GTM Hub
What is Tavus?
Tavus is a developer platform for building real-time, conversational video AI agents (PALs) and asynchronous avatar videos. It combines configurable personas, photorealistic visual replicas, perception, turn-taking, speech recognition, LLM reasoning, text-to-speech, and WebRTC delivery into an end-to-end Conversational Video Interface (CVI) pipeline.
Categories
Key features
Conversational Video Interface (CVI)
End-to-end managed pipeline combining perception, turn-taking, STT, LLM, TTS, replica rendering, and WebRTC for real-time face-to-face AI interactions
Phoenix-4.5 & Phoenix-4 Models
Photorealistic rendering with full-face micro-expressions, context-aware active listening, and explicit emotion control. Phoenix-4.5 offers fast preview (~1 min from image); Phoenix-4 requires 3-4 hours from video
Raven-1 Perception
Multimodal perception analyzing facial expressions, tone, gaze, emotion, and ambient environment in real-time to trigger function calls from visual/audio events
Sparrow-2 Turn-Taking
Transformer-based dialogue management with smart turn-taking, interruption handling, and configurable conversational timing
Stock & Custom Replicas
Access to 100+ stock avatars or train custom replicas from 1-2 minutes of video or a single photo with custom voice
Multilingual Support
30+ languages with accent preservation and auto-detection; one agent can serve global audiences
Cross-Session Memory
Agents remember participant context across conversations using flexible memory stores tied to users, sessions, or shared contexts
Function Calling & Tools
Agents can book meetings, pull records, submit forms, and call external APIs mid-conversation based on LLM decisions
Knowledge Base (RAG)
Upload PDFs, docs, or crawl websites for 30ms retrieval; agents answer from your data, not hallucinations
Bring-Your-Own Models
Use any OpenAI-compatible LLM and TTS (ElevenLabs, Cartesia, or custom); keep logic private and fully customize
Real-Time Performance
Sub-600ms latency (often sub-500ms average) from speech to video; 1080p resolution at 24kHz audio and 40+ FPS video
Developer APIs & React Library
@tavus/react-cvi npm package with drop-in components and full TypeScript support for building custom UIs
Integration with Video Platforms
Native support for Google Meet, Zoom, and Microsoft Teams; agents can join as full participants
White-Labeling
APIs and experience fully white-labeled under your brand; enterprise-grade customization available
Conversation Data & Analytics
Full transcripts, emotion timelines, perception events, sentiment shifts, and recordings exportable and queryable at scale
Tavus pricing
Tavus offers separate pricing for developers/APIs (Conversational Video agents) and consumers (PALs messaging app). Developer pricing includes free tier, pay-as-you-go overage per minute, with minutes and concurrent streams varying by tier.
| Plan | Price | What's included |
|---|---|---|
| Free (Basic) | $0/mo | 25 min conversational video, 5 min video generation, 25 stock replicas, 1 concurrent stream, support for 30+ languages, whitelabeled APIs, developer docs & community |
| Starter | $59/mo + pay-as-you-go | 100 min conversational video, 10 min video generation, 3 custom replica trainings/month, 3 concurrent streams, full conversation transcripts, overages at ~$0.37/min |
| Growth | $397/mo + pay-as-you-go | 1,250 min conversational video, 100 min video generation, 7 custom replica trainings/month, 100+ stock replicas, conversation recordings, 10 concurrent streams, overages at ~$0.32/min |
| Enterprise | Custom | Volume discounts, unlimited custom replicas, 100% white-labeled experience, custom concurrency limits, enterprise support & SLAs, guaranteed uptime, bespoke solutions, dedicated Slack channel, zero data retention option |
Pros & cons
Pros
- + Lowest latency on market (sub-600ms, often sub-500ms average) — industry-leading performance
- + Highest fidelity real-time rendering (1080p at 40+ FPS with studio-quality lip-sync)
- + Strong multimodal perception (emotion detection, facial expressions, gaze, objects in real-time)
- + Modular architecture — swap components (LLM, TTS, voice) without rebuilding pipeline
- + White-labeled APIs for enterprise use; full customization available
- + Comprehensive built-in models (Phoenix, Raven, Sparrow) require no external vision/speech tools
- + Rich conversation data (transcripts, emotion timelines, recordings) for analysis and iteration
- + Easy developer experience with React library, REST APIs, and no-code portal options
- + SOC2, GDPR, HIPAA compliant for regulated industries
Cons
- − Consent and biometric data handling workflows need strengthening for privacy-sensitive deployments
- − Hallucination controls and safety guardrails require careful setup and monitoring
- − Accessibility features should be verified for compliance with modern standards
- − Safe escalation paths to human support not always obvious; disclosure of AI agent required
- − Custom face training from video takes 3-4 hours (Phoenix-4); faster option (Phoenix-4.5) carries preview watermark during training
Who is Tavus best for?
- • Real estate agents and property tour automation (visual perception + navigation)
- • Sales demo agents and conversational sales (24/7, multilingual, sub-500ms latency)
- • Customer support and technical troubleshooting (perception-based intent detection)
- • Medical intake and healthcare triage (HIPAA compliance, perception, memory)
- • Coaching, mentoring, and onboarding (cross-session memory, emotional awareness)
- • Interview and hiring screening (real-time perception, emotional tone detection)
- • Content delivery and educational agents (multilingual, video generation + CVI)
- • Enterprise workflows needing white-labeled, fully customizable AI agents
