Powering 150,000+ developers and the world's most forward-thinking companies

From chat to real-time video, choose how your AI understands and connects.

PALs can be deployed through chat, voice, and video - giving you the flexibility to choose the right modality for every experience.

LESS SIGNAL
Minimal black speech bubble chat icon on white background.

Chat

Ear Icon
Information

How are you doing?

Richness of medium
Speaker icon indicating sound or volume control.

Audio

AUDIO
Speed Icon
SPEED
Ear Icon
TONE
Ear Icon
Information

How are you doing?

FULL PRESENCE
Black video camera icon with a white background.

Video

PRESENCE
Eye Icon
Visual Cues
AUDIO
Speed Icon
Emotion
Ear Icon
Visual Cues
Ear Icon
Information

How are you doing?

INFORMATION ONLY
MEDIUM EXPRESSION
FULL PRESENCE

models

Four models, working in sync.
One seamless conversation.

Tavus unifies turn-taking, rendering, perception, and speech in real time, so every interaction feels immediate, aware, and natural.
Our Research
White background with light blue dotted decorative swirls and curls on edges
Best-in-class TURN-TAKING

Sparrow [2]

Sparrow is the #1 conversational understanding model on the market, built to understand not just what someone says, but when to listen, wait, speak, and respond. It understands the entire audio scene - including pauses, interruptions, backchannels, multiple speakers, and background noise - to make AI conversations feel more naturally human.
#1 on TurnBench - best conversational turn-taking model on market
Lowest error rate - fewer missed turns, false interruptions, and awkward conversational mistakes
Best-in-class rendering

Phoenix [4.5]

Phoenix is the most natural and realistic human rendering model on the market, designed to generate lifelike faces, expressions, gestures, and upper-body movement in real time.  It can render entirely new identities zero-shot from a single image, making photorealistic conversational AI faster and easier to create than ever before.
134ms latency - fastest in market, 25% faster than any other model
Preferred 2:1 - in blind head-to-head testing against nearest competitor
Best-in-class perception

Raven [1]

Raven is a new type of multi-modal perception model, giving machines the ability to see, hear and understand in real-time. It translates facial expressions, tone, gaze, emotion, and environmental context into rich conversational signals, helping your AI respond with empathy, awareness, and intent.
3 simultaneous inputs - understands voice, video, and screen context together
<300ms context freshness - not a snapshot, a continuously rolling read of emotional, environmental and attentional state

Best-in-class interaction

Griffin [Preview]

Griffin is the first human interaction model: a video-to-video model that sees, hears, speaks and moves at the same time, like a person. It keeps listening while it talks, responds to what you show it and gestures as it speaks, so for the first time a machine can hold up its end of a face-to-face conversation.

48% thought it was a real person - on live video, making it the first model to pass the Turing test face to face
#1 on NVIDIA VideoFDB - top score on both perception and generation in NVIDIA's benchmark for face-to-face AI

Giving machines EQ

Humans understand more than words. PALs do too. More than just avatars, they are conversational AI that can look the part, read the room, and understand you.

FACE-TO-FACE VIDEO
NO RANDOM BLINKS
NO CANNED NODS
speech detected
nod in understanding
“Yeah, that sounds fine.”
Excitement  detected
Input: 2min video
cloned
Smiling young woman with long brown hair and light sweater sitting indoors.Smiling young woman with long brown hair and light sweater sitting indoors.
SAM
CUSTOM PAL

Train a PAL tailored to you, with custom faces and voices trained from a short video or a single image.

EMOTION CONTROL

Direct the delivery, not just the words. 10+ emotion states, adjustable mid-conversation.

ACTIVE LISTENING

It's listening while you talk, with advanced turn-taking to know when to speak. Hesitation, interest and confusion register in real time.

Realism and Naturalness

Every frame is generated live, with no random blinks, canned nods, or twitching while it listens.

Stylistic variety

Clone a photorealistic, cartoon, anime character, illustration, and AI image.

