Parallax - The world's first fully autonomous AI operating system open-sourced by Gradient

What's Parallax?

Parallax is the world's first "fully autonomous AI operating system" open-sourced by Gradient, a distributed AI lab. Parallax supports cross-platform deployment of large models on Mac, Windows and other heterogeneous devices, allowing users to fully control the model, data and AI memory. Parallax is compatible with more than 40 open source models such as Qwen3, Kimi K2, DeepSeek R1, gpt-oss, etc. It is the first fully autonomous AI operating system in the world, which supports cross-platform deployment of models and AI memories on heterogeneous devices such as Mac and Windows.

Parallax Features

• Heterogeneous device networking: Supports cross-platform deployment of large models on a wide range of heterogeneous devices, such as Mac, Windows, etc., to enable collaborative work between devices.
• intelligent dispatch (computing): Built-in network-aware slicing and dynamic task routing mechanism can automatically adjust task allocation according to the inference load, realizing seamless switching among three modes: stand-alone, local multi-device, and wide-area clustering.
• Fully Autonomous Deployment: Users can build and run AI applications such as programming assistants, personal intelligences, etc., locally and with complete autonomy, ensuring that data and control are retained locally.
• high compatibility: It is currently compatible with more than 40 open source large models, such as Qwen3, Kimi K2, DeepSeek R1, gpt-oss, etc., providing users with a rich choice of models.

Parallax's core strengths

• Fully autonomous control: Users have full control over their AI systems, including models, data and AI memories, ensuring privacy and data security and avoiding reliance on cloud-based services.
• Cross-platform deployment: Supports deployment on multiple heterogeneous devices (e.g., Mac, Windows, etc.) for flexible localized applications.
• Intelligent scheduling mechanism: Built-in network-aware slicing and dynamic task routing that automatically adjusts task allocation according to load and supports seamless switching between single machine, local multi-device and wide-area clusters.
• Wide range of model compatibility: It has been compatible with more than 40 kinds of open source big models, providing users with rich choices to meet the needs of AI applications in different scenarios.
• Challenging the traditional modelAs the first fully autonomous AI operating system, it challenges the traditional logic of "AI must go to the cloud" and promotes the localization and autonomy of AI applications.
• Community Support and Impact: Widely noticed by the open source AI industry, it has been supported by a number of organizations such as Ali Qianqi and Kimi, and has achieved excellent list results on platforms such as Product Hunt.

What is the official website of Parallax

• Github repository:: https://github.com/GradientHQ/parallax
• arXiv Technical Paper:: https://arxiv.org/pdf/2509.26182v1

Who is Parallax for?

• AI developer: The desire to build and deploy AI applications in local environments, avoiding dependence on cloud-based services, and the need for flexible model selection and device adaptation capabilities.
• Privacy and Data Security Valuers: Have high data privacy requirements and want to be in full control of their data and AI models to ensure information security.
• business user: AI solutions need to be deployed locally to meet data compliance requirements while enabling efficient task scheduling and resource management.
• Technology enthusiasts and researchers: Interested in emerging AI technologies and open source projects, and wants to explore and practice building and optimizing fully autonomous AI systems.
• Multimodal application developers: The need to deploy and run multimodal AI applications such as image, text and speech processing on multiple devices is met by Parallax's cross-platform and model compatibility.