Twilio Conversation Relay

Human-friendly voice AI that keeps customers from shouting "live agent!"

Easily integrate voice AI into your stack for smooth, personalized customer conversations — no complicated infrastructure or awkward AI moments.

Smiling woman speaking on the phone with a virtual agent interface overlay displayed.
Smiling woman speaking on the phone with a virtual agent interface overlay displayed.

Test out Conversation Relay's voice AI

How Twilio Conversation Relay works

Diagram showing integration of Twilio Voice with a ConversationRelay API connecting to an app with TTS and STT components.
Diagram showing integration of Twilio Voice with a ConversationRelay API connecting to an app with TTS and STT components.

Our scalable solution combines fast speech-to-text (STT) and text-to-speech (TTS) capabilities with the AI of your choice, seamlessly orchestrated through a WebSocket API.

  • Avoid long pauses or waiting for voice AI to finish

    Cure awkward digital pauses with Deepgram Flux, automatic language detection, and SSML tags. Low latency keeps the conversation flowing for far less robotic conversations.

  • Deliver natural voice, pacing, and intonation

    Provide customer interactions that sound like a real person, with the option to seamlessly transfer to a live agent for complex issues.

  • Add context for personalized support

    Enable smooth input/output with your large language model (LLM) so your AI agent can recognize customers and recall interactions.

Built for live conversation

Optimized for latency to keep dialog flowing at a natural pace.

<0.5 S

median latency

<0.725 S

at the 95th percentile*

Build AI support that understands your customers

Create customer experiences that are engaging, friendly, and always on point.

Deliver effortless self-service support

Enable context-aware, intelligent virtual agents that handle inquiries
efficiently—and know exactly when to bring in a human.

  • Handle routine inquiries while keeping customers engaged and frustration-free. 

  • Escalate complex or sensitive issues to live agents when needed.

  • Orchestrate customer data to provide personalized, context-rich interactions at scale.

Flowchart showing an incoming call, user data collection, virtual agent interaction, and sentiment analysis.
Flowchart showing an incoming call, user data collection, virtual agent interaction, and sentiment analysis.

Conversation Relay features

Streamline complexity while enabling human-like interactions at scale.

A dashboard showing a virtual agent and sentiment analysis results indicating positive sentiment.
A dashboard showing a virtual agent and sentiment analysis results indicating positive sentiment.

New

Turn every interaction into a continuous, contextual conversation

Conversation Relay is part of the Conversations layer of the Twilio platform. You can turn fragmented customer interactions into a continuous conversation that moves across channels, AI agents and human agents and time without losing any context. In fact, it gets smarter with every interaction through a persistent customer memory.

  • LLM integration that works for you

    Get the flexibility to bring your own LLM so you can control your UX, manage costs, and adopt new tech as it's released.

  • Speech recognition STT

    Convert spoken words into text in real time to supply your LLM with accurate transcription for responsive conversations.

  • Natural human-sounding TTS

    Get pronunciation, intonation, and rhythm right, or bring your own Text-to-Speech capabilities for a custom integration.

  • Interruption handling

    Use adjustable interruption sensitivity to fine-tune exactly how the agent reacts in noisy environments.

  • Global connectivity

    Access flexible, secure connectivity that includes number provisioning porting compliance.

  • Low-latency infrastructure

    Minimize latency to improve the quality of voice AI interactions and ensure a better customer experience.

  • Scale securely in highly regulated industries

    Build PCI-compliant workflows and HIPAA-eligible architectures to deploy compliant solutions faster.

Start bringing your voice AI agent to life

Explore our comprehensive APIs, documentation, and the Conversation Relay Studio Widget, which allows teams to deploy secure AI voice flows using drag-and-drop tools.

<?xml version="1.0" encoding="UTF-8"?>

<Response>
  <Connect action="https://myhttpserver.com/connect_action">
    <ConversationRelay url="wss://mywebsocketserver.com/websocket" welcomeGreeting="Hi! Ask me anything!" />
  </Connect>
</Response>

Need help setting up Conversation Relay?

Work with one of our trusted partners to set up your voice AI solution and start delivering amazing engagement. View partners

Your AI powers the conversations. We handle the voice.

Deploy secure voice AI flows using our intuitive drag-and-drop Studio Widget or comprehensive APIs, so you can focus on designing the smart, meaningful interactions your virtual agents need to deliver.

A smiling man holding a phone to his ear, wearing a dark jacket and green shirt with a red background.
A smiling man holding a phone to his ear, wearing a dark jacket and green shirt with a red background.

Conversation Relay FAQs

Twilio Conversation Relay integrates voice AI into your stack for natural, human-sounding phone conversations. It combines fast speech-to-text and text-to-speech with the LLM of your choice, orchestrated through a WebSocket API. This lets you build smooth, personalized voice agents without wiring up complex infrastructure or living with awkward AI pauses.

In a voice conversation, the delays are what make AI agents feel robotic. Twilio Conversation Relay is built to sound natural with median latency under 0.5 seconds and under 0.725 seconds at the 95th percentile, per internal benchmarks. Deepgram Flux, automatic language detection, and SSML tags cut the awkward pauses, so dialog flows at a human pace.

Twilio Conversation Relay uses Deepgram Flux for low-latency turn detection and adjustable interruption sensitivity, so the agent knows when to listen and when to respond (even in noisy environments). Callers can barge in and be understood mid-sentence, so the conversation feels like talking to a person.

Yes. Twilio Conversation Relay is built for bring-your-own-LLM, so you control the UX, the costs, and can adopt new models as they ship. You get natural human-sounding TTS and real-time STT out of the box. Or, you can bring your own text-to-speech for a fully custom voice.

Twilio Conversation Relay supports HIPAA-eligible architectures and PCI-compliant workflows to help you deploy voice AI in regulated industries like healthcare and finance. 

Twilio Conversation Relay is built for fast deployment. Use the drag-and-drop Studio Widget to build secure voice AI flows without heavy code, or go deeper with the WebSocket API and docs. You bring the AI and design the conversation, and Twilio handles the voice infrastructure underneath.

Twilio Conversation Relay supports multilingual voice AI with automatic language detection that works within a live call. Using multi mode, Conversation Relay detects the language a caller speaks and adapts in real time, passing the detected language to your app and responding in kind.

*Based on internal benchmarks with Conversation Relay (p50 491 ms, p95 713 ms) using different models. Results may vary.

Let's find the right products for you

Generating your product picks...

See your product picks