Tavus

Tavus is an AI platform focused on real-time, human-like conversational AI. Rather than concentrating only on prerecorded AI avatar videos, Tavus has developed a platform for creating AI humans, called PALs, that can see, hear, understand, remember, and respond during live conversations.

The platform combines real-time video, voice, conversational intelligence, visual perception, memory, AI avatars, and developer APIs. Tavus positions this technology as part of its broader Human Computing approach, which aims to make interactions with computers feel more natural and similar to communicating with another person. The company was founded in 2020 by Hassaan Raza and Quinn Favret.

One of Tavus’s central technologies is its Conversational Video Interface (CVI). CVI provides developers with an API for embedding real-time, face-to-face AI conversations into websites, applications, and other digital products. It combines speech, language processing, visual perception, real-time rendering, conversation management, and WebRTC into a single conversational video stack. Developers can also use their own LLM and connect external knowledge sources through the platform.

Tavus’s current platform is built around several proprietary models. Phoenix handles realistic facial rendering, Raven provides multimodal perception, and Sparrow handles conversational timing and turn-taking. Together, these models allow a PAL to respond to both spoken information and visual context during a conversation. Tavus states that Raven can process voice, video, and screen context, while Phoenix generates facial expressions and upper-body movements in real time.

Users can create PALs through PAL Maker, Tavus’s no-code creation environment. A PAL can be configured with a particular appearance, voice, personality, knowledge base, instructions, objectives, and guardrails. Developers who need more control can instead use the Tavus API to build customized conversational experiences directly into their applications.

Tavus also supports AI Rep, AI Interviewer, AI Healthcare Agent, and AI L&D Coach use cases. AI Reps can assist with sales conversations, product demonstrations, qualification, and customer engagement. AI Interviewers can conduct structured recruiting conversations. Healthcare applications include patient intake and support, while L&D applications include role-play, coaching, onboarding, and training simulations.

The platform includes a Knowledge Base for grounding AI responses in uploaded information. It can work with documents and other sources and uses retrieval-augmented generation to provide information to a PAL during conversations. Tavus currently states that its Knowledge Base content support is English-language, even though the conversational interface itself supports substantially more languages.

Another important capability is Function Calling. This allows a PAL to perform actions during conversations, such as booking appointments, logging information, triggering workflows, or escalating an interaction. Tavus also supports persistent memory, objectives, guardrails, conversation transcripts, and structured outputs for business workflows.

Tavus is particularly focused on enterprise deployments. The company states that its platform is SOC 2, GDPR, and HIPAA compliant and supports integrations with business systems such as CRM, ATS, EHR, LMS, and scheduling platforms through its API and function-calling capabilities.

Pricing

Tavus currently offers separate pricing structures for PALs and Developer/API usage.

PAL Plans

• Free — $0/month: Unlimited messaging, 15 minutes of voice and video calls, and support for 30+ languages.
• Plus — $20/month: Unlimited messaging, 150 minutes of voice and video calls, 30+ languages, and MCP Early Access.
• Max — $50/month: Unlimited messaging, 500 minutes of voice and video calls, 30+ languages, and MCP Early Access.

Developer Plans

• Basic — Free: 25 minutes of AI conversational video, 5 minutes of AI video generation, 25 stock replicas, whitelabeled APIs, and support for 30+ languages.
• Starter — $59/month: 100 minutes of AI conversational video, 10 minutes of AI video generation, 3 custom Replica trainings per month, and up to 3 concurrent streams.
• Growth — $397/month: 1,250 minutes of AI conversational video, 100 minutes of AI video generation, 7 custom Replica trainings per month, 100+ stock replicas, recordings, and up to 10 concurrent streams.
• Enterprise — Custom: Volume discounts, fully white-labeled experiences, custom concurrency limits, enterprise security and compliance, and guaranteed SLAs.

Paid Developer plans also use pay-as-you-go billing for additional conversational-video usage. Tavus states that additional CVI usage is charged by the minute, with Starter currently listed at $0.37/minute and Growth at $0.32/minute.

Review

Tavus stands out for focusing on real-time conversational AI humans rather than simply generating prerecorded avatar videos. Its combination of facial rendering, speech, visual perception, conversational timing, memory, and API infrastructure makes it suitable for applications where an AI needs to interact with a person rather than simply deliver a scripted presentation.

The platform is particularly relevant to developers and organizations building AI representatives, recruiting interviewers, training coaches, healthcare assistants, and customer-facing AI agents. Its CVI gives developers access to an end-to-end conversational video stack instead of requiring them to assemble separate speech, avatar, video, and real-time communication systems.

Tavus also provides flexibility through its no-code PAL Maker and developer APIs. Non-technical users can create AI humans through the visual interface, while developers can integrate Tavus into their own products and workflows. Function Calling, memory, Knowledge Base, objectives, guardrails, and webhooks extend the platform beyond basic conversational avatars.

A consideration is that Tavus is primarily designed around conversational AI and AI humans, so organizations looking mainly for traditional text-to-video production or a large template-based video editor may find its product scope different from conventional AI video-generation platforms. Pricing for production use also depends on conversation minutes and other usage, meaning high-volume deployments need to account for usage-based costs in addition to the monthly subscription.

Another limitation is language coverage within the Knowledge Base. Tavus supports 42+ conversational languages, but its current documentation states that Knowledge Base content is English-language only. Organizations with multilingual deployments therefore need to evaluate language support separately for conversation and knowledge retrieval.