, dispute)
  • Objection handlers — scripted responses for common objections that keep the conversation on track
  • Capture fields — variables the agent collects during the call (e.g., preferred callback time, dispute reason)
  • Closing statement — how the agent wraps up based on the disposition it reaches
Keep your opening statement under 20 seconds. Customers are significantly more likely to stay engaged when the agent reaches the key ask within the first 30 seconds of a call.

​
Multi-Turn Dialogue Understanding

Neuron does not treat each customer utterance in isolation. It reasons over the full conversation history on every turn, which means it can:
  • Handle interruptions gracefully — if a customer cuts in mid-sentence, Neuron processes what they said and responds to it rather than repeating itself
  • Track topic switches — if a customer asks an off-script question (“wait, is this the same loan I applied for last month?”), Neuron handles the tangent and returns to the main flow
  • Manage compound objections — customers often give layered responses (“I’d pay but I’m traveling and also there’s a dispute on this account”) — Neuron unpacks these and addresses both components
This full-context reasoning is what separates Neuron from rule-based IVR systems that break the moment a customer deviates from the expected path.

​
Dispositions and Call Outcomes

Dispositions are the structured outcome tags Neuron assigns at the end of every call. You define the disposition set for each agent based on your workflow. Common disposition values include:

Connected

The call was answered and the agent completed at least one full exchange with the customer.

Not Reachable

The number was unreachable — no answer, switched off, or invalid number.

Callback Requested

The customer asked to be called back at a specific time. Neuron captures the preferred time if provided.

Interested

The customer expressed positive intent — agreed to pay, accepted the offer, or confirmed the appointment.

Not Interested

The customer declined and did not request further contact.

DNC

The customer explicitly requested to be removed from your calling list. VInfer automatically suppresses this number from future campaigns.
All dispositions are pushed to your CRM in real time after each call. You can map VInfer disposition values to your CRM’s own lead status fields under Settings → Integrations → CRM Mapping.

​
Fallback to Human Agents

Neuron is designed to know its limits. When a call crosses into territory that requires human judgment — a frustrated customer, a compliance-sensitive statement, or a specific escalation keyword — it transfers the call to a live agent immediately. You configure fallback behavior in the fallback object of your agent configuration:
​
fallback.trigger
string
required
The condition that triggers a transfer. Supported values:
  • "escalation_requested" — customer explicitly asks for a human
  • "sentiment_negative" — Neuron detects sustained customer frustration or distress
  • "keyword_match" — a keyword from your configured list appears in the conversation
  • "compliance_trigger" — a compliance-sensitive phrase is detected (e.g., legal threats, regulatory references)
​
fallback.transfer_to
string
required
The phone number or SIP URI to transfer the call to when the trigger fires. Must be in E.164 format for PSTN numbers.
​
fallback.context_handoff
boolean
When set to true, Neuron passes a structured summary of the conversation to the receiving agent’s screen before the transfer completes. Defaults to true if your CRM integration is active.
If you configure "compliance_trigger" as a fallback trigger, work with your legal and compliance teams to define the exact keyword and phrase list. VInfer’s default compliance keyword list is a starting point, not a substitute for your organization’s own compliance requirements.

​
Supervisor Dashboard

Your supervisors get a real-time view of every agent running across all active campaigns:
1

Open the Supervisor View

Navigate to Neuron → Supervisor Dashboard in the VInfer console. The dashboard refreshes every 5 seconds.
2

Monitor Active Calls

The Live Calls panel shows every call in progress — customer number (masked per your privacy settings), current conversation stage, duration, and language.
3

Track Dispositions in Real Time

The Disposition Board shows a running count of each outcome category as calls complete. Filter by campaign, agent, or language.
4

Review Agent Performance

The Agent Performance panel shows connect rate, average call duration, escalation rate, and disposition breakdown per voice agent — useful for identifying scripts that need refinement.
Supervisors can listen to live calls in real-time monitor mode or review recordings after the fact, subject to your organization’s call recording and consent configuration.

​
Next Steps