Over the years, we have watched countless Linux distributions come and go, and it usually ends with maintainers running short, a project leader retiring, or a lack of traction.

Now, in this AI-fueled era, we are seeing many new distros pop up, and while some of them seem to be trying something different, others are just trying to be macOS clones.

That is why I am excited to see Slax return after years without a release and take on what is expected of a modern Linux distro in 2026. It is a portable distro you run from a USB stick, btw.

A comeback we didn't see coming

The Slax 2026.11 release is based on Debian "Trixie," and is powered by Linux 6.12, an older kernel release that is one of the current long-term support offerings maintained until December 2028.

The previous release came way back in October 2023, where it shipped with two bases, Debian 12.1 and Slackware 15.0.2-current. That's about three years and a month ago, and the recent release post gives no reason for the large gap.

Slax now has a graphical installer for Windows and Linux, where you open the downloaded ISO as a folder, start the installer from there, and pick a drive. It will then create a bootable USB stick or set Slax up alongside an existing system.

Similarly, a single USB stick can hold several named sessions, such as one for work and another for experiments, where you pick one at startup according to the use case.

The Fluxbox desktop now shows window previews when pressing⁣ Alt+Tab and on taskbar icons, and Alt+F3 opens all four workspaces at once. For the launcher, Slax can now promptly download any missing apps if you click on the app icons that have a downward-pointing arrow on them.

A signed bootloader and kernel also let Slax start with UEFI Secure Boot enabled on compatible PCs.

The last addition is Slax Assistant, an AI helper that opens from the launcher. You tell it what you want in plain language, and it can then inspect the system, investigate an issue like disk space running out, and execute commands to fix it.

Though you need to be online and signed in to use it, with email, Google, and GitHub as the sign-in options. The whole thing runs on an online Slax service, details for which are slim.

The sign-in screen offers 100 free credits to try it out, and Slax does not name the AI provider behind it or lay out how they handle data either.

No backsies?

This is not the first time Slax has pulled a disappearing act. In 2022, it came back with a release candidate based on Debian 11 after nothing significant had happened since 2018, apart from a Debian 10.2 test version in 2019.

A Slackware edition followed that summer, and the October 2023 release came after that. The posts published that month never really said anything about what was next.

Here's hoping there are no backsies this time. 🥂

The ISO for Slax 2026.11 is only 536.9 MB, and it runs rather well on a virtual machine if you prefer testing it before committing to a full installation. The Slax user guide is a handy resource to keep by your side if you do get it for your system.

💬 Do you want us to do a review of Slax? It could be an article or video review. Let us know your thoughts below.

A tinkerer, Kathy Korevec, has made ZigZag available under the Apache 2.0 license. She says that the entire app was built with AI, using ChatGPT, and her goal was to get software made to her specs, similar to the Adobe suite, but with MCP support built in.

That MCP support is a built-in server that lets AI agents draw and edit in the same window.

Her post about this on Hacker News has already managed to reach top 10 status (over a 48-hour period), and she has been answering questions there about how this was built and which machines it supports.

ZigZag: Why develop this?

This was built around a Brother Skitch PP1 embroidery machine that she says cost her around $300.

You see, the Skitch needs Brother's Artspira app to operate. While the app is free and works with the machine, a paid plan called Artspira+ adds useful extras such as digitizing images into embroidery patterns and creating cross-stitch patterns.

Kathy says the free version is limited in colors and tools, and she believes some of the clip art is reserved for subscribers. She also guesses Brother sold the machine at a loss at first because of the subscription.

ZigZag imports SVG art and can trace PNG and JPEG images into shapes. It covers 11 stitch types, including fill, satin, and appliqué, with a preview of the stitches being shown in the editor.

As for the MCP server, it lets an AI agent inspect the open design, add and edit objects and text, trace images, preview the stitches, and export files. It can also control a Skitch device, and each action counts as one undo step.

This is not something new.

Open source embroidery tools existed before ZigZag. People have brought up projects like Ink/Stitch, an Inkscape extension, and Respira already lets people send designs to the Skitch without Artspira.

The project also credits Respira for the Bluetooth protocol and sewing flow. Kathy says she reverse-engineered the protocol herself and that her agent used Respira to verify it.

Get sewing

Currently, ZigZag is available for Windows and macOS. You can grab the latest release from the project website or its GitHub releases page.

The README covers installation, Bluetooth pairing, and the MCP server, and it advises against firmware updates on the Skitch. Linux users can follow its steps for building ZigZag from source with Rust.

The Brother Skitch PP1 is the only machine ZigZag sends designs to over Bluetooth, and Artspira needs to be closed first since the Skitch can only talk to one app at a time.

For other machines, ZigZag can export PES, DST, EXP, JEF, or VP3 files.

Suggested Read 📖: Synaptics went hard after a lone open source developer who wanted to get DisplayLink working on Linux.

Traditional operating systems are designed primarily to manage computing resources, including CPU scheduling, memory allocation, storage, networking, and application execution.

Agentic AI introduces another layer of complexity: infrastructure must also support the lifecycle of intelligent agents.

This includes agent deployment, task scheduling, execution isolation, context management, memory retrieval, tool invocation, observability, and recovery.

openEuler is embracing the shift toward agentic infrastructure and evolving its OS capabilities to support the development, deployment, execution, coordination, and optimization of agents across heterogeneous computing environments.

This is what they are discussing at the currently ongoing Agentic Infra OS Conference & openEuler Summit 2026.

AI Agents Need More Than Computing Power

AI agents are moving beyond conversation to execute code, call tools, interact with applications, and coordinate multi-step workflows. As these workloads scale, infrastructure faces a new set of challenges: secure execution, fast environment provisioning, efficient context reuse, persistent memory, and elastic resource management.

Traditional infrastructure was not designed around these requirements. The next evolution of the operating system must go beyond managing processes and hardware resources to support the full lifecycle of agent execution.

This is the direction behind openEuler's agentic OS: building an open, secure, and efficient software stack that connects operating system capabilities with agent runtimes, AI frameworks, data services, and heterogeneous computing resources.

The infrastructure is designed to operate across diverse computing architectures, including general-purpose processors such as x86, Arm, RISC-V, and LoongArch, as well as AI accelerators from multiple vendors. This heterogeneous approach is important because enterprise AI deployments rarely operate on a single hardware architecture or accelerator platform.

The architecture is centered on two layers: Agent Kernel, which provides foundational system-level capabilities, and Agent Service, which exposes reusable services such as secure sandboxing, memory, session management, and Tool Gateway. Around this foundation, complementary infrastructure technologies—including distributed caching and serverless data services—help support the performance, scalability, and data requirements of agent workloads.

Rather than placing every capability inside the OS kernel, the approach combines system-level optimization with reusable infrastructure services and an open ecosystem. Five technical pillars are central to this architecture.

Five Pillars of Agentic Infrastructure

1. Sandbox: Secure, Scalable Agent Execution

Agents need isolated environments to execute generated code, invoke tools, and interact with external systems. These environments must start quickly, scale to high concurrency, and recover reliably when tasks are interrupted.

openEuler brings together lightweight virtualization, snapshot-based recovery, resource prewarming, and high-density scheduling to address these requirements. Technologies such as StratoVirt and the Conch runtime help create and manage sandbox environments, while orchestration capabilities support large-scale deployment.

According to openEuler's performance tests, sandbox startup latency can be as low as 20 ms, with a density of more than 2,000 sandboxes. In a separate large-scale concurrency test, the platform launched 100,000 sandboxes within 10 seconds. These results demonstrate the potential for fast provisioning and high-density agent execution at scale.

2. KV Cache: Making Inference Context Reusable

In multi-turn agent workflows, repeated inference often involves substantial overlap in context. Recomputing or repeatedly transferring the same attention states wastes accelerator memory and increases time to first token (TTFT).

openEuler's KV cache approach combines multi-tier caching, global indexing, NUMA-aware placement, and high-speed cross-node data transfer. With technologies including LMCache and openYuanrong, KV cache can be reused across instances and accessed closer to compute resources.

