AI Video Agents
Tavus vs Soul Machines: how do they compare on replica governance (consent, controls) and data retention?
14 min read
Most teams exploring AI Humans are asking the same question: if you’re going to put a photorealistic “someone” in front of customers or users, who controls that replica, and what happens to the data around every interaction? This comparison walks through how Tavus and Soul Machines stack up on replica governance (consent, controls, and usage) and data retention, so you can decide which approach fits your risk, compliance, and brand standards.
Note: This article focuses on conceptual, product-level differences. For legally binding details, always review each provider’s current terms, DPAs, and enterprise contracts.
Quick Answer: Tavus is built around white‑labeled, governed AI Humans where you (or your organization) own the replica’s identity, permissions, and deployment surface. Soul Machines leans more into branded, often publicly recognizable digital people. Both address consent and retention, but Tavus emphasizes enterprise control, sub‑second real‑time pipelines, and stricter “you own the human you deploy” governance.
The Quick Overview
-
What It Is:
A practical comparison of Tavus and Soul Machines on consent, replica governance, and data retention for real‑time, face‑to‑face AI Humans. -
Who It Is For:
Product leaders, compliance teams, and engineers evaluating AI Human platforms for customer experience, internal enablement, or consumer apps. -
Core Problem Solved:
You want lifelike, on‑screen AI that can see, hear, and respond like a person—but you need to guarantee that the replica is authorized, controllable, and compliant with your data policies.
How It Works (Framing the Comparison)
Both Tavus and Soul Machines live in the same emerging category: AI-driven digital people who can talk, gesture, and respond. Under the hood, though, the governance story depends on three things:
-
Replica creation & consent:
Who can be turned into an AI Human? How is identity verified and consent captured? -
Ongoing controls & governance:
Who decides where that AI Human appears, what it’s allowed to say or do, and how it can be updated or revoked? -
Data handling & retention:
What happens to video, audio, transcripts, and behavioral logs created by each interaction?
For Tavus, these questions sit inside a model‑led, real‑time stack—Phoenix‑4 for lifelike facial behavior, Raven‑1 for perception, and Sparrow‑1 for conversational timing—delivered through one API you embed into your own product. That white‑label focus pushes governance toward “you own, configure, and gate every AI Human.”
Soul Machines, by contrast, has historically emphasized branded, often celebrity or brand‑mascot digital people, where identity rights, likeness licenses, and behavioral templates tend to be negotiated case‑by‑case with the brand or rights holder.
Replica Governance: Consent & Control
1. Consent to Create a Replica
Tavus
Tavus treats identity as a governed asset, not a visual skin. In practice:
-
Explicit replica setup:
You don’t “scrape” someone into an AI Human. You create replicas through a defined onboarding flow or API path where likeness, voice, and behavior are configured intentionally. -
White‑labeled, owned by you:
The system is designed for developers and enterprises embedding real‑time, face‑to‑face AI into their own apps. That means the default assumption is:- Your organization has the right to use the likeness (your own employees, your own brand character, or a fully synthetic persona).
- Tavus provides the models and infrastructure; you govern who gets turned into an AI Human and under what terms.
-
Enterprise‑grade consent workflows (contractual):
For production deployments, consent and likeness rights are typically captured in employment agreements, model releases, or internal policy—then enforced by how you configure Tavus (which replicas you create, where you embed them, how you authenticate users).
Soul Machines
-
Rights‑driven replicas:
Soul Machines often works with brands, celebrities, and licensed characters. Replica creation generally requires explicit rights to the likeness and voice, usually handled in custom contracts. -
Platform‑defined creation process:
Soul Machines provides tools to design and configure a “digital person.” Consent is wrapped in onboarding, licensing, and legal agreements that govern where this digital person can appear and what it can represent. -
Public‑facing orientation:
Because many Soul Machines deployments are forward‑facing marketing or CX agents, consent is frequently tied to public brand representation rather than private, role‑based internal use.
