Asahi fork embraces LLMs and lands Linux on the M4 Mac mini

Gravity Linux, a recently introduced Linux distribution, is now targeting newer Apple Silicon Macs, as announced by its creators. The project's first experimental alpha version has already managed to boot an accelerated desktop on the M4 Mac mini, a model released in October 2024. Gravity Linux is a Fedora remix, designed to provide a fast, modern Linux desktop experience for Apple Silicon devices.

The project was initially forked from Asahi Linux due to policy differences, particularly regarding the use of Large Language Models (LLMs). Despite sharing a common goal of running Linux on Apple Silicon, Gravity Linux and Asahi Linux have taken different approaches to leveraging AI tools in their development processes.

The M4 Mac mini alpha release of Gravity Linux includes GPU acceleration with OpenGL ES 3.0 and OpenGL 3.3 support, catering primarily to developers rather than everyday users. While HDMI output works, USB-C displays, USB4, Thunderbolt, and suspend functionalities remain incomplete, and power management is still in progress. Compared to Asahi Linux, Gravity Linux is ahead in terms of hardware generation, as Asahi Linux only recently added preliminary M3 support, which currently lacks graphics acceleration.

Apple has also moved forward with its own developments, launching M6-based Macs nearly three years after introducing its M3 systems.

The Reg reported on the Asahi Linux project in March 2022 but has since lost several key contributors. Notably, the project lead Hector Martin resigned in February 2025, followed by the GPU effort leader Asahi Lina and the GPU driver effort leader Alyssa Rosenzweig, who left for Intel in April 2025. Gravity Linux, on the other hand, attributes its rapid progress to a development method that Asahi Linux explicitly prohibits.

Project lead Cody Ho, who previously worked at Apple and OpenAI, has utilized AI tools, specifically coding agents, for reverse engineering Apple Silicon components. His two-part account, "I Came, I Prompted, I Left," describes the construction of a custom hypervisor for newer Apple Silicon, as well as the development of an M4 GPU driver in about a month, largely through AI-assisted loops.

This method contrasts with Asahi Linux's Generative AI (LLM) Policy, which broadly forbids the use of generative AI tooling for material contributions. Gravity Linux, however, has a LLM Use and Clean-Room Policy that permits the use of such tools but attempts to contain legal risks through separation. While using an LLM to disassemble Apple's proprietary SPTM component, Ho was ineligible to write the corresponding clean-room implementation, necessitating documentation and hand-off to someone unexposed to the protected material.

The GPU work followed a stricter route, relying on hardware traces, live probing, and self-written shaders, with no inspection of Apple binaries. Despite these efforts, Gravity's first alpha release still requires extensive testing, human review, and refactoring before upstreaming.

Gravity Linux's developers emphasize that their work is more of a demonstration of AI-assisted reverse engineering's potential than a controlled experiment proving Asahi Linux's policies hinder its progress. The comparison between the two projects highlights the significant impact AI tools can have on reverse engineering tasks, though the code still needs significant refinement before it can be integrated into mainstream distributions.