LIVE TICKER
[Live wire]Connecting to the newsroom…
New TechnologyPublished 10 November 2026 • 9 min read • 2,084 words • 7 cited sources

Kora Linux desktop OS: local AI agents, Snapdragon X2 Elite and a 2027 launch

KORA Systems unveiled Kora on November 10, 2026, a Linux desktop OS that runs AI models and agents at system level. It is due on Snapdragon X2 Elite and x86 PCs in 2027, but pricing, OEM partners and independent benchmarks are still missing.

Kora Linux desktop OS: local AI agents, Snapdragon X2 Elite and a 2027 launch
WIRED (photo-illustration; Getty Images) — view at source ↗

Key facts at a glance

Announcement date
November 10, 2026, 6:00 a.m. PTKORA Systems press kit
Launch window
2027, with HUMAIN OS shipping first and Kora later in the yearKORA and Qualcomm joint release
Launch silicon
Snapdragon X2 Elite with an 80 TOPS Hexagon NPUKORA and Qualcomm joint release
On-device model claim
A 26-billion-parameter language model running locally, with no published benchmarkKORA and Qualcomm joint release
Windows desktop share
About 77% worldwide in October 2026; macOS close to 10%TechCrunch, citing StatCounter
Engineering team
About 50 engineers, with fewer than 100 people at the companyTechCrunch and WIRED

What did KORA Systems announce on November 10, 2026?

KORA Systems Inc. unveiled Kora on November 10, 2026, at 6:00 a.m. Pacific. The Palo Alto company calls it "The Operating System for Intelligence." In practice, Kora is a Linux-based desktop operating system that treats AI models and agents as system components rather than separate apps.

Kora is led by CEO Kirt McMaster, chief business officer John Lagerling and CTO Franck De Girolami. The company credits the team with three operating systems: Cyanogen, Android and GameOS. Kora says those systems reached more than a billion devices. That figure reflects the wider Android ecosystem as much as the team's own work.

Kora grew out of Playtron, the gaming-OS startup McMaster founded in 2022. TechCrunch reports the company rebranded as Kora in July 2025 and shifted its focus to AI. Work on the new system began in November 2025, according to Kora's press kit, using the GameOS foundation.

The sales pitch is that Windows and macOS trap AI inside apps and browser tabs. Kora argues those agents cannot see the machine, schedule its chips or reach files and memory. McMaster calls the desktop operating system "the ultimate harness."

Nothing is on sale yet. Kora says the OS is coming to PCs in 2027. The company has not published a price, a download or independent performance tests.

How do Cortex and neural workspaces work?

At the center is Cortex, Kora's hybrid inference engine. It schedules AI work across a PC's CPU, GPU and neural processing unit. It can also spread tasks to other machines on the local network. Local models run by default, and cloud models join only when a task needs more compute.

The OS is built on Linux with the Nix package manager, TechCrunch reports. TechPowerUp describes the base as NixOS, replacing the Fedora Silverblue foundation used by Playtron's GameOS. Linux and Nix are open source; Kora's own layer is not.

The switch matters to Linux developers because Nix treats system configuration as reproducible code. Kora's public materials do not explain how much of that model it keeps. They also do not say whether users can rebuild the system from a configuration file. For now, the Nix detail is a signal about engineering priorities rather than a user-facing feature.

Neural workspaces replace the familiar folders-and-apps desktop. Each workspace isolates a project or client, and the AI sees only the workspace the user is in. Kora says AppArmor enforces permissions for files, hardware, AI tools and connected services.

A neural file system indexes files by meaning as well as by name. That lets a user search for a contract from a half-remembered detail, as WIRED describes it. Each workspace also keeps version history at the operating-system level.

The memory system analyses the email, calendar, contacts and cloud files a person connects. Kora says that analysis runs locally, and the resulting context stays on the device. Cloud access happens only when the user authorises it.

An SDK exposes projects, memory, context, semantic search and email to third-party apps and agents. Kora says an agentic service could run as part of the OS. That service would borrow the system's identity, memory and compute.

Kora projects that about 70% of the model calls it manages will run locally in 2027. That is a company estimate, not a measured result. The split will depend on hardware and workload.

WIRED saw screenshots with a conventional dock and native apps for email, calendar and contacts. Kora is also building editors and chat. De Girolami described starting a project with an OpenAI model and switching to Anthropic's tools for code. He also described using a local model to save tokens.

Which PCs will run Kora, and when can you buy it?

Kora is coming to PCs in 2027, but the first machines will not run Kora itself. They will run HUMAIN OS, an enterprise system Kora is co-developing with HUMAIN. HUMAIN is the Saudi AI company backed by the Public Investment Fund. HUMAIN OS was first shown at LEAP 2026 in Riyadh. Kora follows later in 2027.

HUMAIN OS is not a rebranded Kora. Kora describes it as an enterprise system organised around intent. Users state a goal, and the OS gathers context, chooses local or cloud models and coordinates agents. It also integrates corporate email, calendars and chat, and adds a suite of AI-native office apps. Thurrott reports it is expected in early 2027.