Net takeaway on consent:
- Tavus is optimized for: “Your organization creates and governs its own AI Humans inside your own surfaces, backed by your internal consent and policy framework.”
- Soul Machines is optimized for: “Brands/rights holders license or design digital people for customer‑facing experiences, with identity rights captured in custom engagements.”
2. Controls Over Behavior, Usage, and Surfaces
Tavus
Tavus is built for developers who care about precise, real‑time behavior at the API level:
-
White‑labeled, controlled surfaces:
- You embed Tavus AI Humans inside your app, your site, your product flows.
- Tavus is not a public directory of digital people; it’s infrastructure you wire into your stack.
-
Policy via orchestration & prompts:
- You shape behavior through system prompts, guardrails, and orchestration around the LLM and perception pipeline.
- You can constrain topics, enforce escalation rules, and limit what actions the AI Human can take (e.g., “may schedule meetings but may not send outbound emails without explicit confirmation”).
-
Role‑ and tenant‑based governance:
- In an enterprise deployment, you can define different AI Humans for different roles (support, sales, training) and restrict each to specific data sources and capabilities.
-
Revocation and updates:
- Because the replica is defined in your configuration, you can update or retire an AI Human centrally—cutting off new sessions and updating the behavior model in one place.
This is all anchored by Tavus’s real‑time pipeline—perception → speech recognition → LLM → TTS → Phoenix‑4 rendering—which means behavior controls play out in live conversation, not just static scripts.
Soul Machines
-
Behavioral templates & persona design:
- Soul Machines typically exposes persona configuration tools where you can set tone, style, and domain knowledge.
- For many enterprise deployments, they co‑design a persona aligned with your brand guidelines.
-
Platform‑governed actions:
- The digital person operates within the capabilities Soul Machines exposes (integrations, dialogue logic, and controls).
- For deeper, agentic actions (e.g., interacting with your back‑end systems), you’ll typically implement custom integration and governance.
-
Central management of digital people:
- You can manage when and where each digital person is deployed, though the tools and granularity depend on your contract and plan.
Net takeaway on controls:
- Tavus: governance is infrastructure‑level and developer‑first—you program behavior, guardrails, and permissions directly into your stack, with the AI Human as a white‑labeled, real‑time interface.
- Soul Machines: governance is often more design‑ and persona‑driven, with behavior framed around brand personality and dialogue flows within their platform.
Data Retention & Usage
3. Interaction Data: What’s Collected and Why
Both platforms need multimodal data—voice, video, and language—to make a digital person feel present. That usually includes:
- Audio streams (user speech)
- Video frames (for perception of gaze, gestures, or environment)
- Text transcripts
- Dialogue context and logs
- Performance metrics (latency, error rates, interruptions)
Tavus
Tavus’s research stack (Phoenix‑4, Raven‑1, Sparrow‑1) is explicitly tuned to real‑time, face‑to‑face interaction. Data is used to:
-
Drive live perception and response:
- Raven‑1 analyzes objects, emotion, and attention in real time.
- Speech is recognized, passed to an LLM, then turned into expressive video and audio.
-
Improve reliability and performance:
- Sub‑second latency and enterprise uptime guarantees rely on monitoring and optimization.
- Operational metrics are logged to maintain service quality.
Typical enterprise‑grade expectations (based on how real‑time AI platforms operate, and Tavus’s emphasis on “best‑in‑class performance and reliability”):
-
Configurable retention windows:
- For logs and transcripts, enterprises often negotiate how long data is retained and whether it’s stored in identifiable form.
- Shorter retention and anonymization are common for regulated industries.
-
Model training controls:
- Enterprises usually require that their content is not blended into multi‑tenant base models without explicit opt‑in.
- Instead, data may be used for customer‑specific fine‑tuning or not used for training at all, depending on contract.
-
White‑label privacy posture:
- Because Tavus is an underlying engine, your brand is what the user sees.
- That pushes Tavus toward a “you own the customer relationship and data; we provide the infrastructure under your DPA” model.