Understand what words leave out

Tavus combines words, vocal tone, and facial expression in real time,  giving your agent more context for every response.

Video Perception

Audio Perception

Chat Only

“Yeah, that sounds fine.”
Ear Icon
CONCERN HEARD IN THEIR VOICE
Eye Icon
SUBTLE HESITATION NOTICED
Charlie

Charlie (TAVUS AGENT):  You seem hesitant, and I noticed you looked a bit unsure. Want to talk it through before we make a decision?

“Yeah, that sounds fine.”
Ear Icon
CONCERN HEARD IN THEIR VOICE
VISION OFF
Charlie

Charlie (TAVUS AGENT):  You seem hesitant. Want to talk it through?

“Yeah, that sounds fine.”
Audio off
VISION OFF
Charlie

Charlie (TAVUS AGENT):  Sounds good. Let me know if you have any questions.

Easy to get started. Easy to customize. 
Bring your own, or use ours.

PAL configuration

Curl POST request to tavusapi.com setting a Customer Support persona with empathy and GPT-OSS model.
1 curl --request POST\
2 --url https://tavusapi.com/v2/pals \
3 --header 'Content-Type: application/json' \
4 --header 'x-api-key: $TAVUS_API_KEY' \
5 --data '{
6 "pal_name": "Customer Support",
7 "pipeline_mode": "full",
8 "system_prompt": "Resolve issues with empathy.",
9 "default_face_id": "rc9cff32ceba",
10 "layers": {
11 "llm": {
12 "model": "tavus-gemma-4",
13 "speculative_inference": true,
14 "extra_body": {
15 "temperature": 0.7
16 }
17 },
18 "perception": {
19 "perception_model": "raven-1"
20 },
21 "stt": {
22 "stt_engine": "tavus-auto"
23 },
24 "conversational_flow": {
25 "turn_detection_model": "sparrow-2"
26 }
27 }
28 }'
YOUR CHOICE OF LLM
Tavus optimized
Built in and tuned for latency
Bring your own llm
Connect your preferred model
YOUR CHOICE OF Pipeline
Tavus full pipeline
Everything managed end to end
Bring your own audio
Connect to LiveKit / Pipecat

Docs

Github

Deploy

Tools let PALs take actions during a conversation.

One ask. Watch it get handled while you're still talking.

user

“Convince me this beats what we're using now and if it does, get us switched over.”

tavus

“On it. Here's the plan”

Black silhouette of a hand touching a trident symbol formed by particles on a black background.

Tools / MCP

Wire in your own functions, or anything that speaks MCP, and the PAL calls them mid-conversation.

Hand holding a glowing molecular structure with connected nodes and particles floating around.

Integrations

PALs plug into the systems you already run, from EHRs and CRMs to the rest of the stack.

Screen with a chart showing seats 55%, support 30%, add-ons 15%, and a paused video of a young man.

Present the deck

The PAL pulls the deck into the call and presents it live, slide by slide, pausing wherever the questions are.

Browser Use

Hand the PAL a browser and it drives, walking through the product click by click while you talk.

PAL
When should we switch you over?
Calendar for August 2026 with the 14th day highlighted in purple.

Magic Canvas

The PAL summons the right widget mid-conversation, from a date picker to a chart, when you ask for it.

Tools

Ready. Set. Deploy

Made to customize and fit your setup seamlessly

FILE    EDIT   VIEW   GO   WINDOW   HELP

Caller.App

Interview.exe

Kiosk.exe

Terminal

Make it yours.

The room API underneath, your interface on top. Every pixel is yours.

Icon of an open book with text lines on the left page and bullet points with horizontal lines on the right page.
LIBRARY

Drop it in.

Prebuilt React components: video, transcript, controls. Styled and shipped in an afternoon.

CHARLIE

Send it to meetings.

Your PAL joins like any other participant. Screening calls, onboarding, support hours.

Wire in your stack.

Already on LiveKit or your own media pipeline? Tavus plugs into it.

