A new way to run powerful AI privately and efficiently across devices you already own
Singapore, Nov. 06, 2025 (GLOBE NEWSWIRE) — Gradient, an AI R&D lab building open intelligence infrastructure, has open-sourced Parallax, a new operating system for AI that lets people and teams build, host, and access powerful AI applications on their own devices and networks.
Backed by leading global investors such as Pantera, Multicoin and HSG (previously Sequoia China), Gradient aims to make advanced AI infrastructure accessible to anyone, anywhere. Parallax turns everyday hardware – laptops, desktops, and GPUs – into one adaptive network that behaves like a single high-performance computer. It allows users to deploy their own copilots, assistants, or creative tools locally while scaling seamlessly when extra compute is needed.
Large AI models that normally require expensive cloud infrastructure can now run collaboratively across multiple devices, automatically splitting and routing workloads for maximum speed and efficiency. At launch, Parallax supports over 40 open models across Windows, macOS, and Linux, running on both NVIDIA GPUs and Apple Silicon. In large scale tests across 14 connected machines, Parallax demonstrated significant efficiency gains – up to 3.6× higher throughput and 2.6× lower latency compared with leading local hosting frameworks.
“Parallax is the easiest way to host your own agents and applications – privately, efficiently, and without relying on the cloud,” said Eric Yang, Founder and CEO of Gradient. “The release of Parallax marks a turning point for the AI industry. For years, progress has meant bigger models and bigger clouds. Sovereign AI is the next phase: open systems that anyone can run, verify, and improve. We’re moving from rented intelligence to owned intelligence, where people and teams build, host, and evolve AI on their own terms.”
Parallax is designed for the next generation of AI applications that can stay private, verifiable, and open. As people, developers, and businesses increasingly rely on AI for coding, communication, design, and research, the ability to run those systems securely and independently has become essential.
The system keeps data and memory local, uses open tools, and tracks every step of computation so results can be reproduced and audited. With its Lattica networking layer live, and verification and multi-agent features coming next, Gradient is building the backbone of a sovereign AI ecosystem that can scale globally.
The company has also released the research work for Echo – a distributed reinforcement learning framework that decouples training from inference by leveraging Parallax to deliver efficient, consistent inference performance across networks of machines.
Developers and researchers can access Parallax now at github.com/GradientHQ/parallax and explore the full technical paper on arXiv.
About Gradient:
Gradient is an AI R&D lab dedicated to building open intelligence through a fully decentralized infrastructure – OIS (Open Intelligence Stack), encompassing distributed training, serving, agentic systems, and more.
We are solving the hardest and most ambitious problem in the intersection of Blockchain x AI: How do we train/serve good LLMs over the public internet – Is it even possible to build an OpenAI competitor over a permissionless, autonomous network? Our current OIS includes Parallax for distributed serving, Echo for distributed reinforcement learning, and Gradient Cloud for enterprise solutions.
Backed by top investors and a team of world-class researchers, Gradient is committed to releasing more frontier research that will unlock a future where intelligence can be assembled, scaled, and evolved by anyone, anywhere.