The launch silicon is Qualcomm's Snapdragon X2 Elite. The platform combines Qualcomm Oryon CPU cores, an Adreno GPU and an 80 TOPS Hexagon NPU. Kora and Qualcomm say it can run a 26-billion-parameter language model entirely on the device, plus speech-recognition and embedding models.

Those figures come from the companies, not from independent testing. No model name, token rate, memory footprint or battery measurement has been published.

The 80 TOPS figure refers to the Hexagon NPU alone, not total system performance. Running a large model on a laptop also depends on memory bandwidth and capacity, which Kora has not specified.

Qualcomm engineers worked with Kora for close to a year, according to the joint release. The chipmaker is also upstreaming Snapdragon X2 Elite support into the mainline Linux kernel. It's FOSS notes that Qualcomm's wider Linux effort for X2 laptops is still early. Its developer preview targets kernel developers, and NPU drivers are still being pushed upstream.

x86-64 is also supported, WIRED reports, so the OS should install on many current Intel and AMD machines. Kora is optimising for configurations with at least 64GB of RAM, WIRED says. That is a demanding target for mainstream PCs.

Windows software is handled by compatibility and translation layers. Kora says selected Windows x86-64 applications will run inside the Linux environment on Snapdragon. TechPowerUp reports the layer is a custom tool based on Wine and Proton, with local models handling configuration.

Gaming survives as GameFrame, a one-click mode inherited from GameOS. It turns the desktop into a console for PC games. TechCrunch reports Android app support is on the roadmap, along with AI tools for creating and publishing games.

There is no ISO to download today. Kora's press kit says pricing and the business model will be shared closer to availability. TechCrunch reports the OS will be free to use, with subscriptions bundling access to models and agent services.

Kora has not named any PC makers. TechCrunch reports that partnerships will be announced later, without a timeline.

Is Kora really open source?

Kora markets itself as open. The promise is that any model, from any lab, can run locally or in the cloud. Third-party agents can plug into the same system primitives as Kora's own apps.

That openness has limits. Cortex and the desktop interface are proprietary, TechCrunch reports. Kora's own Q&A says Cortex is not open source. It's FOSS calls the closed engine a major caveat for a product that leans on Linux's openness.

So Kora is open at the model layer, not at the code layer. Users may choose models, but they cannot inspect or fork the engine that schedules them. That is closer to Android's model than to a community Linux distribution.

The Linux kernel is covered by the GPL, but Kora's proprietary layer sits above it. That is a familiar pattern: an open kernel, a closed user experience. It lets Kora take upstream drivers and security work while keeping its differentiation private. The pledge to support every model concerns integration, not model weights or licensing.

The security story is also a company claim. Kora says each workspace is a sandbox by default. It argues that Windows applies comparable isolation only to apps built on its UWP model. It says AppArmor and granular permissions enforce access.

Independent security researchers have not audited the design. An OS that gives agents system-level reach into files, email and calendars creates an obvious target. Sandboxing and local processing reduce some risks, but they do not remove them.

McMaster's Cyanogen experience hangs over this. Cyanogen was an Android fork that depended on Google Mobile Services. TechCrunch reports he now argues that a challenger which still relies on the incumbent has no chance.

Can a small startup take on Windows?

Windows held about 77% of the worldwide desktop operating-system market in October 2026, according to StatCounter data cited by TechCrunch. Apple's macOS was close to 10%. The remainder is split among Linux and smaller systems.

Kora is not trying to win that whole market. McMaster targets the 45 million to 50 million people who pay heavily for AI and run agents and coding tools all day. He told TechCrunch that Microsoft "can keep its billion users."

Even that niche is large. Kora's press kit estimates 45 million people account for 80% of AI spending. It projects 150 million such users spending more than $400 billion a year by 2033. McMaster gave TechCrunch a higher 2033 figure of more than 200 million users. Both numbers are company forecasts.

The team is small. TechCrunch puts engineering at about 50 people, and WIRED describes a company of fewer than 100. McMaster told TechCrunch that AI coding tools such as Codex and Claude have accelerated development at least fivefold.

Kora is venture-backed. Investors include Samsung Next, Square Enix, Polychain, VanEck, Circle and Alumni Ventures, all of which backed Playtron. McMaster said the company raised more since pivoting but declined to disclose the total or the valuation. He said the funding covers several years.

Thurrott's Paul Thurrott is unimpressed so far. He calls Kora "more of a plan than a product" and notes that the website offers promises and logos but no software.

The competitive field is also filling. Google launched its Googlebook OS with Gemini in October 2026, WIRED notes. Kora's answer is model choice and local compute, not a single assistant.

Microsoft and Apple are not standing still either. Both ship on-device AI tied to their own assistants and cloud accounts. Kora's argument is that those systems limit how far an agent can reach into the machine.

Distribution may be the harder problem. Preloading an OS on retail PCs requires OEM deals, support staff and hardware certification. Kora has none of those publicly.

