KORA announces a Linux desktop with built-in AI for Snapdragon PCs in 2027

KORA Systems announced KORA, a Linux desktop operating system that lets AI search files by their contents and run models on the PC or in the cloud, on November 10. The company is targeting a 2027 launch. Its website offered an email sign-up at the time of the announcement. A download and a public testing date have yet to appear in its launch materials.

The first product to ship on the shared platform is scheduled to be HUMAIN OS, an enterprise system. In a joint announcement, KORA and Qualcomm said PCs with Snapdragon X2 Elite chips would ship with HUMAIN OS in 2027. KORA, aimed initially at developers and creators, is due to follow on Snapdragon PCs later that year. The announcement identifies the chip platform and the order of release. It leaves the individual PC models, minimum requirements and exact dates open. Qualcomm says it is continuing to contribute X2 Elite support to the mainline Linux kernel.

The software draws on GameOS, the Linux gaming system the team built under Playtron. KORA says it has been developing the desktop on that foundation since November 2025. The gaming technology will live on as a play mode called GameFrame. The new desktop also uses the operating system to coordinate AI computation, file search and access to project information.

An early version of HUMAIN OS was demonstrated at the LEAP conference in Riyadh in August. The partners’ announcement then focused on a system for enterprise information and services, including authorized access to workplace data. The November announcement specifies which product is due to reach Snapdragon PCs first and the engineering the two systems share.

What an agent can see inside a project

KORA groups applications, agents and files in separate workspaces. The company says the file index stays within those boundaries and each app or agent gets its own access permissions. In a hypothetical example, a freelancer could keep client work and personal files on the same computer and grant an agent access to a single project. Searching that workspace by the contents of a document would let the agent find it even when the filename offers little clue.

The company’s product Q&A also describes what happens when a user connects a calendar, inbox or cloud files. A local model processes the information to build context the applications can draw on, and that context is kept on the device. The company says relevant information can go to a cloud service when the user authorizes it. The Q&A also describes handing a task to another agent within the same conversation, with context carried over within the user’s permissions.

KORA plans to offer software interfaces for outside developers to tap into project context and content-based search. The same permissions would apply to the company’s own apps and those built by others. Details of the software development kit will follow as the product develops. The engine that schedules AI work, called Cortex, is proprietary.

AI inference on Snapdragon X2 Elite

KORA and Qualcomm say they have built a working Linux desktop on Snapdragon X2 Elite. Their joint statement describes distributing inference across the processor’s CPU, GPU and dedicated neural processing unit, or NPU, and says the setup can run a 26-billion-parameter language model on the device. Parameters are the internal values a model adjusts during training. The materials give no memory requirement or speed measurement for a retail configuration running it.

Windows app compatibility and pricing

The companies also described a compatibility layer that runs some x86-64 Windows applications within the Linux environment on Snapdragon. They did not publish a list of those applications. The prospect of moving an existing workflow to KORA therefore depends on two separate details yet to come: which computers will support the system and which of the user’s applications will run there.

KORA says it will disclose pricing and how it will charge for services closer to release. Its product Q&A describes local models as the default and cloud models as an option. The company has yet to specify which models will come with the system or what cloud access will cost.