Soul Machines
While implementation details vary by deployment and plan, Soul Machines also requires multimodal data for:
- Real‑time rendering and perception
- Dialogue orchestration
- Performance monitoring
In typical enterprise digital‑human setups, you can expect:
- Platform‑wide logging of interactions for quality and performance.
- Configurable retention and access controls for transcripts and session data under enterprise agreements.
- Optional training/fine‑tuning where your data is used to improve your specific digital person.
The specifics—like exact retention length, regional storage, and training opt‑outs—are contractual and can differ across customers and regions.
Net takeaway on data use:
- Tavus aligns with the “white‑label, infrastructure” model: data is primarily processed to power your real‑time AI Humans and maintain reliability, with enterprise contracts controlling retention and training use.
- Soul Machines generally follows a “platform‑hosted digital person” model: data powers their digital people and platform analytics, with enterprise options to constrain retention and training.
4. Retention, Deletion, and Compliance
Because exact numbers and defaults can change, think in terms of patterns:
Tavus
-
Enterprise retention policies:
- Log retention, transcript storage, and analytics windows can be configured or contractually specified.
- Enterprises can commonly request shorter windows, anonymization, or even log‑free modes for sensitive domains (with tradeoffs in analytics and debugging).
-
Data subject rights & deletion:
- For GDPR/CCPA and similar regimes, Tavus, as a processor, supports data subject rights via the enterprise’s request workflows (export, delete, restrict processing).
- Your app is the front door; Tavus is the processor implementing your instructions.
-
Regional hosting and segregation:
- To support enterprise and regulated deployments, data region and tenant isolation are typically handled at the infrastructure level and spelled out in agreements.
Soul Machines
-
Platform‑defined defaults:
- Soul Machines maintains its own default retention for logs and analytics, usually with enterprise override options.
- Longer retention may be useful for longitudinal analysis of customer interactions but may be constrained by compliance requirements.
-
DPA‑driven obligations:
- Like Tavus, they will commit to data subject rights, deletion, and export under a DPA, with your organization as controller and Soul Machines as processor.
Net takeaway on retention:
Both vendors support enterprise‑grade DPAs and compliance workflows. Tavus’s white‑label, infrastructure orientation often maps more directly onto strict “our product, our governance, our retention rules” requirements, whereas Soul Machines provides similar controls within a more vertically integrated “digital person platform.”
Features & Benefits Breakdown (Governance & Retention Lens)
| Core Feature | What It Does | Primary Benefit |
|---|---|---|
| White‑Labeled AI Humans (Tavus) | Embed real‑time AI Humans into your own app with your branding and flows. | You control identity, consent, and surfaces; Tavus is the engine, not the face. |
| Rights‑Managed Digital People (Soul Machines) | Design and deploy branded digital characters via Soul Machines’ platform. | Strong fit for marketing/brand personas with clear licensing and representation. |
| Configurable Data Retention (Both) | Set how long logs/transcripts persist and how they’re stored. | Aligns AI Human deployments with internal data retention and compliance rules. |
| Policy‑Driven Behavior Controls (Tavus) | Use prompts, orchestration, and access controls to govern live behavior. | Fine‑grained, developer‑level governance over what AI Humans can say and do. |
| Persona & Brand Templates (Soul Machines) | Define personality, tone, and typical dialogues in a visual/UX layer. | Rapidly align a digital person’s style with brand voice and CX flows. |
Ideal Use Cases
-
Best for regulated, product‑embedded AI (Tavus):
Because it’s a white‑labeled, API‑driven platform with enterprise‑grade performance and reliability, Tavus is a strong fit when:- You need strict control over identity, consent, and where AI Humans appear.
- You’re embedding AI Humans deep into your own app (support tools, SaaS workflows, internal enablement) and must align retention and governance with existing policies.
- You want multimodal, real‑time presence (screenshare, surroundings, tone, micro‑expressions) inside your own UX, not on someone else’s platform.
