tavus-overview
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseTavus Overview
Tavus 概览
Tavus is a San Francisco-based AI research lab pioneering Human Computing — teaching machines the art of being human.
Tavus是一家总部位于旧金山的AI研究实验室，开创了**Human Computing（人类计算）**领域——教授机器掌握“成为人类”的艺术。
Mission
使命
"Automation has scaled efficiency but stripped away empathy, nuance, and presence from digital interactions."
Tavus exists to close the gap between humans and machines by creating AI that can see, hear, understand, and respond with emotional intelligence in real time. Not chatbots with faces — authentic, face-to-face digital presence.
“自动化提升了效率，但却从数字交互中剥离了同理心、细微差别和真实存在感。”
Tavus的存在是为了缩小人类与机器之间的差距，打造能够实时观察、聆听、理解并以情商做出回应的AI。不是带脸的聊天机器人，而是真实的、面对面的数字存在感。
What is Human Computing?
什么是Human Computing？
Human Computing is a paradigm shift: computing that adapts to humans, not the other way around.
Core principles:
- Human UI: Interact with AI as naturally as talking to another person — no commands, no learning curve
- Presence over automation: AI that feels like someone, not something
- Emotional intelligence: Reading tone, expressions, context — not just words
Human Computing是一种范式转变：让计算适应人类，而非让人类适应计算。
核心原则：
- Human UI：与AI的交互就像和另一个人交谈一样自然——无需指令，无需学习曲线
- 存在感优先于自动化：AI给人的感觉是“某个人”，而非“某个事物”
- 情商：解读语气、表情和语境——而非仅仅识别文字
The Conversational Video Interface (CVI)
Conversational Video Interface（CVI）
CVI is Tavus's flagship product — an API-first platform for real-time, face-to-face AI conversations.
What makes CVI different from chatbots/avatars:
- Real-time interactive conversation (not pre-rendered video)
- ~600ms latency utterance-to-utterance
- Reads facial expressions, interprets tone, adapts in real-time
- Full orchestration: function calling, RAG, memories
- White-labeled, embeddable, enterprise-ready
CVI Components:
| Component | What it does |
|---|---|
| Replica | The visual avatar — your AI's face and appearance |
| Persona | Behavior, personality, LLM config, system prompt |
| Conversation | A live WebRTC session connecting replica + persona |
CVI是Tavus的旗舰产品——一个基于API的平台，用于实时面对面AI对话。
CVI与聊天机器人/虚拟形象的不同之处：
- 实时交互式对话（而非预渲染视频）
- 话语间延迟约600毫秒
- 可读取面部表情、解读语气并实时调整回应
- 全流程编排：函数调用、RAG、记忆功能
- 支持白标、可嵌入、面向企业级场景
CVI组成部分：
| 组件 | 功能 |
|---|---|
| Replica | 视觉虚拟形象——AI的面部和外观 |
| Persona | 行为、性格、LLM配置、系统提示词 |
| Conversation | 连接Replica与Persona的实时WebRTC会话 |
The Model Stack
模型栈
Tavus builds proprietary models that work together to create human presence:
Tavus构建了一系列专有模型，协同工作以打造真实的人类存在感：
Phoenix-4 (Rendering)
Phoenix-4（渲染）
Gaussian-diffusion model for photorealistic face rendering. Synthesizes high-fidelity facial behavior with:
- Micro-expressions and subtle movements
- Full-face animation (not just lips)
- Real-time emotional response
- Identity preservation
基于高斯扩散模型的照片级真实感面部渲染模型。可合成高保真的面部行为，包括：
- 微表情和细微动作
- 全脸动画（而非仅嘴唇动）
- 实时情绪回应
- 身份特征保留
Raven-1 (Perception)
Raven-1（感知）
Multimodal perception model that lets AI "see":
- Reads facial expressions and body language
- Detects emotions and intent
- Analyzes environment and screen content
- Contextual awareness
多模态感知模型，让AI能够“看见”：
- 读取面部表情和肢体语言
- 检测情绪和意图
- 分析环境和屏幕内容
- 语境感知
Sparrow-1 (Turn-Taking)
Sparrow-1（对话轮次管理）
Transformer-based dialogue model for natural conversation flow:
- Knows when to listen, pause, or speak
- ~600ms response latency
- Handles interruptions naturally
- Multilingual support
基于Transformer的对话模型，实现自然的对话流程：
- 知晓何时倾听、停顿或发言
- 响应延迟约600毫秒
- 自然处理打断
- 支持多语言
Products
产品
Conversational Video Interface (CVI)
Conversational Video Interface（CVI）
API-first platform for developers to embed real-time AI video conversations.