Intelligence included.

Every PAL ships with the brains built in.

Get Started

Super intelligence toolkit.

ITEM
SPEC
VIEW
Documents

Knows your docs.

KNOWLEDGE BASE · LOWEST-LATENCY RAG
<40ms
RETRIEVAL
Save Disk

Remembers you.

MEMORY · BUILT IN, ACROSS SESSIONS
Zero
SETUP
Language

Speaks their language.

MULTILINGUAL · SAME VOICE, SAME EQ
42+
LANGUAGES
Search

Thinks fast.

SOTA LLMS · OURS, OR BRING YOUR OWN
~600ms
REPLY TIME
Lightning bolt icon

Looks it up

INTERNET SEARCH · FRESH ANSWERS, MID-CONVERSATION
Live
WEB DATA

Production Ready. Scale Ready. Enterprise Ready.

Run your PALs at volume.

Get Started
COMPLIANCE
Four compliance badges for SOC 2, HIPAA, GDPR, and EU AI with respective logos in a 2x2 grid.

Regulatory Compliance

SOC 2, HIPAA, GDPR and EU AI Act, covered out of the box.

CONCURRENCY
Bar chart with four pink bars increasing in height, labeled Conversations on left and Usage on bottom.

Concurrency

Run thousands of simultaneous conversations without queueing.

UPTIME
System status shows all green bars indicating the Tavus app is fully operational.

99.9% Uptime

Production infrastructure with live status on every system.

GLOBAL INFRASTRUCTURE
Pixelated world map with colorful dots scattered across North America, Europe, Asia, and Australia.

Global Infrastructure

Deployed close to your users, wherever they are.

Built by devs, for devs.

A best-in-class API serving millions of conversations every month — stable, scalable, and actually pleasant to build on.

TESTIMONIAL

Tavus has the first AI rendering model that gets everything right—full-face animation, seamless identity preservation, and real emotional nuance. It's a high-fidelity, real-time model with no shortcuts, no limitations—just the best AI-driven human expression out there.

Engineering Leader

Alibaba Cloud

TESTIMONIAL

Before Tavus, AI would interrupt or lag, making conversations feel really awkward. Now, they adapt to each user’s rhythm, making mock interviews flow effortlessly. Our users engage longer, have more in depth conversations, and get a practice experience that truly prepares them for the real thing.

Michael Guan

Co-Founder, Final Round AI

TESTIMONIAL

We're truly impressed by the speed of development cycles at Tavus. We're thrilled to have a front-row seat to access such high-quality AI video. The overall customer service is excellent, and implementing your APIs was so easy. The AI works seamlessly and it was easily integrated into our current tech stack. Thank you!

Aliosha Milsztein

Co-Founder & CEO of Aurio

TESTIMONIAL

Tavus makes AI interviews feel real. Their Conversational Video Interface brings a sense of presence that completely changes the experience for candidates. We implemented CVI in just two days—no complex setup, no headaches. It's an API we actually love to work with.

Richter Brzeski

Engineer, Mercor

TESTIMONIAL

Integrating Tavus into ACTO’s platform enables real-time analysis of facial cues and contextual signals during patient interactions with healthcare professionals. This enhancement allows ACTO to deliver more adaptive, intelligent, and personalized experiences for patients, ultimately improving engagement and decision-making in the healthcare sector.

Kumar Erramilli

CTO, ACTO Health

1

1

6

1

API-first, always.

Everything you need to build with Tavus, accessible through a simple API.

Docs devs actually like.

Best-in-class documentation, quickstarts that work, components that drop in.

White-labeled, end to end.

Your product, your brand. Tavus stays invisible under the hood.

What if you could talk to your computer like a friend?

THE HUMAN COMPUTING MOMENT

A detailed painting of a left hand with fingers slightly bent, palm facing down.Pixelated right hand pointing with index finger extended to the left.
Close-up of a flexed muscular left arm with a visible bicep.

Face-to-face conversational AI in two API calls.

Get started now