Agent-aware scheduling takes this further: by mapping agent context to KV cache chunks, the system can predict reuse and proactively coordinate prefetching and eviction.

The performance gains are significant: inference throughput increases by 1.5x–3x, while time to first token (TTFT) drops by 30%–70% in targeted scenarios.

3. Memory Agent: Giving Agents Persistent Context

KV cache accelerates inference by reusing computed states. Agent memory addresses a different challenge: helping agents retain and retrieve useful knowledge across turns, tasks, and sessions.

openEuler's memory infrastructure combines memory extraction and orchestration frameworks with hybrid retrieval and storage capabilities. Solutions involving OpenViking, oGMemory, and openGauss support structured context management, vector and full-text retrieval, and knowledge sharing across agents.

This enables agents to retrieve relevant information when needed instead of repeatedly processing lengthy histories. It also supports shared knowledge accumulation in multi-agent workflows.

By retrieving only the context an agent needs, the system can cut token costs by 30%–70% and improve task success rates by 1.5×–3× in targeted scenarios.

4. Serverless Database: Enabling Elastic Data Services for AI Agents

Beyond the core Agent Kernel and Agent Service layers, data infrastructure plays an important supporting role in agentic workloads. Serverless database capabilities provide the elastic, branchable data environments needed by agents that create, modify, compare, and roll back data across dynamic workflows.

Built on openGauss, this approach combines elastic compute provisioning with copy-on-write storage and database branching. Multiple branches can share baseline data while maintaining independent changes. Point-in-time recovery, branch comparison, and merging further support flexible data workflows.
The result is a more dynamic data foundation for AI agents, where database environments can be provisioned, modified, and recovered as workflows evolve.

The solution is designed for rapid, elastic data provisioning, with millisecond-level instance startup and second-level database branch creation and rollback. These capabilities, enabled by compute elasticity and copy-on-write storage, allow agents to create and manage isolated data environments with minimal overhead.

5. An Open, Heterogeneous Ecosystem

Agentic infrastructure cannot depend on a single model, framework, or hardware platform. It must connect diverse components while making their resources manageable and accessible.

openEuler's ecosystem brings together agent frameworks, inference engines, orchestration platforms, memory systems, and heterogeneous processors and accelerators. Through collaboration with projects and technologies such as Kubernetes, openFuyao, vLLM, MindSpore, and openGauss, it aims to support different agent workloads across diverse computing environments.

The objective is not simply compatibility. It is to provide a consistent infrastructure foundation on which developers and ecosystem partners can build, deploy, and scale agent applications.

From an Operating System to an Agentic Infrastructure Foundation

The transition to Agentic AI is not simply another change in application development. It is reshaping the infrastructure requirements of enterprise computing.

As agents become more autonomous, infrastructure must support more than model inference. It must provide secure execution, efficient data movement, persistent memory, elastic data services, and reliable coordination across increasingly complex workloads.

openEuler's agentic infrastructure initiative addresses these requirements through five interconnected technology areas: an open infrastructure vision, high-performance sandbox execution, distributed KV cache acceleration, persistent agent memory, and serverless databases.

Each addresses a distinct bottleneck in agent deployment and execution. Together, they represent an approach to making agent capabilities more scalable, reusable, and manageable across heterogeneous computing environments.

• Sandbox: Secure and recoverable agent execution.
• KV Cache: Efficient reuse of inference context.
• Memory Agent: Persistent context and knowledge retrieval.
• Serverless Data Services: Elastic and branchable data infrastructure.
• Open Ecosystem: Integration across AI frameworks, orchestration platforms, and heterogeneous hardware.

Together, they point to a broader evolution in the role of the operating system. As AI agents become more autonomous and workloads increasingly dynamic, infrastructure must do more than allocate compute resources: it must help agents execute safely, operate efficiently, retain context, and scale reliably.

This is the infrastructure direction openEuler is exploring with agentic infra OS, building an open foundation for the next generation of intelligent enterprise computing.

First a quick appeal. If you use Google, please add It's FOSS as a preferred source. This way, you won't miss the reliable and trusting tutorials and news from It's FOSS in the noise of AI-generated content.

Speaking of AI, Asahi Linux, the project behind Linux on Apple Silicon Macs, is moving every repository from MIT to GPL v2 or later. The team says people were lifting unreleased code and rewriting it with AI to hide its origin. That hints at Omarchy but DHH told us in an interview that he really doesn't care about this license move.

Linux Mint 27 “Adrien,” due around Christmas. We have created a roundup of its main features so that you can see all the new stuff it will be offering.

Firefox 160 can send a full tab, or just a box you draw around part of a page, into a Mini Window. We tested it and shared this new feature.

Closing a kitty session no longer means losing your layout. The 0.50 release can save every window, tab, and scrollback at logout and restore them at the next launch, once you enable the setting.

Hate Chromium? Domicile is a Rust-based Wayland compositor that draws your entire desktop as a web page using a patched Chromium build ☠️. It only runs on x86_64 machines for now.

Nautilus, or Files, GNOME's file manager, may soon get a neater way to browse long folders. Developers are testing optional section headers that sort files into groups, though this is an early build, so things could change.

Synaptics has sent a cease-and-desist to developer Mike Lothian over Vino, his open source DisplayLink driver. The letters claim patent infringement, and Mike has had to pull the repository from public view.

The company behind GameOS, Playtron, is now KORA Systems. Its new desktop OS, also called KORA, puts AI models and agents right into a Linux base, with a launch planned for later in 2027.

This edition of FOSS Weekly is supported by LabEx.

🧠 What We’re Thinking About

Is de-googling your life worth it? This is the topic of discussion in the It's FOSS Community.

🧮 Linux Tips, Tutorials, and Learnings

Not sure which backup tool to pick? Déjà Dup suits GNOME desktops with almost no setup, while Kopia and Restic handle cloud storage for those who want more control. UrBackup covers fleets of machines.

If you ever wanted to look for a way to help Linux grow, I give you ten 😄

Package manager and repositories are some of the most common terms you will come across as a Linux user. Learn the concept in this article.

👷 Homelab and Hardware Corner

If you want to start homelab without too much hassle, ZimaBoard is a good starting point. The device is 20% off on Amazon for Black Friday early deal.

✨ Apps and Projects Highlights

Teams looking for a password management solution can host Passbolt themselves for free, with no user cap or use its managed cloud.

Helix ships with language server support and tree-sitter ready to go. Just know there’s no plugin system yet, so you’re working with what ships.

📽️ Videos for You

Did you miss the Fedora 45 release? It was a loaded one with many changes.

💡 Quick Handy Tip

On GNOME, you can:

• Open the notification panel using Super + V
• Access the quick settings panel using Super + S
• Get into Activities Overview using the Super key

🎋 Fun in the FOSSverse

There's no dearth of shells in the world of Linux; do you know them all?

📆 Trivia Bit: Firefox 1.0 arrived on November 9, 2004, as Mozilla’s open source alternative to Internet Explorer. It came with tabbed browsing, a built-in pop-up blocker, and phishing protection, and was released for Windows, Mac OS X, and Linux.

Keep on Linux-ing 😄

Till a few years ago, learning Linux was mainly through ebooks and videos. Those methods are still valid, but things have evolved.

Interactive learning makes it easier. Hands-on labs without setting them on your own give you the option to jump straight into the learning. Less intimidating for beginners as well because setting up servers and lab scenarios by hand could be too much when you are just learning the concepts.

Of course, doing things by hand teaches a lot but that could come up later when you have learned the concept, practiced some labs. You are better equipped then.

This is where a platform like LabEx comes in. It is an AI-assisted educational platform that focuses on learning-by-doing by providing you hands-on labs, flashcards, quizzes, and playgrounds to practice in simulated environment.

What can you learn at LabEx?

A lot actually. The platform offers lab-based learnings on a wide variety of tools and topics. Some of the main ones if I can categorize them are:

• Platforms: Linux, Kali Linux, RHEL, etc.
• Frameworks: Shell, Git, Docker, Kuberneter, MySQL, PostgreSQL, SQLite, MongoDB, etc.
• Certifications: CompTIA certificates, RHCSA, CKA, etc.
• Tools: Nmap, Hydra, Wireshark, etc.
• Cybersecurity courses
• Languages: Python, Rust, Java, C, C++, etc.

