Skip to main content

How Vaani Works

Vaani connects your AI agents to phone calls. When someone calls your agent’s number — or when your agent calls out — the platform handles the entire conversation automatically.

Key Entities

Workspace

Your team’s organizational unit. All data — agents, calls, phone numbers — is contained within a workspace. You can belong to multiple workspaces.

Agent

An AI-powered virtual assistant configured with a system prompt, an LLM provider, voice settings, and custom tools. Each agent has a unique phone number for receiving calls.

Phone Number

A provisioned phone number from Twilio or Vonage, connected to your agent. Incoming calls to this number are automatically handled by the assigned agent.

Call Log

A record of every call — including transcript, recording, cost breakdown, AI-generated summary, and disposition (outcome).

The Call Lifecycle

Every voice call follows this lifecycle:
PhaseWhat Happens
ConnectThe call is routed to the assigned AI agent
GreetingThe agent delivers its configured first message
ConversationThe agent listens, understands, and responds in real-time. It can use custom tools, query the knowledge base, or transfer the call
EndEither party hangs up, or the agent ends the call
Post-CallA recording is saved, costs are calculated, and the conversation is analyzed for sentiment, summary, and disposition

Roles & Permissions

Vaani uses Role-Based Access Control (RBAC) within each workspace:
RoleWhat They Can Do
MemberView agents, call logs, reports, and analytics
DeveloperEverything a Member can do, plus create/edit agents, phone numbers, campaigns, and batch jobs
AdminEverything a Developer can do, plus manage workspace settings, invite/remove members, delete workspace
Each user can belong to multiple workspaces with different roles in each. Switch between workspaces using the workspace selector in the dashboard.

AI Providers

Vaani lets you mix and match AI providers for each agent:
ComponentPurposeAvailable Providers
LLMUnderstands conversation and generates responsesOpenAI GPT-4o, Groq
STTConverts caller’s speech to textDeepgram, Sarvam, Cartesia
TTSConverts agent’s text to natural speechOpenAI, ElevenLabs, Deepgram, Rime, Cartesia, Sarvam, Murf, Inworld
Choose GPT-4o for accuracy, Groq for speed. For TTS, ElevenLabs offers the most natural voices while Deepgram provides the lowest latency.

Dynamic Variables

Personalize conversations by injecting variables into agent prompts:
System Prompt:
"You are calling {{customer_name}} about their appointment on {{appointment_date}}."
Variables are filled in automatically from:
  • Outbound calls — passed when initiating the call
  • Batch calls — extracted from your uploaded CSV columns
  • Dynamic data — stored in the platform and matched by phone number

Key Terms

TermMeaning
DispositionThe outcome of a call (e.g., “completed”, “transferred”, “no-answer”)
Knowledge BaseDocuments uploaded to your agent so it can answer questions using your content
Batch JobA set of phone numbers to call automatically, uploaded via CSV
CampaignA simpler version of a batch job for quick outbound calling
Custom FunctionA webhook the agent can call during conversations (e.g., check inventory, book appointments)
Warm TransferThe agent briefs the human before connecting the caller
Cold TransferThe caller is connected directly to the human without briefing
See the full glossary for more terms.