The Conversational Video Interface (CVI) is an end-to-end pipeline for creating real-time multimodal video conversations with a digital twin that can see, hear, and respond similarly to how a human would. Developers can deploy video AI agents/digital twins in minutes using CVI.

CVI is the world’s fastest interface of its kind, allowing you to put a human face and conversational ability to your AI agent or personality. With CVI, you can achieve utterance-to-utterance latency of under ~900ms, which is the full roundtrip time for a participant to say something and for the replica to speak back.

CVI provides a complete pipeline to have a conversation while also allowing you to customize and plug in your existing components where necessary.

​
Key Features

Face-to-face interactions

The first interface that speaks our language. CVI is multimodal and understands and uses facial expressions, body language, and has natural conversational awareness including interrupts and turn-taking.

World’s lowest latency

The world’s fastest interface of its kind, with less than 600 millisecond latency utterance-to-utterance.

End-to-end solution

CVI provides a turn-key solution, delivering all the components to easily deploy AI video agents without having to worry about WebRTC, ASR, or anything else.

Focused on naturalness

Easily create high-quality AI replicas of you or your customers, powered by our state-of-the-art replica model, Phoenix-2.

​
What does a conversation with CVI look like?

​
Here’s a sample:

​
Try it out!

You can try chatting with Carter on our website to get a taste of what a conversation with CVI looks like.

Try Out CVI Now!

Note that Carter can see and hear you.

​
What components does CVI provide, and what can I customize?

CVI provides a full pipeline allowing you to easily create video conversations. You can immediately jump into a real-time conversation with the generated Daily meeting URL. CVI provides the following layers:

  • WebRTC/video conferencing (using Daily)
  • Vision
  • Speech recognition (ASR), with interrupts
  • Optimized, conversational LLM
  • Text-to-speech (TTS)
  • Replica video output

You can choose to customize or bring your own layers as well. For example, you can:

  • Use OpenAI real-time API or other voice-to-voice models and only use Tavus to drive the replica video.
  • Bring your own LLM/conversation logic or enable function calling for Tavus-optimized LLMs.
  • Customize the TTS or ASR engine.
  • Use text parrot mode to directly drive a replica video.
  • Directly access the video streams and create a custom UI. See Custom UI / Getting Raw Streams.

Learn more about the layers and different modes in CVI Modes and Layers.

​
Key Concepts

​
What is a conversation?

A conversation is a single ‘session’ or ‘call’ with a digital twin using CVI. When you create a conversation, you receive a Daily meeting URL. This URL provides a full video conferencing solution, allowing you to avoid managing WebRTC or websockets. Navigating to this URL lets you directly join a prebuilt meeting room UI to chat with your digital twin.

Learn more about creating and customizing conversations.

​
What are personas?

Personas are the ‘character’ or ‘AI agent personality’ and contain all the settings and configuration for that character or agent. For example, you can create a persona for ‘Tim the Sales Agent’ or ‘Rob the Interviewer’. Personas let you customize CVI’s layers and prompt the LLM with personality and context.

Learn more about creating a persona.

​
What are replicas?

A replica is a talking-head/avatar of a human containing a voice and face clone, used as the video output layer for CVI. You can use stock replicas from Tavus or create your own with a few minutes of training data. A replica is key for video generation and CVI.

Learn how to create a great replica.

​
What is a digital twin?

A digital twin is an AI-powered digital version of a human, which looks and sounds like a person and can see and respond similarly to a human.

​
Getting Started

​
No Code

You can easily try out CVI using the Tavus dashboard. Note that not all settings and modes are available via the dashboard.

​
API Quick Start

Check out the Quick Start Guide to learn how to use the APIs to create a persona and conversation. Be sure to grab an API key first!

Visit platform.tavus.io for more information.

ScriptingQuick Start