Tavus

Tavus

Tavus is a San Francisco AI research lab, founded in 2020, that builds real-time conversational video agents it calls PALs. Its Conversational Video Interface API combines speech recognition, turn-taking, perception, an LLM, TTS and face rendering in one pipeline.

Key Features

• Conversational Video Interface: A developer API for real-time face-to-face AI conversations with bring-your-own LLM, knowledge base retrieval, memory, guardrails and function calling.

• Sparrow-2 turn-taking: An audio-native model that handles pauses, interruptions, backchannels and background noise to decide when the agent should listen, wait or speak.

• Raven-1 perception: Reads tone, prosody, facial expression and gaze from voice, video and screen inputs to inform responses.

• Phoenix-4.5 rendering: Real-time lip-synced faces built from a single image or two minutes of video; custom replica training includes a custom voice model.

• Multiple channels: PALs run over chat, voice and video, with support for 30+ languages and a no-code PAL Maker builder.

Use Cases

• AI SDRs: Qualified built its AI SDR, Piper, on Tavus to hold sales conversations with buyers.

• Interview practice: Final Round AI runs mock job interviews with lifelike interviewers on Tavus CVI.

• Training and intake: Role-play partners for sales and pharma rep training, and patient intake agents in healthcare.

Getting Started

Developers can start on the free Basic plan with 25 minutes of conversational video, or sign up for PALs: https://www.tavus.io