There are hundreds more, all of which are listed right on the home page.

Ways of learning

Of course, too much could be overwheling. That's why LabEx has the concept of Skill Trees.

Skill Trees

Skill Tree are like roadmaps. Multiple courses are grouped under one umbrella and in a given order, too. Within each tree, there are a definite number of "skills" and courses, which helps keep track of progress.

Skills are basically an essential concept of a course topic. Let's say you learned pipe redirection under Linux course. That pipe redirection is a 'skill'.

Courses are made of skills (think of them as chapters on a mini concept) and skill trees are groups of courses.

If you look at the DevOps skill tree, it has 24 courses. It covers everything from basic Linux to Kubernetes. You get comprehensive learning.

Practice Labs

And then there are practice labs, which take a more project-oriented approach to learning.

For example, "Searching for Specific Files" project makes you look for certain filenames within a directory, following the instructions given on the project page.

Playgrounds

LabEx also has several "Playgrounds" where the concerned system starts up and guides you through the various aspects of the same. It is basically a virtual machine running on LabEx servers.

You would assume that a "playground" means it will let you roam around and explore whatever you might want to freely, but that isn't quite the case.

While you can enter whatever commands you want to, it only lets you through to the next stage if you've completed the tasks on the page, making it pretty much the same as a regular tutorial but with hands-on practices.

Labby to guide you

There is an AI teacher to help you with guided learning. You will find it in the sidebar of the labs and playgrounds. It shows the supported text form of the concept you are learning.

When you run the provided commands for hands on, it also validates your response. You can also ask it questions to clear your doubts.

Challenges

There are thousands of challenges on LabEx that test various skills that you have. These span over all the courses that are present, and add to your overall skill count if you finish them correctly. You can find all challenges here.

Keeps you motivated to continue learning

As any good educational service, LabEx paves your learning progress with tiny achievements. This game-ification of progress helps keep a steady track of progress.

Firstly, on the front page itself, there's a display of how many days in the week the lessons have been accessed. This is accompanied by milestones, which are measured by "skills earned" and "streaks completed". There's also a leaderboard right next to these, which shows a global ranking based on skills earned.

LabEx also gives you flashcards to revise whatever you've learned, which is a very well-proven method that helps retention. It also gives you badges, like "First Lab Conqueror" and "Rising Talent Badge", which help with the feeling of progress, much like a game.

My learning experience with LabEx

The tutorials are all based in a virtual machine which run Ubuntu with Xfce with a minimum number of options present (unless you are working on the Kali Linux labs).

Most of the work will be done on the terminal, of course. Do note that you need good and stable internet connection so that there is no latency between typing in the command and it appearing on the screen of the VM. The execution of the commands is pretty jiffy.

There are 4 tabs on the top, the first one being the desktop itself, as described. The second one gives you a web-based IDE for your programming needs. This is followed by a terminal, which you can imagine as an only-CLI desktop, and finally there's Web 8080, which shows the localhost for any web-services you might host.

About the lessons, the entire structure is supported by the Labby AI, which checks your work at every step, and encourages you (a bit too much sometimes, as AI tends to do). This is aimed at beginners to guide them at every point.

As someone used to a little hardship in learning technical stuff with books, this was quite unusual for me. The tutorials are split into very easily digestible chunks, where Labby AI makes you use the learned commands after they're taught, making sure you retain what you've learnt.

After completing the course, I tried the related playground. Playground gives you hands-on exercises but still in guided fashion. The main test of your learning is with the Challenges where you are given a scenario and you have to troubleshoot or provide a solution in the given playground/VM.

Getting LabEx

LabEx offers free and paid tiers.

The free tier offers 3 virtual machines per day, and access to all hands-on labs. That's a limitation because it means you can't really exceed 3 tutorials in a day. This is good enough to explore the platform and see if it's worth getting the paid subscription or not.

The paid subscription plan gives you everything LabEx has to offer. It gives you access to unlimited VMs, project labs, challenges and certificates of completion.

Annual subscription costs $99.9 and there is also an option to buy 2 year subscription for $149.

Is it worth trying LabEx?

LabEx offers a very wide variety of courses, covering pretty much every major software skill you might need. More than software, I am glad to see that their focus is on Linux, devops, and cyber security.

The interface is pretty smooth, and so are the tutorials.

The guided learning is helpful for beginners as it gives them the foundation of the topic. Combine it with labs, playgrounds and challenges, LabEx has pretty much everything a modern online tech education platform offers. So if you are a beginner, LabEx would suit you.

Experts or people who are accustomed to learning with books or even videos, might feel different because guided learning may feel like spoon feeding to some. But even an expert can benefit by practicing skills with LabEx Challenges here.

You would need a high speed, stable internet connection here to avoid lags with the VMs.

If you are looking for learning Linux and getting certifications, the hands-on learning by LabEx should help you for sure. Besides, $99.9 a year is one of the most inexpensive learning subscriptions you will get in this field.

Give it a try with the free tier and see if LabEx is worth it or not.

Of course, you'll only know if it is for you if you try it out yourself. Let us know what you think of LabEx in the comments. Cheers!

Agent Name Service (ANS) now has a home at the Linux Foundation. The launch happened at KubeCon North America in Salt Lake City, and you might remember that the foundation first announced its plans for the project way back in June.

GoDaddy, Cloudflare, Salesforce, and DigiCert are among the organizations now involved with ANS, bringing the total number up to eleven.

One of the early deployments of ANS has been in AGNTCY, another project under the foundation, where its directory of AI agents can now check an agent’s ANS identity.

Announcing the move, Jim Zemlin, CEO of the Linux Foundation, had this to share:

Autonomous AI agents are transacting and operating across global systems, but they can't succeed without verifiable identity.

Under the neutral governance of the Linux Foundation, ANS provides a common foundation for trust that the entire industry can adopt, implement and improve together.

An explainer

Platforms identify AI agents in their own ways, and a service that receives a request from one has no standardized way to confirm who the agent represents.

ANS tackles this by linking each agent to a domain name its operator owns, the same way a website is tied to its domain. Each version of an agent gets its own name, such as ans://v1.4.2.travel.example.com.

To register an agent, the operator first has to prove it controls that domain. Certificates are then issued for the agent, and every registration goes into a public log that anyone can read.

ANS also publishes details about each agent in DNS, the internet's address book, and other systems can then look them up with tools they already use.

The main DNS records use a format from a separate project called DNS-AID. ANS adds one record of its own that points to the agent's entry in the public log.

When an agent sends a request, the receiving system can check that its registration exists, that it has not been revoked, and that the agent really holds the key tied to it.

ANS only covers who the agent is. Deciding what it can do is left to the receiving system, and platforms have to add the check themselves.

The setup behind ANS

A technical steering committee now oversees the project. It starts with five seats, held by three people from GoDaddy, Ken Huang of DistributedApps.ai, and Michael Kantor of Hashgraph Online (HOL).

That lineup is only a starting point. The first election has to be held within 12 months, and after it, the charter prevents any one company from holding half or more of the seats (to ensure vendor neutrality, I assume).

The charter also mentions that meetings are meant to be open to the public and that discussions and decisions should be visible to everyone.

As for all the code in the project, the Go, Rust, and Java libraries currently use the MIT license, but new code contributed from now on has to use the Apache 2.0 license.

Some lore info

ANS began in 2025 as a draft sent to the IETF, the internet standards body, and a follow-up came in October. Drafts like these have no formal standing at the IETF and are not endorsed by it.

At present, the Linux Foundation calls the specification formalized. The reference code and a local demo are on the project's GitHub page, and the specifications are on ansinfo.ai.

Before we dive into the matter, you first have to understand what DisplayLink is. It is a proprietary graphics transport technology that lets computers connect to one or more external displays over USB, using docks or adapters built around DisplayLink chips.

It works by compressing display data on the host computer and sending it to a chip in the connected device. The chip then decodes it for output to monitors through interfaces such as HDMI or DisplayPort.

