What Decart does
Decart (decart.ai) is an AI lab and platform company focused on “live” generative systems—models and the infrastructure required to run them with low and predictable latency. Decart’s positioning centers on a vertically integrated stack it calls the Decart Optimization Stack (DOS), which it describes as optimizing end-to-end performance across major accelerator hardware (including NVIDIA GPUs, AWS Trainium, and Google TPU) so that generative models can operate at the “speed of reality.”
Decart’s model lineup is organized around two world-model products and one video/video-to-video “editing/transformation” product: Lucy (real-time video and world editing), and Oasis (interactive real-time world model intended for “physical AI” and robotics/autonomous systems), both built on DOS; and Lucy’s broader real-time video editing workflow as exposed through Decart’s API platform.
Business model-wise, Decart sells access to its optimization and model capabilities as software/platform products. It describes DOS licensing/revenue via contracts with cloud providers, AI labs, and large hyperscalers, and it also publishes API platform pricing for pay-as-you-go usage of its real-time and non-realtime models (e.g., Lucy 2.5, Lucy VTON 3.5, Oasis 3 Preview).
In terms of technical differentiation, Decart emphasizes (1) low-latency, frame-rate capable generation and responsiveness (e.g., claims about under-30ms responsiveness for Lucy deployments), (2) hardware utilization and performance improvements (including DOS claims like >1,600 tokens/sec for agentic inference and up to 100 high-definition video frames/sec for world model inference), and (3) an infrastructure-first approach where models are co-designed and productized around the performance stack.