The pitch arrived today in a familiar voice, with a familiar amount of overreach. Kirt McMaster, who ran Cyanogen and in 2015 promised to put "a bullet in the head" of Google's Android, announced Kora: a Linux-based desktop operating system built on the premise that AI should live inside the machine rather than in a browser tab.

Kora comes out of Playtron, the company McMaster founded to break Windows' grip on gaming PCs, and his framing has not softened with age. "50 engineers with superpowers can take on the Windows dynasty," he says. What the company has actually shown is a website, a list of partners and a great deal of rendering. German outlet heise, reporting the announcement, put it bluntly: apart from partnerships and colourful marketing material, Kora is not showing much of its new AI Linux yet.

The technical claims are where the substance sits, and they deserve to be taken seriously even if the product does not yet exist.

The first is Cortex, described as a "nervous system" for the OS: a hybrid inference engine optimised for local models that unites every chip in the machine — CPU, GPU and NPU — plus every computer on the network into a single computer. Local models run on your own hardware by default; frontier models are called in only when a task requires them.

The second is the disappearance of the folder. Kora proposes replacing the folder-and-app desktop with "neural workspaces and filesystems" organised around intent, context and memory, where each workspace is a sandbox for applications and agents with permissions the user administers. Memory and context, the company says, stay on the user's own machine. A Kora SDK is meant to expose projects, memory, context, semantic search and email to outside developers under the same user-controlled permissions, and the company plans its own suite of mail, calendar, contacts, tasks, team chat and text and image editing built for daily AI use.

The third is gaming and Windows compatibility. GameFrame is a one-click mode that turns a Kora desktop into a console running the "PC catalog" — an evolution of Playtron's GameOS. On Snapdragon PCs, compatibility and translation technology is supposed to run selected Windows x86-64 applications inside the Linux environment, so that switching platforms does not mean losing the software you depend on. Kora says it will bring the system to PCs, including Qualcomm-based machines, in the course of the next year, and it has a second arrangement with the Saudi AI company Humain to build Humain OS for enterprises and public authorities, due commercially in 2027. Investors listed include Samsung, Square Enix, Polychain, VanEck, Circle and Alumni Ventures.

The argument underneath the marketing is the sharpest thing here, and it is unflattering to the incumbents. On Windows and macOS, Kora says, AI runs "in a browser tab or an app sandbox"; it has "no insight into the machine it is executed on", cannot control the underlying hardware, and has no access to the files, memory and applications in its environment — and because both platforms are closed, integration stops there. That is broadly true, and it is exactly the ground Apple and Microsoft are now contesting with their own on-device frameworks.

The obvious objection is that AI probably should not have access to core system components for security reasons. Kora's announcement skips past it, as heise notes. The company's answer is architectural rather than evasive: workspaces are sandboxes, permissions belong to the user, memory stays local. That is a defensible design — and it is the hardest thing to prove, because the security of an operating system that mediates all context depends entirely on whether the permission model holds when the agent is trying very hard to be helpful.

The history is worth holding onto. Cyanogen raised money on the promise of an open Android, shipped a modified Android rather than a new platform, and ended up folded into the community LineageOS project in 2016. The difference this time is that the foundation is Linux, so there is no counterparty to fight. The similarity is that the plan still requires silicon partners, OEMs and developers to arrive on schedule.

And one engineering caveat: pooling "every computer on the network into a single computer" is easy to demo and brutal to productise. Latency, thermals, memory coherence and the fact that most people's second machine is a phone that spends its life asleep all sit between the demo and the shipping feature.

What to watch: a real build, shown running on real hardware rather than rendered; whether Cortex's cross-device pooling ships as a feature or stays a slide; and whether the sandbox model survives contact with users who want their agent to simply do the thing.