On Linux, this support can rub some people the wrong way due to its closed-source nature. That is where Mike Lothian's (FireBurn) Vino comes in as an open alternative.

Unfortunately, Synaptics owns DisplayLink and has sent Mike cease-and-desist letters alleging that he infringed its DisplayLink patents with his project.

What's Vino?

Vino is a driver that lets the Linux kernel talk to DisplayLink docks directly. It is meant to replace DisplayLink's closed-source software and the open source EVDI kernel module that sits beside it.

Mike wrote it with help from Claude Code, and the first version reached the kernel mailing list in June, unfinished. He had faced a communication issue between the dock and the driver and asked other developers for help.

In early July (courtesy of LWN.net), he emailed a Synaptics employee for guidance and mentioned his Rust rewrite of EVDI. About two weeks later, a different employee replied.

That reply said his Rust rewrite of EVDI was not a thing of concern. Rather, the concern was the DisplayLink protocol and codec, which Synaptics believes are covered by its patents.

Lawyers get involved

A cease-and-desist letter followed on August 17. It told Mike to stop development and withdraw anything he had submitted to maintainers and hosting sites, with a signed confirmation due by August 28.

An appendix listed over 125 patents without saying which applied. Mike posted a working version 3 of Vino on August 26 and asked Synaptics the next day which claims they had in mind.

The answer came in October as claim charts. They map one claim from each of two patents, GB2484736 and GB2485613, to his source code.

The October 5 letter said Synaptics would seek $25,000 for the cost of preparing those charts if Mike missed an October 11 deadline, with the figure rising by at least $25,000 each week.

Mike replied on October 6 (message 8) with an offer to pause public work on the encoding code the charts point to if Synaptics paused the deadline and the weekly charge. He also asked about licensing that code for non-commercial use.

On October 7, Synaptics said it would grant no patent license or promise not to sue and set October 18 as a final deadline.

Clean room design quirks

Mike describes Vino as a clean-room driver and says he reverse engineered it from captured USB traffic and vendor programs.

If you didn't know, in a clean-room recreation, one party studies the original implementation, then writes a description, and a separate party builds the new code from that description alone.

It is meant to act as a defense against copyright claims because it shows nothing was copied. Synaptics’ July reply argues that clean-room or independently written code can still infringe a patent. 🤷

The timing

Vino also lands in a busy stretch for AI-assisted reverse engineering, from smart TV cleanups to office suite clones. Legal pressure has reached other AI-built projects too.

Mike has already taken down the Vino repository in response to the demands, and Synaptics has not made a public statement so far.

Only time will tell what happens to the work, and I hope someone forked it before it went offline, as that would give others a good starting point to build on.

Three weeks ago, I wrote that if you can reverse engineer it, it's open source. Looks like the frontier labs didn't love that idea.

Claude Code users are hitting "safety checks" in the middle of decompiling things. Some Codex users got warnings and account restrictions for poking at their own smart TVs. 404 Media calls it a crackdown on the coolest thing people were doing with AI and I agree with them.

This is precisely the reason why local AI and open source/weight models are going to be the next big thing for people who want control over their data and computer.

So this edition is about who gets to say 'no' on your machine.

🐧 AI at the core of new Linux distro

Remember Playtron, the Linux gaming OS from Cyanogen co-founder Kirt McMaster? It rebranded as Kora in July 2025 and has now unveiled Kora OS, a Linux desktop pitched at people who primarily use AI.

Here's the catch. Kora OS will be free when it ships in 2027, but the money comes from subscriptions that bundle AI models and agent services. Cortex and the desktop interface are proprietary. So it's Linux-based, not open, and only "local when possible".

Kora won't be the only one. Omarchy is already there and so do openKylin and openEuler. There will be more distros offering AI-first opearting systems.

🔐 Agent security

You know that "Yes, and don't ask again" option in Claude Code? It saves an allow rule in .claude/settings.local.json inside your project. Sometimes that rule is the exact command you approved, password or token included, in plain text.

Security firm Apiiro searched public GitHub and found 104,185 copies of the file.

The bug report about this goes back to August 2025. A bot closed it as "not planned" after 60 days of no activity. So the same tool that won't help you decompile your own TV firmware will happily remember your production database password.

🔍 Discover AI tools

Memory Poisoning Lab is an MIT-licensed teaching lab for memory poisoning attacks on AI agents. You replay six scenarios, try defenses and see what repair actually restores. It uses a scripted agent and synthetic data, so it runs offline with no API keys. There's a browser demo too.

TurboDiffusion is a video generation acceleration framework that can speed up end-to-end diffusion generation upto 100 times on a single RTX 5090.

Ram-Altman relates to last week's edition. Since Sam Altman was unhappy with consumers not fully utilizing their RAMS, someone created a bash script you run from cron. Any day your Linux box has more than 10% of its RAM available, it mails Sam Altman a note. It's silly but awesome 😄

🗂 AI jargon: Abliteration

In AI, Abliteration is a way to remove the refusal behaviour from an open-weight model without retraining it. Researchers have found that refusals are typically tied to a single "direction" inside the model's activations. Find that direction, subtract it from the weights, and the model mostly stops saying no.

That's why you see "abliterated" versions of popular models on Hugging Face. You see, with open weights LLMS, the guardrails are yours to keep or remove. With a cloud model, they're someone else's call.

😂 Meme

In case you didn't know already, NVIDIA CEO Jensen Huang is accused of dating a Chinese spy.

See you next week 😄

Kirt McMaster founded Playtron in 2022, and the company went on to build GameOS, a Linux-based operating system for gaming PCs. It now goes by KORA Systems Inc. and is launching a new AI-focused desktop operating system.

The OS shares the KORA name, with GameOS living inside it as GameFrame, a dedicated mode for turning the system into a game console. We received a press release regarding this weeks before the launch, so this write-up might be missing some updates.

KORA, explained

KORA is a desktop OS developed on a Linux base, with AI models and agents built right into the system. Don't think that it is an app or a layer over another one; it's a standalone offering.

The company says AI workloads on macOS and Windows stay boxed into a single app or tab, with those having no view of the hardware underneath. KORA is built on the idea that AI belongs inside the OS.

The desktop is organized around neural workspaces, where each one keeps a project or client apart from the rest, and the AI only sees the workspace that is open. Files in it also keep their own version history.

Each workspace doubles as a sandbox too. The user sets what files, devices, AI tools, and online services every app and agent can access, while AppArmor enforces it all.

Search in the neural file system works on meaning, so a person or an agent can find a file by what it contains.

Its memory system builds context from the email, contacts, calendar, and cloud files the user connects. Local models run that analysis on the device, and data goes to the cloud only with the user's approval.

How open is it?

Cortex is the engine behind the models, and it spreads AI work over the CPU, GPU, and NPU, then passes some of it to other computers on the network.

Local models run by default, and cloud ones only join when a task needs them. KORA speculates that local models will handle roughly seven in ten of the calls it manages in 2027.

They also say that models from any outside lab can be integrated, and a chat can move from its own agent to Codex or Claude Code with its context intact. A planned SDK would also let third-party apps and agents tap into projects, memory, search, and email.

Sadly, Cortex is not open source. :(

Still, the wider ecosystem is open, and developers will get the same building blocks KORA's own apps use, such as workspaces, file versioning, and collaboration.

You mentioned Qualcomm?

Yes! Qualcomm engineers worked alongside KORA for nearly a year to bring a Linux desktop to the Snapdragon X2 Elite platform.

Though Qualcomm's wider Linux effort for X2 laptops is still early. Its own developer preview targets kernel developers and not end users, and the drivers meant to open the NPU to local AI are still being pushed upstream.

Even so, KORA says the PCs can run a language model with 26 billion parameters locally, with HUMAIN, the Saudi AI company, working with KORA on an enterprise version.

That version is called HUMAIN OS, with the first OEM PCs expected to arrive preloaded with it in 2027, while KORA itself follows later that year.

So, yes, for now, there is no ISO to download in case you got too excited.

Some thoughts

If you look closely, they are going at it differently than Omarchy, which began in June 2025 as a ready-made setup of Arch Linux and Hyprland that has since landed millions in backing for its foundation.

