Voice Agents Overview
Two-way phone calls answered by your AI agent
Voice Agents turn your sender's AI agent into a phone agent. It answers inbound calls to your number and places outbound calls, holding a real conversation over the phone — not a recorded message.
The same agent you already use for SMS and WhatsApp powers the call: its provider, model, system prompt, tools, and knowledge bases all carry over. You enable voice on the agent, and Zavu handles the call.
How It Works
Each call runs through Zavu's managed voice pipeline:
- Speech recognition transcribes the caller in real time.
- Your agent processes the transcript with its configured model, tools, and knowledge bases — exactly as it does for text channels.
- Speech synthesis speaks the agent's reply back to the caller.
The pipeline supports barge-in: the caller can interrupt while the agent is speaking, and the agent stops to listen. The agent can also call your tools mid-conversation, transfer the call to a human, or end it.
You never manage telephony, audio streaming, or transcription. You configure the agent and call the API.
Inbound and Outbound
| Direction | How it starts |
|---|---|
| Inbound | A caller dials a phone number assigned to a sender whose agent has voice enabled. Zavu answers automatically. |
| Outbound | You call POST /v1/calls. Zavu dials the recipient and the agent takes over once answered. |
Both directions use the same agent configuration and produce the same call record and transcript.
Pricing
Voice calls are billed per minute of connected time plus telephony, deducted from your prepaid balance. A short-duration estimate is reserved when a call is placed, and you are charged for the actual duration when it ends. A hard duration limit (maxCallDurationMinutes, default 15) caps every call.
Calls and Transcripts
Every call is recorded as a call resource with its status, duration, cost, and a turn-by-turn transcript. Retrieve a single call to read its full transcript, or list calls to review recent activity.
Subscribe to the call.* webhook events to react to calls in real time — for example, to log completed conversations or trigger follow-ups. See Webhooks.