Omarchy has been on something of a roll now. Funding for its foundation has been climbing fast, from an $8 million launch to $10 million a few weeks later, and now crossing $13 million with the most recent pledge.

What's drawn my attention are the first two Corporate Patrons, 1Password and 37signals, both of whom have decided to pitch $100,000 a year for the next three years. Now, 37signals I get; DHH is heavily involved there, but 1Password was a surprise entry.

Naturally, not everyone's onboard with their pledge.

Signs of disagreement

This has not gone well with employees working at 1Password, with company leadership having to take certain damage control measures to assure their staff.

An internal Slack message has surfaced (courtesy of The Verge), which shows cofounder Roustem Karimov downplaying the criticism, telling team members that:

people have different personal opinions. You believe in your heart that DHH is evil, that you have the moral high ground, and that nothing will change your mind.

However, not everyone believes that. It is not fair to claim a monopoly and ostracize team members who might disagree with you. There are people who are afraid to speak up simply because they will be personally attacked.

CEO David Faugno responded on a different note, telling employees the company doesn't endorse DHH's views, while also noting that Omarchy is the second most used Linux distribution among 1Password's own users.

Why the backlash?

DHH is known to be someone who firmly falls on the "right" side of the political spectrum, someone who doesn't shy away from putting his opinions in public view.

He regularly posts blogs that show where he stands on certain societal issues, and some of his recent writeups are what's fueling this particular backlash.

His recent July post, titled "Wolves, sheep, and gypsies," compares Denmark's wolf population with the Romani people camping in Copenhagen parks, where he argues: "When gypsies take over public spaces, you deport them."

Then there's "As I remember London" from September 2025, where he laments about London losing its native Brit majority and a nod of approval for a Tommy Robinson march.

Of course, these aren't the only factors behind his disapproval, but you get the gist of it, right?

Users could jump ship

If you search for the terms "1Password" and "Omarchy" right now, you are bound to run into the many comments made by disgruntled 1Password users on Reddit and Hacker News.

They are calling the choice tone-deaf, given how many smaller open source projects could've benefited from the money. Some are already jumping to alternatives like Bitwarden, while others suggest self-hosting Vaultwarden or going with KeePass and its forks instead.

I see the issue compounding too. Back in February, 1Password raised subscription prices, taking effect at renewals from March 27. Individual plans went up 33 percent from $35.88 to $47.88 a year, and family plans went up 20 percent from $59.88 to $71.88.

Its community forum already has a long thread full of longtime users saying they were leaving over it. If 1Password continues playing with its users' trust like this, who knows what kind of exodus it will see next?

Debian has gotten some good press thanks to Federico Vaga and Nikos Tsipinakis, who presented a talk at MiniDebConf Winterthur this past weekend.

The two CERN engineers laid out why the organization is moving its particle accelerator control computers onto Debian 13, with more than 2,200 industrial computers and embedded systems set to run it by the end of 2026.

Though this doesn't mean that the whole CERN IT infrastructure is being moved. The migration is specific to the accelerator control layer—the front-end computers that talk directly to the equipment running the beam.

Why move?

CERN ran a risk analysis in the second quarter of 2023, putting a number on what staying in the Red Hat ecosystem would cost them. The estimate came out to roughly CHF 5.4 million, with around 11 hardware boards needing a full redesign, plus hiring two electronic engineers, two software engineers, and two technicians just to handle it.

Racks would need reorganizing and rewiring, and most systems would be affected during commissioning. Even assuming bug-free replacements, CERN's own estimate put the odds of success at an optimistic 20 percent.

The reason for all that was a compiler flag. When Red Hat builds RHEL, it picks a minimum CPU generation the software will run on. RHEL 9's cutoff already excludes CERN's oldest boards, the kind running chips like Intel's old Core 2 Duo, which comprises about 47 percent of their fleet.

RHEL 10 raises that cutoff again, and this time even newer Ivy Bridge-generation boards, another 17 percent, fall on the wrong side of it.

The release strategy

CERN weighed two paths for keeping Debian aligned with its accelerator schedule. Plan A develops on Bookworm through 2026, deploys Trixie as the long-term support release from 2026 to 2030, then moves to Debian 15 "Duke."

Plan B skips that last jump, staying on Trixie under extended long-term support (ELTS) out to 2033 instead. A third option, running Bookworm itself under extended support the whole way, got ruled out.

Either path depends on Debian's LTS and ELTS programs staying healthy, which is part of why CERN has started sponsoring Freexian, the Debian-focused services company behind those support programs.

How it will be built

The old setup ran on NFS and tftpd, a bootserver model that traces back to around 2005.

The new one decouples the bootloader, kernel, init RAM disk, and userspace into separate pieces, then rebuilds delivery around Kubernetes.

A controller compares the configuration CERN wants against what's actually deployed and redeploys automatically when the two fall out of sync, instead of relying on someone pushing files over NFS by hand.

None of this touches CERN's data centers; this is specifically about the machines that keep the accelerator itself running. By the time the last quarter of 2026 closes out, CERN expects to have completed the migration.

Gaming on Linux keeps getting better, Wine and the Proton compatibility layer keep opening up more of the Windows catalog, and community projects continue filling the gaps that big publishers leave behind.

Take Xodus, for instance. The open source effort is looking to bring Xbox and Game Pass titles to Linux. Even though there are no playable builds yet, its direction says a lot about where Linux gaming is headed.

The money aspect of the equation is catching up too. While not strictly a gaming-focused move, the recently launched Omacom Foundation, which backs Omarchy and the open source projects it relies on, already has $12 million in funding.