Omarchy's Quattro release lets users pick from several third-party agents, while KORA builds its own agent into the system and lets others plug in their models. They are basically taking different routes to chase the same bet.

Both seem convinced that AI agents are where the desktop market is headed.

Helix is a modal text editor that operates much like Vim or Neovim. While switching from a GUI like VSCode takes some practice, mastering a terminal editor truly boosts your daily productivity.

Starting with Helix is a smooth experience if you have never used modal editors before. The default settings are intuitive, provided you are familiar with your keyboard layout.

You can avoid the hassle of learning a specific programming language just to configure your workspace. Helix delivers a complete and powerful experience right after you install it.

In my opinion, I suggest Helix if you want the Neovim vibe without the massive time investment. It includes modern tools like language server protocols and tree sitter support by default.

Let's look at how you can get started with Helix as your very first modal text editor.

Installing Helix

Helix is available in the official repositories for several Linux distributions, like Arch Linux, Fedora, and Ubuntu. To install it on Arch Linux, use the command:

sudo pacman -S helix

If you are a Fedora user, install the editor with:

sudo dnf install helix

Ubuntu 26.04 LTS and above also includes Helix in its official repositories. You can install it using:

sudo apt install hx

For all other users, pre-built binaries are available in the official Helix GitHub repository. You can follow the official documentation to complete the installation of these binaries.

Modes in Helix

Helix is a modal editor. It means Helix has different modes with dedicated purposes.

Without having an understanding of the modes, it is hard to grasp how the editor works. So, let's see them one by one.

The Normal Mode

Normal mode is the default state in most modal text editors, where you perform all text manipulations.

Whenever you are elsewhere in the editor, simply press the ESC key to return to this home base. Unlike a traditional editor where you are always in a typing state, modal editors use this mode as a starting point.

For example, you can scroll down a couple of lines by pressing j while in normal mode. You can also remove a character under the cursor by pressing d.

You cannot type any text while you are in this Normal mode. To start typing, you must enter insert mode by pressing i while the editor is in its normal state.

The Insert Mode

Insert mode is the state where you actually type your text. Most traditional GUI editors, like VSCode, have this mode only where all text actions are carried out with the help of keyboard and mouse.

While in insert mode, you generally focus on entering text rather than deleting or copying it. You can still use the arrow keys to move the cursor around in both normal and insert modes.

Whenever you finish typing, simply press ESC to return to normal mode.

Select Mode

Select mode is the equivalent of Visual mode in Vim and is used specifically for highlighting text. You enter this mode by pressing v while in the Normal mode.

The character under your cursor is automatically selected as soon as you enter this mode. You then use navigation keys to extend that selection across your text.

For example, pressing l will extend the selection to the next character, and pressing it again adds the third. You cannot type any new text while you are in this state.

In my experience, combining this with movement shortcuts is incredibly powerful. For instance, I can press v followed by ge to instantly select everything from my cursor to the last line.

You can even press the END key to include the entire last line in your selection. This allows you to highlight massive blocks of text with just a few keystrokes.

Once your selection is finished, use the Normal mode keys to perform an action.

Pressing d will delete the entire highlighted block, while y will copy it for later use.

Other modes of purpose

The above three are the main modes in Helix. But there are some more modes that you may come across.

Command Mode

When you press : in normal mode, you enter the command mode. Here, you can enter supported commands.

Some important operations, like saving a file, quitting the editor, etc., are performed in the command mode.

To save the current file, and quit the editor:

:wq

Helix offers a lot of commands that you can use. Its official documentation on commands has listed it more efficiently.

Picker Mode

Picker mode is triggered by pressing the Space key in normal mode. It is a fuzzy finder interface, where you can invoke a file picker, buffer picker, etc.

If you know Neovim, this picker mode can be a minimal close equivalent of the Telescope plugin.

For example, we can use Space + f to open a file picker.

Also, if I select a block of text and press Space + Shift + c, then you can block comment that section, and so on.

Helix shows the actions

For supplementary modes like picker mode, Helix shows what to press next in a neat pop up.

When you press space, you can see that a pop-up comes up that says all keys and their uses, so that you can read and follow.

Another one I suggest you try is press the g key and read all the subkeys and their functions.

Access the tutor

Helix has an interactive tutorial available to get you started, like the Vim tutor.

Open a terminal and run the command:

helix --tutor

This will take you to a text, which you will read and perform actions as described.

A beginner should complete this tutorial since it gives you a lot of day-to-day shortcut practice inside a sample text.

Essential Keybindings

Helix works flawlessly right out of the box. You do not need any configuration files to make any part of the editor work in the beginning.

But, since it depends on keybindings, before going to other areas, let's see some essential shortcuts to get you started.

Shortcuts to move around the text in Normal mode.

Shortcuts to change the text in Normal Mode

Other routine shortcuts

You can refer to the official keymap documentation for a lot more keyboard shortcuts across modes for specific use cases.

Create a config file

Helix doesn't need a user-written config file to work since it comes with many usable defaults. Still, changing away from defaults needs a config file.

Helix uses the TOML configuration file, located at ~/.config/helix/config.toml.

First, create the directory if it does not exist.

mkdir -p ~/.config/helix/

Now, create a file inside this directory like:

touch ~/.config/helix/config.toml

A different theme

Helix comes with a lot of themes preinstalled, and you don't want to create a long config for it.

Open Helix in a terminal:

helix

Inside Helix, in Normal mode, enter :theme, press a space, and press the TAB key.

You can see that themes begin to show preview. Press TAB to scroll through the live preview of themes. When you spot a theme of your choice, note the name of the theme and press Enter.

Upon pressing the enter key, the theme will be applied to the current session.

To make it permanent, you need to add one line to the Helix config file.

Edit the config file with your favorite text editor.

helix ~/.config/helix/config.toml

And add the line

theme = "theme-name"

For example,

theme = "tokyonight"

Language support

Helix comes with Language Server Protocol Support preset, so there is no need to worry about configuring the LSP.

Still, you need to setup the language server though.

I am using Arch Linux in this example, and in Arch Linux, most of the language server you need are available in the official repository.

To set up CSS support, install the vscode-css-language-server

sudo pacman -S vscode-css-languageserver

Helix will spot this , and you can start working on CSS files from your next start.

To set up Rust support, install the rust-analyzer package:

sudo pacman -S rust-analyzer

Refer to the Language support page in the official documentation to know what package needs to be setup to get the language support working.

Multiple Cursors

Helix supports multiple cursors out of the box. You can duplicate your cursor downward by pressing Shift + c in Normal mode.

This lets you edit several lines at once. If you end up with multiple cursors by accident, press , to collapse back to a single cursor.

Wrapping Up

Helix is a decent modal editor, that offers most of the modern features you would expect in the convenience of an easier learning curve.

I won't suggest replacing Neovim or Vim with Helix if you have mastered those editors. But, if you are into modal editing, or terminal-based text editing in general, Helix is the best place to start.

Will you use Helix for all your coding purposes? Or you have another editor that meets all your needs? Do mention in the comments.

Mozilla has been experimenting a lot with Firefox lately, and a good chunk of that effort has gone into AI. The sidebar can house a chatbot of your choice, and Smart Window has an assistant that lets you pick the AI model.

Luckily, they have stopped tunnel-visioning into AI, and the UX side of the browser has seen some big upgrades.

Containers keep work, shopping, and personal browsing separate; Split View shows two pages side by side in one window; and Vertical Tabs move your tab list into the sidebar.

I believe that these upgrades have made Firefox a much more compelling option when compared to other popular web browsers.

Of course they are not stopping there. Mini Window is the latest addition to that list, which builds on the work done in Picture-in-Picture, the floating video player.

Firefox Mini Window

Using Mini Window on Linux with Firefox 160 feels like a beta from the start. It needs a regular browser window with a tab open to work because a standalone mini window is not possible yet.

It comes in two modes. Move full tab sends the whole page into a small window that floats above your other windows, and Move selection lets you draw a box around the part of a page you want and sends only that.

Either way, the moved tab is not available in the main window until you send it back.