-
Best for branded, marketing‑heavy digital people (Soul Machines):
Because it emphasizes designed “digital people” and persona‑driven deployments, Soul Machines fits when:- Your primary goal is a brand‑aligned spokesperson or front‑of‑site experience.
- Consent and governance are handled via brand/likeness licensing and contractual controls.
- You’re comfortable with a platform‑hosted approach where most UX runs inside the vendor’s framework.
Limitations & Considerations
-
Limited public documentation on deep policy specifics:
For both Tavus and Soul Machines, the most detailed rules around retention, training, and replica governance live in enterprise contracts and DPAs, not on marketing pages. Always request current documentation. -
Your internal governance still matters:
Neither platform can solve consent and retention on its own. You still need:- Clear policies for who can be replicated and why.
- Contractual likeness rights or internal permissions.
- Data mapping, retention schedules, and deletion workflows that match your risk profile.
Pricing & Plans (Governance‑Relevant Context)
Specific pricing tiers and SKUs change over time, but the structure typically looks like this:
-
Tavus Developer Accounts:
Best for builders, founders, and teams integrating Tavus into a product.- Pay for usage of real‑time AI Humans via API.
- Governance and retention are set at the account/tenant level.
- Ideal if you want to experiment with consent models, behavior policies, and custom integrations before scaling.
-
Tavus Enterprise Deployments:
Best for organizations needing strict governance, uptime guarantees, and compliance.- Custom contracts that spell out data retention, training, and replica governance.
- Enterprise uptime guarantees, sub‑second latency, and 30+ language support.
- Tailored to deployment across your organization—internal tools, customer‑facing products, and more.
-
Soul Machines Plans:
While details vary, expect:- Project‑based or seat‑based commercial terms for specific digital people.
- Enterprise contracts that define data usage, retention, and likeness rights.
- Pricing aligned with deployment scale, interaction volume, and design complexity.
Frequently Asked Questions
Does Tavus let me control exactly who can be turned into an AI Human?
Short Answer: Yes. You decide who is replicated and under what terms.
Details:
Tavus is built for developers and enterprises who already own or govern the identities they deploy—employees, brand characters, or fully synthetic personas. Creation of AI Humans happens through your account and configuration, not public scraping. In practice, you’ll pair Tavus with your own consent stack: HR/legal approvals, model releases, or internal policy. Tavus then provides the real‑time perception and rendering stack to make that replica feel present in your app, while you remain the authority on who can be replicated and where.
Can I make sure interaction data isn’t used to train global models?
Short Answer: For enterprise deployments, you can contractually govern how your data is used, including training opt‑outs.
Details:
Both Tavus and Soul Machines support enterprise agreements where data usage is explicitly defined. On Tavus, that typically means your organization can:
- Prohibit use of your data to train multi‑tenant, global models.
- Allow limited use for customer‑specific fine‑tuning or analytics.
- Set retention windows and anonymization rules.
The exact knobs—whether data is retained, anonymized, or excluded from training entirely—are spelled out in your DPA and master services agreement. If you’re in a regulated industry, push for clear language on training, retention, and deletion; both platforms are accustomed to those requirements.
Summary
Replica governance and data retention are where AI Humans either earn trust or fail it. Tavus approaches the problem as infrastructure for “human computing”: white‑labeled, real‑time AI Humans you embed into your own experiences, with your organization owning identity, consent, and deployment surfaces. Data is processed to power perception and presence at the speed of human interaction, with enterprise‑grade controls over retention, training, and uptime.
Soul Machines takes a more platform‑centric stance on digital people: branded, often licensed personas operating within their environment, with governance and retention tuned through enterprise contracts and platform tools.
If you need deeply embedded, developer‑controlled AI Humans with tight governance around who is replicated and how data is retained, Tavus aligns naturally with that model. If your primary focus is a highly produced, brand‑forward digital spokesperson, Soul Machines may map better to your needs—provided you lock down the right consent and retention terms.
Next Step
Ready to experiment with real‑time AI Humans under your own governance and data policies?
Get Started