Plivo is a cloud telephony platform whose voice API includes bidirectional audio streaming, the piece that lets an AI voice agent sit on a live phone call. A element in the call's XML opens a WebSocket to your server with bidirectional="true"; the caller's audio arrives as base64-encoded media events, and your server returns playAudio messages to speak. A keepCallAlive attribute holds the call open while the agent thinks, and statusCallbackUrl reports stream lifecycle events.
The typical pipeline is speech-to-text, an LLM, and text-to-speech, with a stated end-to-end target under one second. Three content types are supported; μ-law at 8 kHz is the native telephony format and avoids transcoding.
| Content type | Codec | Rate |
|---|---|---|
| audio/x-mulaw;rate=8000 | μ-law (PCMU) | 8 kHz |
| audio/x-l16;rate=8000 | Linear PCM 16-bit | 8 kHz |
| audio/x-l16;rate=16000 | Linear PCM 16-bit | 16 kHz |
Plivo also sells a hosted AI Agents product: an agent studio where you describe a use case in plain language and a builder generates the conversational flow for voice, SMS, WhatsApp, or chat. You can pair Deepgram Nova-3 or your own STT, OpenAI, Gemini, or custom LLMs, and Cartesia, ElevenLabs, or custom TTS; attach a knowledge base synced from files, websites, and apps such as Notion; and hand off to a human mid-call over SIP REFER. Telephony covers 150+ countries with SIP trunking.
Features
- Bidirectional: WebSocket audio in both directions on a live call, with keepCallAlive and status callbacks
- Event model: start with a streamId, media payloads from the caller, and playAudio from your server
- Audio formats: μ-law 8 kHz, linear PCM 8 kHz, and linear PCM 16 kHz
- Latency target: under one second end to end for the STT, LLM, and TTS loop
- AI Agent Studio: describe a use case and get a generated flow across voice, SMS, WhatsApp, and chat
- Pluggable providers: Deepgram or custom STT, OpenAI, Gemini, or custom LLMs, and Cartesia, ElevenLabs, or custom TTS
- Knowledge base and integrations: sync files, websites, and apps; 30+ connectors including Salesforce, HubSpot, Shopify, and Zendesk
- Human handoff: SIP REFER transfer to a person by agent, team, or custom rule
- Turn detection: voice activity detection for interruption handling
- Telephony reach: 150+ countries and SIP trunking