A full tab mini window looks like a tiny browser window. The top bar has the page address along with reload, back, forward, send back, and close buttons. Pages adapt to the smaller size as well.

During testing, websites like ours, Wikipedia, and Associated Press were successfully switching to their compact layout.

A selection mini window is more compact and shows just the chosen area under the page address, with a reload button and the same close and send back buttons.

Both kinds can be dragged around and resized. Selection mini windows have some rough edges because they can appear too small and refuse to resize well.

They are still handy for research sessions where you need to check small bits of information without opening more tabs and cluttering the desktop.

Try it Out

Before you get started using Mini Window, you first have to ensure that you are on Firefox 160 or later. To do that, go into the hamburger menu and open "Help and Report." Here, click on "About Firefox" and take note of the version shown in the dialog that opens up.

If you find that an older build of Firefox is installed, then get the latest release from the official website or the official FTP mirror.

Next, you just have to start a normal browsing session with multiple tabs and follow any one of these three approaches to get started:

• Add the Mini Window icon to the toolbar.
• Right-click a tab and go into "Move Tab."
• Right-click the page content and choose Move Section to "Mini Window."

For the toolbar method, right-click the toolbar and open "Customize Toolbar." Find the Mini Window icon in the list and drag it onto the toolbar, then hit "Done." After that, the feature is just one click away on any page.

Suggested Read 📖: Firefox can do a lot more than you realize. Here are 21 features most people overlook.

If you have followed desktop Linux this year, you could not have missed Omarchy. And if you have followed Omarchy, you could not have missed DHH.

David Heinemeier Hansson, better known as DHH, created Ruby on Rails and co-owns 37signals, the company behind Basecamp and HEY. In 2025, he released Omarchy, an opinionated Arch Linux setup built around the Hyprland tiling window manager. It looked great and had a keen following. But it was still one more Arch-based setup.

Then AI moved into its core. With Omarchy Quattro, agents became a first-class part of the desktop, and Omarchy took off. It drew in people who had never touched Linux before. Its plugin marketplace crossed 500 plugins in its first week and now has thousands, and the Omacom Foundation has more than 20 million in pledges, including AI tokens from Anthropic, OpenAI and Meta.

That rise has not been quiet. Many long-time Linux users distrust AI-generated code, and the flood of vibe coded plugins has definitely raised real security worries.

Asahi Linux, the project that makes Linux run on Apple Silicon Macs, has a no-AI contributions policy and has just moved from MIT to GPL, citing AI-assisted code laundering. And DHH, never shy of a fight on Twitter, is a polarising figure in his own right.

So I sent him questions over email about all of it: the AI bet, the Asahi licence change, plugin security, the foundation money, his critics and what comes next. Here are his answers, lightly edited for typos.

Omarchy was a nice-looking tiling setup on Arch until AI was added to its core, and then it got extremely popular, even among non-Linux users. Why do you think that happened? What did Omarchy do differently that hundreds of other distros didn't?

The age of agents is changing everything. We clearly saw its potential with Omarchy Quattro, and it's no surprise to me that anyone who's interested in agents is therefore also interested in Omarchy. We've simply made the best of a wonderful opportunity that's been offered to Linux. Meanwhile, lots of other open source projects in this ecosystem have decided to take a very anti-AI stance, and their results are as you would expect.

"Lots of other open source projects in this ecosystem have decided to take a very anti-AI stance, and their results are as you would expect."

Ubuntu is carefully adding AI to its desktop. We have a couple of examples from China in the form of openKylin and openEuler. Other than those, I don't see any other major distros doing that. Why has no other major distro gone as far as you?

Ubuntu is built on top of GNOME, which is extremely hostile to the idea of agent development. So it's not a surprise to me that they're having a hard time integrating the best of AI into their distribution. The same goes for many other distributions that are deeply entwined with open source projects that are hostile to AI. It's difficult to fully embrace the age of agents when you're steeped in such thinking.

Omarchy chose Superwhisper for voice input. Ubuntu has an open source solution for the same. Sometimes it seems that your focus is less on open source and more on Linux.

I don't have the repulsion to commercial software that many in the old Linux world have. I love commercial options when they give us the best software. I've worked on commercial software my entire life. I am happy to integrate the best commercial offerings, whether it's Superwhisper or Spotify or Dropbox or anything else, into Omarchy. I'm creating a distribution that includes all the software that I personally want to use, and plenty of that software is commercial.

You've said that you respect Asahi but disagree with its no-AI policy, and you called for an agent-driven effort to finish M3 to M5 support. This month, Asahi moved from MIT to GPL, citing AI-assisted code laundering. Do you read that as a response to Omarchy?

Asahi changing their license may be a response to Omarchy, but it also doesn't matter. The Linux kernel is already licensed under the GPL, and we are in full compliance with those terms. We have a bunch of patches to the kernel that are available as open source. If Asahi is licensed in the same way, we will continue as we have all along and build on top of the best open source software. Isn't that the point of open source? That any project is free to incorporate open software as long as they abide by the license? That certainly is how we approach it within Omarchy. I don't see how anything changes at all.

Isn't that the point of open source? That any project is free to incorporate open software as long as they abide by the license?

Many Omarchy supporters seem angry with Asahi. But does the licence change affect Omarchy at all? Omarchy can continue its Mac support work with the MIT-licensed Asahi code it already has.

I don't think there's any reason to be angry with Asahi, but it's fair to be disappointed that our advances can't be upstreamed such that we could get a single body of work that enables Linux to run well on Macs. But personally, I can't get too excited about it either, because whatever they choose to do, it doesn't really have any bearing on our mission. We will simply build on top of the best open source solutions that are already out there and then fill in the gaps.

If Asahi didn't have a "no AI contributions" policy, would Omarchy have contributed back the work it has done to bring Omarchy to Apple M hardware?

Of course.

Omarchy's plugin market passed 500 plugins in its first week, and it now has thousands. Many of these are vibe coded by hobbyists, unlike AI-assisted development by experienced programmers, and this worries a lot of people. What stands between a malicious plugin and an Omarchy user today? And what do you say to people who think Omarchy is unsafe because it's "vibe coded" and not properly tested for security issues?

The plugin marketplace is agentically reviewed. There is surely more we can do to lock things down, but it's already far more secure than, say, something like the AUR. I would say that it's a feature, not a bug, that the majority of these plugins have been created by AI, as AI is incredibly good at sticking to established conventions. And it's also very good at reviewing incoming code for security vulnerabilities.

It's a feature, not a bug, that the majority of these plugins have been created by AI.

The Omacom Foundation has about $18.5 million in pledges, including AI tokens from Anthropic, OpenAI and Meta, to spend over three years. How do you plan to spend it? If the donations run out, is there a plan to make money in some way? How would you keep Omarchy going?

We have already hired four full-time employees. We have started the AIR project to sponsor artists. We are paying for infrastructure, and we are supporting community members with hardware and tokens to pursue integration projects like Omarchy M and Omarchy Dragon. There will be a lot more to come, including contests and other forms of community engagement. We will find a way to spend all the money. And when it's gone, we will either have succeeded so well that it'll be easy to raise more, or the world has changed to such an extent that Omarchy is no longer necessary. Both of these outcomes are just fine with me.

Much of the opposition to Omarchy is about two things: AI, and you. What would you tell a Linux user who likes Omarchy but not its creator?

I'm not here to persuade any haters. I'm simply offering the gift of Omarchy to anyone who wants it. And if someone can't accept it because they have a problem with me, their loss. Thankfully that's a tiny but vocal minority in the existing Linux community. The ambition for Omarchy is much, much larger than just converting people over from other Linux distributions. In fact, I don't really spend much if any time thinking about that. The way to make Linux win is by converting Windows and macOS users.

The way to make Linux win is by converting Windows and macOS users.

Fragmentation is often blamed for holding desktop Linux back, and you've added another distribution, even if it does things differently. What do you make of the "there are way too many distros already" argument?

