Tavus capabilities

Front-end options for dropping a live conversational video agent into a web app.

An API-first framework for building real-time, face-to-face conversational video agents: a lifelike digital human that sees the user, listens, and responds in a live two-way video call.

Lets an AI persona remember context about a user across separate conversations and devices, so interactions feel continuous and personal.

Ground a conversational persona in your own documents so the AI human answers from proprietary, domain-specific content via retrieval-augmented generation.

The rendering model that synthesizes a photorealistic digital human face in real time, including lip-sync, micro-expressions, and natural listening behavior.

Tavus's developer (API) pricing tiers for conversational video and video generation, plus the consumer PALs tiers.

Choose the language model that drives a conversational persona — Tavus-hosted models or your own LLM — and let it call external tools/functions mid-conversation.

The perception layer that lets the AI human see and interpret the user — facial expressions, gaze, emotion, posture, surroundings, and shared screen — and feeds that context to the LLM in real time.

Create a photorealistic AI avatar (replica) of a real person that can speak, listen, and respond, or use a ready-made stock replica.

The dialogue-timing model that manages natural conversation flow — when the agent should speak, pause, or yield when the user interrupts — so the back-and-forth feels human.

Render a replica with a transparent or green-screen background, or composite it over a website or supplied video, so the AI human can be overlaid on slides, demos, or branded scenes.

Batch (non-real-time) generation of a finished, downloadable talking-head video by pairing a written script or audio file with a replica.