Tavus Conversational Video Interface (CVI)

What it does

Tavus CVI is a developer platform for building conversational AI agents, called PALs, that run over chat, voice, or real-time video. Developers supply a system prompt, a face and voice trained from a short video or a single image, and optionally an LLM. The PAL takes in live audio and video and returns synchronized speech and rendered video of a human-like face.

Developers typically create a PAL with a single API call and embed the conversation in their own app using Tavus's prebuilt React components or LiveKit and Pipecat media pipelines. Notable capabilities include turn-taking detection, emotion control adjustable mid-conversation, a knowledge base for retrieval, memory across sessions, function calling and MCP tools, and optional browser control or on-screen widgets during calls.

This suits developers and product teams building customer support, recruiting, intake, sales, or interview experiences that need a visual agent. Getting started requires a Tavus developer account and an API key. The product is an API for builders rather than a finished end-user app, and it is aimed at teams integrating conversational video into their own products.

What it can do

• Create PALs (face-to-face AI agents) with a single API call using a system prompt, face, and voice.
• Real-time turn-taking, perception of voice and video, and emotion control during conversations.
• Knowledge base retrieval, memory across sessions, and internet search during calls.
• Function calling and MCP tools that let a PAL take actions mid-conversation.
• Chat, voice, and video deployment modes with a choice of LLM or Tavus-managed pipeline.
• Prebuilt React components for video, transcript, and controls.

Works with

• LiveKit
• Pipecat
• Google Meet
• Zoom

Example workflows

Customer support PAL

Input: a support system prompt and a trained face. Steps: create the PAL via the v2/pals endpoint with raven-1 perception and sparrow-2 turn detection, then embed it in a web page with the prebuilt video component. Output: a face-to-face support agent that listens, responds with empathy, and can call tools mid-call.

Recruiter screening interview

Input: interview questions in the system prompt and a candidate-facing face. Steps: deploy the PAL into a meeting through the Google Meet or Zoom integration and enable a knowledge base of role details. Output: a screening conversation that can be recorded and transcribed.

Before you choose

• Choose the Developer Account type for API-based builds; PALs Accounts are aimed at individual personal use.
• Vendor performance and latency figures are reported by Tavus; validate them against your own use case.

Listed under

• Coding
• AI sdr
• AI video agent
• Conversational video
• Customer support
• Developers
• Healthcare agents

Get started

Create a Developer Account, then generate an API key to call the PAL endpoints.

1. Create a Tavus Developer Account at platform.tavus.io and generate an API key.
2. Create a PAL by sending a POST request to the v2/pals endpoint with a system prompt, default face ID, and layer settings.
3. Embed the conversation in your app using the prebuilt React components or connect to LiveKit or Pipecat.

Current rates, billing basis, and any starting allowance are listed on the live pricing page and should be confirmed there before committing.

We couldn’t find published prices for Tavus Conversational Video Interface (CVI). See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Build Face-to-Face AI Video Agents | Tavus CVI ↗www.tavus.io
• Plans and Pricing - Tavus ↗www.tavus.io

Company website

tavus.io: product, pricing and policy pages

Hacker News

Launch posts and discussion

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.