There are very few Linux distributions that matter in any grand scheme of things. I don't think fragmentation is a problem at all. What's a problem is having such a tiny user base as Linux on the desktop has had for decades. Omarchy is on a mission to make a major dent in that by converting large numbers of Windows and macOS users to a distribution that is not only materially different from what they already have, but is the right answer for the age of agents. That turns out to be a pretty unique proposition in the world of Linux at the moment.

We all know how vocal and passionate you are about Omarchy on Twitter. Do you still develop new features of Omarchy, or is it completely handled by the team you have formed? What role do you play on the development side?

Designing features and defining the user experience of Omarchy is the main thing I spend my time on. We have a wonderful community helping with a thousand different aspects of what it takes to run a successful distribution, but defining that key experience is still the main work that I do.

Elon Musk says AI will soon skip code and generate binaries directly. You built Rails around programmer happiness. Do high-level languages matter when the programmer is an AI?

They clearly don't matter in the same way they did when we were writing everything by hand. I think the future of programming is still up in the air, though. But I also think it is very worth listening to what someone who has been as correct as Elon Musk has been on a wide variety of topics thinks about this future. I am certainly spending a lot more of my time developing in lower-level languages I would never have engaged with prior to the age of agents.

Servers, phones, hardware or something else entirely: what's one plan for Omarchy that nobody outside the core team knows about yet?

First, we take Manhattan, then we take Berlin.

What would you like to tell It's FOSS readers, many of whom are long-time Linux users wary of AI?

Take another look with fresh eyes and an open mind. AI is the most amazing thing we've made computers do. Don't miss out because of your priors, because of politics, because of what the technology looked like a year ago.

The Agentic Infra OS Conference & openEuler Summit 2026 will take place November 13-14 at the Zhongguancun International Innovation Center in Beijing.

Under the theme "Building Agentic Infra from the OS Up," the event will bring together developers, open-source contributors, and industry partners to explore how operating systems can support large-scale AI computing and the deployment of AI agents.

The two-day event pairs an industry-focused first day, headlined by the Agentic Infra OS Conference keynotes, with a community-focused second day built around the openEuler Summit.

It's FOSS is going to follow the event remotely and will report on the interesting bits of the event again when the event has concluded.

Operating systems are becoming part of AI infrastructure

As AI applications grow more complex, infrastructure teams face challenges that extend beyond model performance. The challenges include coordinating computing resources across tightly integrated systems, running agent workloads securely, and understanding how systems behave in production.

Only a few months back, openEuler shipped 24.03 LTS SP4, an update built on Linux kernel 6.6 that added an agent sandbox runtime for enterprise AI agents, SuperPoD reliability and usability enhancements, NPU resource partitioning, rapid fault recovery for inference services, and AI compiler and performance work.

The release reflected contributions from 2,006 developers across 39 community member organizations.

The reason why I am mentioning this is because that release frames the questions at the center of this Summit.

As agents run longer, multi-step task loops and hold state and memory across systems, the process and resource abstractions at the heart of today's operating systems are being stretched.

The conference will examine these challenges through two connected areas: SuperPoD OS and agentic infrastructure.

SuperPoD OS

SuperPoD OS, openEuler's operating system for SuperPoD computing clusters, addresses the system software needed to coordinate resources across large, tightly integrated computing systems.

At the event, openEuler is going to showcase its latest developments in this area, building on the UnifiedBus, memory-pooling, and fault-localization work introduced in 24.03 LTS SP4.

The published agenda includes JD.com's experience with openEuler-based SuperPoD systems for advertising workloads, alongside a session on Ubuntu and openEuler's work to develop the SuperPoD ecosystem.

JD.com is a Fortune Global 500 company and China's largest direct-sales retailer.

Agentic infrastructure

The agentic infrastructure program will focus on the foundations needed to run AI agents reliably and securely - from sandboxed execution environments and OS-level memory and KV-cache management to integration across the AI software stack.

A scheduled presentation from Meituan, a food delivery tech giant, will explore its use of openEuler for agent sandboxing, offering a practical perspective on managing agent execution environments in production.

Event overview

Together, these sessions will connect operating system development with the needs of organizations building and deploying AI applications.

They will also give developers an opportunity to discuss engineering challenges directly with industry practitioners and open-source contributors and to see the technology hands-on in the event's demonstration zone.

The two-day event will feature more than 30 keynote presentations and panel discussions, complemented by technical sessions and a massive 1,000-square-meter demonstration and interaction zone.

There are plenty of subforums with topics on kernel, AI infrastructure, development tools, performance optimization, edge computing, and software security.

Day 1 also includes invitation-only sessions for internet and financial-industry scenarios.

Registration

Developers, infrastructure engineers, and ecosystem partners can view the agenda and register through the official event website:

Linux Mint is an Ubuntu-based distribution equipped with its own desktop environment and a set of XApp tools that make it a truly unique package that caters to both beginners and power users alike.

For years it followed a familiar rhythm of new releases roughly every six months, with girls' names being used in alphabetical order from Ada in 2006 to Zena in 22.3 for naming individual releases.

That flow has since shifted as the developers settle on a longer development cycle, which sees them move to one major release per year with no point releases in between.

Version numbering and naming also change with this, as the number now points to the year ahead, taking users from 27 to 28 to 29, and boys' names take over from the girls' names.

Linux Mint 27 "Adrien" is the first release under this new strategy. It is planned for Christmas, builds on Ubuntu 26.04 LTS and comes with a five-year support period.

So, with this feature rundown, we are following the monthly news posts since January and are sticking to changes that are new in Linux Mint 27.

1. Wayland is Here

Cinnamon's Wayland session is no longer experimental. Both X11 and Wayland are fully supported in the next Cinnamon release, with Wayland nearly level with X11. This makes the flagship experience of Linux Mint Wayland-ready, bringing it up to the level of other mainstream distros like Fedora and Ubuntu.

Much of the groundwork revolves around everyday interactions. Windows and menus map properly, focus stealing is prevented, and crashes across Cinnamon, Muffin, cinnamon-session, and Xwayland are fixed.

Multi-monitor setups and KVM switches are covered too. 🔥

Graphics get attention as well. Hardware acceleration is supported, including GBM over EGL for NVIDIA cards, while HiDPI scaling reaches cursors and Chromium-based apps such as Slack and VS Code.

The final gap in ensuring Wayland support was the screensaver.

The new one is rendered by the compositor and handled by Cinnamon itself in place of a separate cinnamon-screensaver process, and it works on both Wayland and X11. The old screensaver stays as a fallback for one release before it goes away.

2. Desktop Upgrades

Nemo Changes

The default file manager, Nemo, gets a preview pane that can be opened with F7 or from the status bar. It handles audio files, pictures, and videos and can show metadata details such as the capture date and the camera used.

Similarly, Interactive Search, the feature that kicks in when you open a folder and start typing, becomes a filtered view that narrows the list as you type, in place of jumping to the first match and leaving you to navigate through the rest.

Screenshot Tool

A dedicated screenshot tool arrives for Cinnamon. It can capture a window with or without its shadow, and it can capture all monitors or just one.

You can also crop a shot before copying or saving it, and the tool deals with shadows differently for windows that draw their own titlebars and windows that rely on the system's.

Library Redesign

The Library is how the XApp app Thingy appears on the desktop, and it gets a redesign that groups recent documents by type instead of by the application that opened them.

A sidebar holds predefined groups that you can show, hide, and reorder. You can also add custom groups based on file type and limit them to a single directory if you like. Single items can be removed, and the whole recent documents list can be cleared at once.

3. System and Installer Changes

A Common Installer

Linux Mint switches to the live installer that Linux Mint Debian Edition (LMDE) already uses, replacing Ubiquity. The new installer handles OEM installs, BIOS and EFI systems, Secure Boot, and LVM and LUKS setups.

Hibernation Returns

After Ubuntu dropped it, hibernation has made a comeback in Linux Mint 27 through a new "Sleep Modes" page in the System Administration tool that makes it easy to set up again.

Though it still needs a large swap to work. ⚙️

That page also adds two more options. Hybrid Sleep suspends the machine while also writing the session to disk, and Suspend-then-Hibernate starts out as a normal suspend before it hibernates once the battery reaches a critical level.