What are the biggest unknowns before 2027?

Four questions will decide whether Kora matters. First, performance: does a 26-billion-parameter model run fast enough on a laptop, and what does it do to battery life? Second, compatibility: how many Windows apps actually work, and how well do games run on Arm?

Third, price: what does the OS cost, and what do subscriptions include? Fourth, distribution: which PC makers will preload it, and when? Kora has promised OEM partnerships but named none, TechCrunch reports.

The enterprise-first plan adds another dependency. HUMAIN OS must ship before Kora does. It must also satisfy corporate buyers on manageability, compliance and support. Those buyers are less forgiving than enthusiasts.

Privacy claims need proof. Kora says memory and context stay local. The system is still designed to send selected context to cloud models. Users will need clear controls and an audit trail, not just a permissions screen.

Developers will want stable APIs. Kora says its SDK will expose primitives such as memory and semantic search. The details are likely to change before 2027, so building on the pre-release platform carries risk.

A desktop OS also needs years of driver support, security patches and hardware certification. Kora has not published a support policy or an update cadence.

The hardware target matters too. A 64GB RAM target puts Kora near premium workstations. That limits the addressable market at launch.

Why an AI-native OS matters beyond Kora

The wider significance is that the fight over AI agents is moving down the stack. Microsoft and Google put assistants in apps, browsers and cloud services. Kora wants the operating system to own the agent, the memory and the permissions.

That shift could make agents more useful. An OS can see files, schedule silicon and enforce boundaries in ways an app cannot. It could also concentrate risk. An agent with system access is a bigger prize for attackers and a bigger source of mistakes.

For buyers, Kora is a signal of where premium PCs may go. Local inference, private memory and model choice are becoming selling points, not just features. Whether Kora or an incumbent delivers them first is still open.

The enterprise-first order is unusual for a consumer operating system. It gives Kora a distribution channel and a demanding test environment before it asks enthusiasts to switch.

Kora's own framing is generational. It says the desktop was built 40 years ago and has never been rebuilt for intelligence. That is a claim about design, not a proven advantage.

For now, the honest description is a credible team with a detailed design and no shipping product. The next 12 months should bring the evidence that matters: silicon, software, prices and independent tests.

Books behind the context

  • Glyn Moody

    Traces how Linux and open-source collaboration challenged proprietary platforms, the same bet Kora makes on an open kernel.

  • Eric S. Raymond

    Explains the open-versus-closed development tension that Kora reflects, with an open model layer around a proprietary engine.

  • Geoffrey G. Parker, Marshall W. Van Alstyne and Sangeet Paul Choudary

    Provides the platform-economics frame for judging whether an operating system can attract developers and partners at scale.

Sources used for this dispatch

  1. KORA Systems - KORA press kit (announcement, Qualcomm and HUMAIN releases)
  2. WIRED - This New Linux-Based Desktop Operating System Is Built for AI Nerds
  3. TechCrunch - After challenging Google's Android with Cyanogen, its co-founder takes aim at Windows for AI era
  4. SiliconANGLE - Kora builds AI models and agents into a new Linux OS bound for Qualcomm Snapdragon PCs
  5. Thurrott - Kora Promises to Replace Windows With an AI-Infused Linux
  6. It's FOSS - The Company Behind GameOS is Now Building an AI Desktop OS Based on Linux
  7. TechPowerUp - KORA Introduces AI-Focused Operating System with Local AI and Agentic Workspaces

This article is original analysis and synthesis. Facts, figures and quotations are attributed to the sources above. Images remain the property of their publishers and are credited in the caption.

Questions readers ask

What is Kora?

Kora is a Linux-based desktop operating system from KORA Systems Inc. It builds AI models and agents into the system itself, with a hybrid inference engine called Cortex, sandboxed workspaces and a file system that indexes content by meaning.

When will Kora be available?

Kora is due on PCs in 2027. HUMAIN OS, the enterprise version built with Saudi AI company HUMAIN, is expected to ship first, and Kora follows later in the year.

Will Kora run Windows apps?

Kora says selected Windows x86-64 applications will run through compatibility and translation layers, including on Snapdragon PCs. TechPowerUp reports the layer is based on Wine and Proton, but the company has not published a compatibility list.

Does Kora work on Intel and AMD PCs?

Yes, Kora supports x86-64 as well as Qualcomm Snapdragon. WIRED reports the OS should install on many current Intel and AMD machines, although Kora is optimising for systems with at least 64GB of RAM.

Is Kora open source?

Only partly. Kora uses open Linux and Nix foundations and lets users choose models, but its Cortex inference engine and desktop interface are proprietary, as TechCrunch reports.

How much will Kora cost?

Kora has not announced a price. TechCrunch reports the OS will be free to use, with subscriptions bundling access to AI models and agent services. Kora says pricing will be shared closer to availability.

Topics:KoraLinuxAI PCsSnapdragon X2 EliteOperating systemsLocal AI

More from the research desk