Kora plans an OS with AI agents for Intel, AMD and Snapdragon PCs in 2027

AI can help with code or analyze a document, but setting up its working environment often remains a manual job: uploading files, recapping a previous conversation and moving the result into another application. Kora Systems wants to build that work into the operating system. The company has announced a Linux desktop where models and agents will be able to access files, act across applications and retain context between tasks.

According to TechCrunch, a free release is scheduled for 2027. Support is promised for computers running Intel, AMD and Qualcomm Snapdragon processors. For developers and product managers, the interesting part is the underlying bet: if an agent constantly needs documents and several applications, a standalone chat window may not be enough. For now, however, Kora offers a direction and its team's promises, rather than a proven replacement for the familiar desktop.

What changes when an agent lives inside the OS

Imagine a product manager who needs to prepare a plan for the next release. There are interview notes, a spreadsheet of user requests and an earlier discussion of priorities. Today, that scenario often means gathering the materials in one place and explaining the task to the assistant all over again. Kora's idea is to let an agent work with those sources through a shared environment and draw on context already accumulated.

This is an example of a proposed use case, not a test result described in the article. The company says agents will work across applications and access files, but the report does not establish how well the system handles specific programs or long sequences of actions. Those transitions are precisely where its practical value will be tested.

An interface called Chord will provide a way to interact with models. The Cortex engine is intended to run tasks locally whenever possible and turn to cloud models when additional computing resources are needed. Kora plans to support multiple model and service providers. However, the report does not disclose the criteria for choosing between local processing and the cloud.

The company has also announced workspaces that restrict agents' access to files and other resources. This is an important part of the design: an assistant working on one project has no reason to see everything on the computer. How precisely users will be able to configure those permissions remains to be seen.

Linux underneath, a proprietary product on top

Kora is led by Kirk McMaster, a co-founder of Cyanogen. His company, Playtron, previously developed GameOS, a Linux gaming system. In July 2025, it changed its name and pivoted to AI, retaining the ability to draw on nearly three years of prior development work.

The new system is built on Linux and uses Nix; the earlier GameOS was based on Fedora Silverblue. But an open-source foundation does not make the entire product open source: Cortex and the desktop interface remain proprietary. The company has made no commitment to open-source its own code. For a team considering this OS as a work platform, that is a significant factor in its future dependence on the vendor.

About 50 engineers are working on the system. According to McMaster, Codex and Claude have accelerated development by at least five times. That is the company leader's assessment: the report provides no measurement methodology, so it is too early to use it as a benchmark for your own team.

What users will pay for and what needs testing

Kora plans to make money from subscriptions providing access to models and agent services. A free OS installation therefore does not necessarily mean free everyday AI use. The report gives no prices or specific terms.

The company also reports working with Qualcomm and developing an enterprise version, Humain OS, together with Humain. Its launch on Snapdragon laptops is expected in 2027. These plans give the project more substance, but do not answer questions about application compatibility or agent reliability.

The practical step for now is to define a scenario for future testing. Take a task that involves moving data between several applications and count the manual steps. Once a build becomes available, check how many of those steps Kora actually eliminates, what permissions it requires and how much execution costs. That is what will show whether an agent needs an entire desktop.