Gregory Kurtzer, founder of the CentOS and Rocky Linux distributions, introduced the Ginux project, which develops the operating system kernel in Rust, aiming for compatibility with the Linux kernel. The released research prototype can already boot the system environment of the Alpine distributions with BusyBox and Rocky Linux 10 with systemd in a multi-user console mode. The project's code is distributed under the Apache 2.0 license. It supports operation on x86-64 architecture systems.
The prototype was created in 11 days as part of an experiment applying the HDAE (Human-Directed Agentic Engineering) development technique to create complex projects at the Linux kernel level. Using the HDAE method, code is generated by an AI agent (utilizing Codex with the GPT-5.6 Sol model), while a human focuses on architecture design, decision-making, coordinating work, forming a development plan, and defining readiness criteria for the components being developed. AI agents are also involved in testing, reviewing, and diagnosing issues.
The project does not aim to replicate technologies from the Linux kernel but instead creates its own compatible implementations while referencing them. Ginux provides an independent memory management system, task scheduler, boot process, process model, file systems, network stack, device drivers, and a layer for compatibility with the Linux ABI. The Ginux kernel boots directly via UEFI without involving a host Linux system.
Ginux does not yet claim to replace Linux and is not ready for use in production systems, but it is already suitable for experimentation and testing. In the long term, the project is considered an operating system for running isolated containers of components for AI infrastructure. The name Ginux is derived from a combination of GPT/Gregory and Linux. The testing platform is based on a virtual machine Q35 (Intel Q35/ICH) running with OVMF (Open Virtual Machine Firmware), Intel IOMMU, and virtio devices. KVM virtual machine Q35 (Intel Q35/ICH) with OVMF (Open Virtual Machine Firmware), Intel IOMMU, and virtio devices.
Of the 375 system calls in the Linux 6.12 kernel, 82 are fully implemented, 245 are partially implemented, and 48 are blocked for security reasons. To identify failures that synthetic tests cannot uncover, kernel performance is evaluated by running real applications and libraries such as musl, glibc, BusyBox, and systemd.
Among the plans:
- Ensuring the launch of user-space components of Linux distributions without using the Linux kernel.
- The ability to create isolated domains environments with their own sets of processes, address space, file systems, devices, resources, and independent life cycles.
- Preparing a self-contained bootable image.
- Providing a minimal user-space control plane that allows kernel management without executing code in ring zero.
- An access control system operating at the level of named objects.
- Support for the isolated launch of container images in OCI format.
Source: opennet.ru
