We build the infrastructure and models that make AI run at the speed of reality.
Everything humanity is about to build runs on a limited number of chips.
Introducing
Decart Optimization Stack.
Frontier optimization technology that extracts up to 100× more performance from the chips you already run.
Google TPUs
AWS Trainium
Nvidia GPUs
Hardware agnostic
Run optimized workloads across every major chip architecture. Your workload shouldn't care what it runs on.
Peak hardware performance
Extract more from the compute you already have. No new racks. No new contracts. New ceiling.
Ultra-fast inference
Frontier models at lower latency and higher token throughput. Intelligence at the speed of reality.
Deploy the Live AI layer
Contact usOne stack powers every layer of Live AI.
DOS powers Decart's proprietary models as well as external LLMs and agentic workloads.
World Editing Model
Edit reality as it streams. Transform people, objects and environments live.
World Generation Model
Create live, interactive worlds that evolve instantly, with every action.
What Live AI unlocks
Explore what becomes possible when models can edit, generate, and respond live.
Math to Matter
New research, launches, and experiments.
October 2026
Learning a Spatial Memory Codec for Video World Models
Store the whole journey compactly, and read only what each view needs. Each view reads a few learned entries, even from archives of nearly two hours.
Read more
September 2026
Step-Wise Refusal Dynamics in Autoregressive and Diffusion Language Models
Diffusion language models are markedly harder to jailbreak than comparable autoregressive models. We show the difference is tied to the way they generate text, and turn that finding into SRI Guard, a jailbreak detector that matches or outperforms existing defenses at 0.01–0.04% overhead.
Read more
July 16, 2026
Lucy 2.5: Raising the Bar for Live AI
Higher-quality edits, physically-aware VFX, stronger temporal consistency, and even lower latency, all generated live at 30 FPS.
Read more
June 10, 2026
Introducing Oasis 3: First Interactive World Model for Physical AI
Today, we’re introducing Oasis 3 — Decart’s flagship interactive world model, a major step towards a future where robots are an everyday reality.
Read more
May 18, 2026
Decart Raises $300M: Tech Leaders Back the Company as Both Customers and Investors
With funding led by Radical Ventures, Decart is building the infrastructure layer for the next generation of low-latency AI systems, through three product lines: DOS, an ultra-optimized inference and training stack that enables agents and reasoning models to run smarter and faster; and the models Lucy, its World Model for Immersive Experiences; and Oasis, its World Model for Physical AI – both powered by DOS. Today, we’re also announcing DOS 2.0, with new versions of Lucy and Oasis launching in the coming weeks.
Read more
February 2026
What if video editing had a volume knob?
Text-driven video editing lets you describe what to change. But what about how much? We introduce Adaptive-Origin Guidance (AdaOr), a method that turns edit intensity into a smooth, continuous slider.
Read more
January 26, 2026
Introducing Lucy 2: SOTA realtime world transformation model
Today, we’re releasing Lucy 2.0 — a real-time world transformation model that shifts high-fidelity video editing from offline rendering to live interaction. This opens up an unlimited set of possibilities ranging from character swaps and product placement to high-fidelity real-time data augmentation and simulation for robotics.
Read more
October 22, 2025
Building Living Characters With Decart and ElevenLabs
Decart syncs the lips. ElevenLabs gives them soul. Together, they turn digital avatars into living characters.
Read more
October 31, 2024
Oasis: A Universe in a Transformer
Powered by our Oasis video model, your actions shape the environment in real time.
Read more