Decart is a frontier AI lab focused on building real-time world models—that is, environments that can respond instantly, run with millisecond-level latency, and achieve unprecedented efficiency. Its official website positions it as "the real-time AI lab the world runs on," with its core work being to build the infrastructure and models that let AI operate at the speed of the real world.
Purpose and Positioning
Decart targets AI scenarios that require real-time generation and interaction, emphasizing low latency, high efficiency, and physical accuracy. It builds both infrastructure (inference and training stacks) and world models, spanning the entire computing stack from hardware to world models.
Main Product Lines
Information shows that Decart builds the infrastructure layer for next-generation low-latency AI systems through three product lines:
- DOS (Decart Optimization Stack): A hyper-optimized inference and training stack, claimed to unlock the full performance of major chips, helping hardware providers and AI teams run faster, more cheaply, and with higher utilization, and built on NVIDIA, AWS Trainium, and Google TPU. The official website also mentions DOS 2.0, as well as a claim of "100x efficiency improvement."
- Lucy: A world model for immersive experiences, supporting real-time video transformation, deployed in scenarios such as virtual try-on, real-time advertising, social platforms, and gaming. Information mentions that Lucy 2.0 pushes high-fidelity video editing from offline rendering to real-time interaction, and supports character replacement, product placement, high-fidelity real-time data augmentation and simulation for robots, and more. There is also mention of Lucy World Editing.
- Oasis: A world model for physical AI, providing real-time responsive, physics-compliant interactive environments for robotics, autonomous driving, manufacturing, drones, and more. The official website introduces Oasis 3 as its flagship interactive world model and calls it a step toward a future where robots are part of everyday life. It also mentions that Oasis supports real-time, infinite video and generates in real time at 30 FPS, with higher-quality editing, physics-aware VFX, stronger temporal consistency, and lower latency.
Research and Technology
The official website mentions the Adaptive-Origin Guidance (AdaOr) method, used to turn "editing strength" in text-driven video editing into smooth, continuous slider adjustment. It also mentions Aha-related research and expressions such as "what you see is what you can reach."
Partnerships and Ecosystem
Decart has partnered with Comcast, bringing high-performance inference to the edge based on NVIDIA GPUs, delivering generative AI experiences under 35 milliseconds. The official website emphasizes that AI-native services are exposing new bottlenecks in AI infrastructure: as millions of users, agents, and devices need access to intelligence, the challenge is shifting from peak training throughput to large-scale deterministic inference—predictable latency, jitter, and a sustainable token economy. The material also mentions the pairing of Decart and ElevenLabs, the former synchronizing lips and the latter granting a soul, together turning digital avatars into living characters.
Funding and Background
The material shows that Decart is an Israeli AI company that emerged from stealth and raised a $21 million seed round, with investors including Sequoia and Oren Zeev. It then raised another $32 million within four months and appeared in an exclusive report: Decart raised $100 million at a $3.1 billion valuation, chasing the future of real-time creative AI. Co-founders include Dr. Dean Leitersdorf and Moshe Shalev.
Applicable Scenarios
According to the official website, its products can be used in robotics, autonomous driving, manufacturing, drones, virtual try-on, real-time advertising, social platforms, gaming, real-time data augmentation and simulation, character replacement, product placement, and more.
Limitations
This material does not provide information such as specific pricing methods, prices, usage steps, or platform availability, and these contents are not yet provided.