- Full pipeline: perception → STT → LLM → TTS → rendering
- Customizable layers (bring your own LLM/TTS)
- Knowledge base (RAG) and memories
- Function calling for external integrations
面向开发者的API优先平台，用于嵌入实时AI视频对话。
- 全流程覆盖：感知 → STT → LLM → TTS → 渲染
- 可自定义层级（支持接入自有LLM/TTS）
- 知识库（RAG）和记忆功能
- 支持函数调用以对接外部集成
Video Generation API
视频生成API
Async video generation from scripts or audio.
- Personalized videos at scale
- Custom backgrounds and watermarks
- Transparent background support
通过脚本或音频异步生成视频。
- 大规模生成个性化视频
- 自定义背景和水印
- 支持透明背景
Replica API
Replica API
Create digital twins from 2 minutes of training video.
- Studio-grade fidelity
- Stock replicas available
- Identity preservation
仅需2分钟训练视频即可创建数字孪生。
- 工作室级保真度
- 提供现成Replica
- 身份特征保留
PALs (Personal AI Lifeforms)
PALs（Personal AI Lifeforms）
Consumer-facing AI companions that remember, evolve, and connect.
- Text, call, or video chat
- Persistent memory
- Proactive check-ins
面向消费者的AI伴侣，能够记忆、进化并建立连接。
- 支持文字、电话或视频聊天
- 持久化记忆
- 主动问候关怀
Use Cases
应用场景
- Sales & Recruiting: AI SDRs, interviewers, qualification flows
- Education: Tutors, trainers, onboarding
- Healthcare: Patient companions, training simulations
- Customer Support: 24/7 face-to-face assistance
- Personal: Companions, coaches, productivity assistants
- 销售与招聘：AI销售开发代表、AI面试官、资质审核流程
- 教育：AI导师、培训师、入职引导
- 医疗健康：患者陪伴AI、培训模拟
- 客户支持：7×24小时面对面协助
- 个人场景：AI伴侣、教练、生产力助手
Key Stats
关键数据
- 2B+ interactions powered
- ~600ms utterance-to-utterance latency
- 30+ languages supported
- SOC 2, GDPR, HIPAA compliant (enterprise)
- 支持超20亿次交互
- 话语间延迟约600毫秒
- 支持30+种语言
- 符合SOC 2、GDPR、HIPAA合规要求（企业级）
Links
链接
Platform & Docs
平台与文档
- Homepage: https://www.tavus.io
- Developer Portal: https://platform.tavus.io
- Documentation: https://docs.tavus.io
- API Reference: https://docs.tavus.io/api-reference/overview
- CVI Overview: https://docs.tavus.io/sections/conversational-video-interface/overview-cvi
Resources
资源
- Research: https://www.tavus.io/research
- Blog: https://www.tavus.io/blog
- Example Projects: https://github.com/Tavus-Engineering/tavus-examples
- Status: https://status.tavus.io
Community
社区
- Discord: https://discord.gg/5Y9Er6WNN5
- GitHub: https://github.com/Tavus-Engineering
Getting Started
快速开始
- Sign Up: https://platform.tavus.io/auth/sign-up?is_developer=true
- Free Tier: 25 conversational minutes + stock replicas
- 注册：https://platform.tavus.io/auth/sign-up?is_developer=true
- 免费套餐：25分钟对话时长 + 现成Replica
Company
公司信息
- Founded: 2021 by Hassaan Raza & Quinn Favret
- HQ: San Francisco
- Backed by: Sequoia, Scale Venture Partners, Y Combinator, HubSpot
- Category: Human Computing / AI Research Lab
- 成立时间：2021年，由Hassaan Raza和Quinn Favret创立
- 总部：旧金山
- 投资方：Sequoia、Scale Venture Partners、Y Combinator、HubSpot
- 领域：Human Computing / AI研究实验室
Pricing Tiers
定价套餐
| Tier | Minutes | Replicas | Concurrency |
|---|---|---|---|
| Free | 25 | Stock only | 1 |
| Starter ($59/mo) | 100 | 1 custom | 3 |
| Growth ($397/mo) | 1,250 | 3 custom | 15 |
| Enterprise | Custom | Custom | Custom + SLAs |
| 套餐 | 时长 | Replica | 并发数 |
|---|---|---|---|
| 免费版 | 25分钟 | 仅现成Replica | 1 |
| 入门版（59美元/月） | 100分钟 | 1个自定义Replica | 3 |
| 成长版（397美元/月） | 1250分钟 | 3个自定义Replica | 15 |
| 企业版 | 自定义 | 自定义 | 自定义 + SLA |
Ethics & Trust
伦理与信任
Tavus is built on:
- Informed consent: Every likeness used with permission
- Transparent systems: No hidden levers
- Full disclosure: You know how the magic works
- Bias reviews: Active monitoring and advisory oversight
Tavus的构建基于以下原则：
- 知情同意：所有使用的肖像均获得许可
- 透明系统：无隐藏操作机制
- 完全披露：让你了解技术原理
- 偏见审查：主动监控并接受顾问监督