Keyframe Labs

Competitive Intelligence & Product Roadmap

Turn agents into lifelike real-time video calls.

Company Overview

Keyframe Labs is a video-agent API that renders AI agents as lifelike video-call personas. Serving enterprise training, support, recruiting, telehealth, and consumer AI buyers where no named customers are public.

Latest Intel

Zeitgeist tracks private signals to determine where the company is heading strategically.

May 18, 2026

Technical Approach Signal

Confidence:

Medium

New Intel: Keyframe Labs is centering the product on real-time persona rendering, not agent reasoning. If the $0.06 per minute claim holds, it pressures Tavus, Simli, D-ID, and HeyGen on cost-sensitive training, support, and recruiting workflows.

What They're Building

The company's public product roadmap & what they're committed to building.

persona-1.5-live

The current live persona model adds emotion control and replaces the older live model for better lip sync and expression.

Hosted agent calls

Customers can bring an existing OpenAI Realtime, ElevenLabs, or LiveKit agent while Keyframe hosts the video-call layer.

Self-managed SDK

Teams can keep agent credentials in their own stack and use Keyframe browser primitives for custom embeds.

Widget deployment

Nontechnical teams can design the agent flow in Keyframe and embed the video persona through a no-code widget.

Custom personas

Enterprise buyers can request custom personas with volume commits, higher concurrency, and SLA terms.

Founder and Key Execs

Kaahan Radia

Co-Founder & CTO (former Zipline perception engineer with camera-based autonomy and computer-vision research experience)

Parth Radia

Co-Founder & CEO (former Ericsson AI lead for voice and video synthesis, with prior 3D telepresence work at AT&T Labs)

Founder Force Multiplier

Kaahan brings real-time perception experience from autonomous drones, while Parth brings frontier voice and video synthesis experience from Ericsson and AT&T Labs. The pairing fits a company whose bottleneck is real-time photoreal rendering, not generic agent reasoning.

Funding History

2025 | Founded
2026 | Joined Y Combinator Spring 2026

Competitors

Tavus:

Tavus sells a conversational video interface API with real-time video agents and swappable voice, LLM, and knowledge components.

Simli:

Simli focuses on real-time streaming avatars for AI systems.

D-ID:

D-ID offers interactive AI agents and avatar video tools with a broader creative and enterprise surface.

HeyGen:

HeyGen pairs avatar video creation with LiveAvatar and API products that overlap with real-time persona use cases.

Keyframe Labs

's Moat:

Technical infrastructure is the clearest moat path: low-cost real-time photoreal video rendering could compound if custom persona assets and latency tuning become hard to replace.

How They're Leveraging AI

Emotion-Controlled Live Agent Avatars

Keyframe exposes emotion control for live personas, letting agent experiences shift facial expression states such as happy, sad, angry, and neutral. This makes the video layer useful for roleplay, coaching, customer experience, and digital health interactions where affect changes the experience.

Audio-Driven Photoreal Persona Rendering

Keyframe uses AI to turn spoken agent output into lifelike video-call personas. The system makes an AI agent appear as a real-time human video presence for training, support, recruiting, telehealth, and consumer AI workflows.

Custom Persona Adaptation From Short Video Data

Keyframe can create enterprise custom personas from recorded input data, turning a specific identity into a reusable live video agent surface. This matters for branded trainers, executive avatars, influencer twins, and customer-facing persona experiences.

AI Use Overview:

Persona-1 uses a two-stage flow-based image transformer that maps audio embeddings into compressed image latents, making the video layer the core model rather than a wrapper on an LLM.

Marketing, Sales, and Creator AI

Unify

AI system of action that turns buyer signals into automated, personalized outbound at scale.

A proprietary first-party contact dataset built through waterfall enrichment turns data cost from a line item into gross margin, making Unify a competitor to its own suppliers.

Roblox

Immersive UGC platform where users create, play, and monetize 3D social experiences.

A structural lead in UGC gaming combined with a creator economy paying out $1B+ annually and AI tooling that compresses developer production cycles by roughly 35%.

Humans&

Build a foundation model that maps social dynamics and long-horizon multi-user collaboration.

$480M at seed buys the only team merging social cognition research with frontier model training before the market prices that combination in.

Runway

Generative AI video and world model platform for creative and enterprise workflows.

Proprietary video models combined with embedded studio relationships give Runway both halves of a moat that OpenAI, Google, and Adobe each only have one side of.

Back To All Companies >