Kernel Management

This has moved out of the Update Manager and into the System Administration tool on both Mint and LMDE, and it is built around kernel series.

You track a series, choose how many older kernels to keep, and let a weekly cleanup remove the rest, or run a manual cleanup whenever you like, while a History page keeps a record of what happened.

Custom, OEM, and Real-Time are some of the kernel series' that can be marked as tracked, while untracked series are removed at the next cleanup round unless their kernels are exempted.

User Settings

Account settings also move into the System Administration tool, with it taking over the handling of users and their account details. That includes home directory encryption at account creation, webcam selfies with a mirror option, and HiDPI avatars.

4. Core App Additions

Superlight

Superlight is the new system monitor. Its graphs track CPU load, memory and swap use, GPU, network traffic, disk activity, temperatures, and fan speed, while a dedicated "Battery" page shows status and charge history.

An app page and a process list sit alongside them. Columns are configurable and sortable, and the choices are carried across sessions. The "Storage" page shows a usage breakdown and gives quick access to options like emptying the trash, clearing the cache, and accessing the file system.

Velocitty

The new default terminal for Linux Mint, called Velocitty, offers a tab overview in the style of Ghostty and Ptyxis, color palettes, and tabs that can be reordered, moved, and detached.

Paths can be shortened, and tab colors can flag root sessions, remote sessions, and running processes. Custom tabs can also be assigned a name, color, or command.

Xepub

If you prefer reading EPUB books, then you will love this one. Linux Mint 27 offers a dedicated app called Xepub that supports chapter navigation, search, bookmarks, and zoom.

You also get a choice of fonts, and it remembers where you stopped reading.

Selecting a word opens a menu that can be used to search the dictionary, Wikipedia, or a translator, and the search engine to be used is configurable.

Calendar and Clocks

Calendar is a new app from the Clockenstein project, and it connects to local calendars, Google Calendar, and CalDAV servers like Nextcloud. You can use it to create and edit events, and any sync errors on remote calendars are shown, with unreachable ones turning read-only.

Reminders can be dismissed or snoozed, and a separate daemon and notification agent deliver them while the app is not running, with the Cinnamon clock applet being compatible with those to keep events visible in the panel.

Clocks is the part that was still under development when we last heard about it. The September update previews it as an Android-inspired alarm app from the same project, with recurring alarms and custom sounds.

Unfold

Handling archives gets simpler with Unfold, a very portable archive manager that lets you view, create, and extract archives while using 7zip and 7zip-RAR to do the actual work behind the scenes.

Encrypted and password-protected archives are supported as well, and the app is released under GPLv3+.

5. Misc Changes

Before we wrap up, here are a few smaller changes that could be of interest to you:

• WPA3 and OWE support in Cinnamon.
• Fingerprint login in Slick Greeter without needing to press Enter.
• A new "Sensors" page in System Reports with fan speeds and temperatures that refresh automatically.
• A Mint-Y refresh with better dark colors and contrast, plus rounded frames, scrolled areas, treeviews, and listviews.

Can't Wait to Get Started?

Well, you will have to wait, oops. Linux Mint 27 is supposed to drop sometime around December 24 on Christmas Eve and will be offered in three officially supported variants:

• Cinnamon (the flagship ⭐)
• MATE
• Xfce

Seeing that this article has gone live before the release is available, the download portal linked above only features the earlier releases of Linux Mint. Though, if you have the skills, you could build the release from source.

Asahi Linux is the reason you can run Linux on an Apple Silicon Mac today. For years, the project put most of its own code under the MIT license so that anyone, including the BSDs, could reuse it as needed.

That is changing because Asahi Linux is tired of code laundering. They have announced that Asahi is moving all of its repositories to the GNU General Public License version 2 or later. Going forward, all Asahi Linux code will be under strong copyleft licenses only.

This is because people were taking in-development code straight out of their personal forks, publishing it without credit, or rewriting it with AI to hide its origin for "replacing the project and aggressively monetising Linux on Apple Silicon."

What is code laundering?

Code laundering (also called license laundering) is the practice of taking someone else's code and passing it through a process that hides where it came from, so you can use it without honoring the original license or crediting the author.

Before AI, this usually meant manually rewriting code line by line, renaming variables and stuff, and shuffling functions around until it no longer looked like the original.

LLMs made it much more easy and cheaper of course.

Learning from history

The move has come after the Asahi team was made aware "some time ago" of multiple attempts to exploit its permissive licensing to hijack its work.

In the most extreme cases, feature code that was still under development was lifted from developers' personal forks before it was even publicly released. The report points out that the MIT license "explicitly allows this, and has no carveouts to protect against bad faith actors."

Asahi didn't name anyone in the report. It also didn't spell out how existing contributions will be relicensed or whether every past contributor agreed to it.

Asahi draws parallels with WINE's situation two decades ago. WINE, the compatibility layer that lets you run Windows apps on Linux, was originally MIT-licensed too. But in the late 90s and early 2000s, companies took the code, built proprietary forks on top of it, and sold them commercially. So, in 2002, the WINE team relicensed to the LGPL to discourage proprietary forks and push companies to contribute upstream.

Those who do not learn from history are doomed to repeat it

Asahi has been upset lately

Asahi didn't name anyone so we are left with guesswork on who they are targeting with this license move.

There have been at least two public forks of Asahi's work in recent months, both tied to AI-assisted development. In case you did not know, Asahi Linux forbids AI generated code contribution.

The first is Aurora Silicon and its Vulkan driver, Borealis. Borealis is a fork of Asahi's HoneyKrisp, the Vulkan driver for Apple GPUs. Aurora Silicon submitted a pull request to register its own Vulkan driver ID with Khronos (VK_DRIVER_ID_AURORA_BOREALIS), says the driver runs on M1 to M5 and the A18 Pro, and keeps it in its own Mesa fork rather than upstream.

The second is Gravity Linux, a Fedora-based distro for Apple Silicon. It was forked from Asahi over the disagreement about LLM use.

To be fair to Gravity, it is open source, it isn't selling anything, and it credits its AI use openly. It even says it plans to sunset the distro once its driver work lands upstream.

It isn't the kind of bad actor the progress report describes. But it does represent the "permissive fork that embraces AI" approach that Asahi has consistently pushed back against.

Why this license change matters

As far as I can tell, this is the first time a major open source project has moved to copyleft specifically because of AI-assisted code laundering. Projects have relicensed before, but usually to stop cloud providers from reselling their work as a service. This one is specifically about AI.

I wouldn't be surprised if more hardware enablement and reverse-engineering projects follow. Making the first changes takes a lot of time and effort but they are easier to "rewrite" once someone has done the hard part.

Will the GPL actually help?

If the problem is people who already ignore the MIT license, why would they respect the GPL? After all, GPL cannot stop a determined bad actor. Nothing does, actually.

Stripping attribution from MIT code already breaks the MIT license, since keeping the copyright notice is its one real condition.

What the GPL changes is the commercial benefit. Under MIT, someone can take Asahi's code, close it up, and sell a product, and that's perfectly legal as long as the notice stays in.

Under the GPL, anyone who distributes a modified version has to release the source under the same license. You can't legally build a closed, paid Apple Silicon distro on top of GPL code. And if a product's source has to be public, the copying is easier to spot.

Not everyone is happy

The move has drawn criticism, particularly from the BSD side, which was one of the main reasons Asahi chose MIT in the first place. OpenBSD's Apple Silicon support relies on Asahi's m1n1 bootloader.

OpenBSD developer Bryan Steele wrote on X:

"Sad to see Asahi out there burning bridges for no reason.. relicensing m1n1 and drivers under GPL2-or-later, seemingly without the consent of all prior contributors, under the guise of "defending against AI bad actors", seems entirely disingenuous and misguided."

Wrapping up

Asahi is betting that copyleft license provides a better defense in this changed world where anyone can launder code with a prompt.

It may not stop the bad actors from rewriting its code, but at least it reduces the chances of a closed commercial fork.

It would be interesting to see whether the BSD community finds a way to keep working with Asahi under the new terms.

Suggested Read 📖 These Linux Distros Can Run on Apple Silicon Mac