Then you have Ubisoft, fresh off the success of Assassin's Creed Black Flag Resynced (a mouthful, I know), who are going the other way, breaking something that worked just fine before their meddling.

They kicked out Linux

Ubisoft has announced that For Honor is leaving the Linux platform. Starting September 10, the game will no longer be playable on Linux desktops or the Steam Deck. The removal of support lands the same day its Ranked Dominion and Ranked Duel modes go live.

They are calling it a move intended to safeguard players by introducing additional measures for ranked play to ensure a fair competitive environment, adding that they "cannot meaningfully deploy protection on this platform."

Ubisoft goes on to say that they:

will continue to investigate different ways to improve our anti-cheat with the possibility of reopening the platform in the future.

Which is PR speak for. This has been put on the back burner and is not a priority for us.

That makes me think that they would rather push out new pricey skins than support a minority portion of their player base.

The same old shenanigans

If you have been following Linux gaming-focused news, you might already know the script such big game publishers follow.

Roblox blocked Wine users in 2023, while GTA V Online barred Linux players in 2024, and Apex Legends kicked out Steam Deck owners the same year. For Honor is the newest name on a list that keeps growing, and the provided rationale never really changes.

What makes this one different is that Ubisoft once opened the same door it is closing.

The company enabled Easy Anti-Cheat support for the Steam Deck back in July 2023, and the game has been playable there since. Now it is closing that door, saying protection cannot be meaningfully deployed on Linux, even though Easy Anti-Cheat has supported the platform for years.

Heck, even Electronic Arts seemed to be coming around earlier this year when a job listing outlining a path for its Javelin anti-cheat to support Linux and Proton appeared.

Though it is unclear how the new owners and leadership will steer the company, that plan may not survive in the end. Meanwhile, the handheld market keeps getting stronger, with SteamOS spreading to more devices than just the Deck.

And you know what's the more troubling thing here?

Cheaters always find a way in, on Windows or anywhere else.

If your Raspberry Pi can no longer connect to Wi-Fi because the SSID or password changed, you can update the Wi-Fi configuration directly from its SD card. This helps when you cannot connect the device to a monitor and keyboard.

That's what I did. I had switched to a different internet provider, and that changed the router, its SSID and the password. I was using a Raspberry Pi Pico W in my Pi Dog. Connecting it to a screen and keyboard would have been a pain.

The easier option was to just take out the SD card and edit the WiFi details directly there. The entire process is composed of the following step:

• Insert the SD card with Raspberry Pi OS on it on your (Linux) computer
• Go to the mounted rootfs partition and go to /etc/NetworkManager/system-connections/ location
• Here, either create or edit .nmconnection file and enter the wifi name (SSID) and password

Let me show that in detail for you.

Step 1: Connect the SD card to your computer

Shut down the Raspberry Pi, remove the microSD card, and insert it into your Linux laptop or desktop using an SD card reader.

Raspberry Pi OS usually has two partitions:

bootfs
rootfs

The Wi-Fi configuration is stored inside the rootfs partition.

Step 2. Find the SD card mount points

Open a terminal and run:

lsblk -f

You should see something similar to:

sda
├─sda1   vfat     FAT32 bootfs B2F0-82D2                             438.1M    13% /run/media/abhishek/bootfs
└─sda2   ext4     1.0   rootfs 15f4c6be-1102-4331-9904-f78e78afd1fd   21.9G    18% /run/media/abhishek/rootfs

If you only see bootfs, mounted, you can mount rootfs from the file explorer.

As you can see, for me, the Raspberry Pi root filesystem is available at:

/run/media/abhishek/rootfs

You can also confirm it with:

mount | grep rootfs
/dev/sda2 on /run/media/abhishek/rootfs type ext4 (rw,nosuid,nodev,relatime,errors=remount-ro,uhelper=udisks2

Step 3. Check the existing Wi-Fi configuration

Recent Raspberry Pi OS releases store NetworkManager connection profiles here:

/etc/NetworkManager/system-connections/

Since we are accessing Raspberry Pi OS through its SD card, run:

sudo ls -la /run/media/$USER/rootfs/etc/NetworkManager/system-connections/

If you already see a .nmconnection file, you can edit that file.

No nmconnection file? Create one

If the directory is empty, create a new Wi-Fi profile.

touch /run/media/$USER/rootfs/etc/NetworkManager/system-connections/home-wifi.nmconnection

And then use a terminal-based text editor like Nano or even Micro and add the following details:

[connection]
id=home-wifi
type=wifi
interface-name=wlan0
autoconnect=true

[wifi]
mode=infrastructure
ssid=YOUR-WIFI-SSID

[wifi-security]
key-mgmt=wpa-psk
psk=YOUR-WIFI-PASSWORD

[ipv4]
method=auto

[ipv6]
method=auto
EOF

Step 4. Set the correct file permissions

Your work is almost done. You just need to set the correct file permissions. That's because NetworkManager expects connection files to be readable only by root.

To give it the correct file permissions, use the following command:

sudo chmod 600 /run/media/$USER/rootfs/etc/NetworkManager/system-connections/home-wifi.nmconnection

Verify the permissions:

sudo ls -l /run/media/$USER/rootfs/etc/NetworkManager/system-connections/

The file should look something like:

-rw------- 1 root root ... home-wifi.nmconnection

Step 5. Flush the changes to the SD card before removing

This is an important step. After making all the changes, run this command:

sync

The sync command makes sure any pending writes cached in memory are actually written to the SD card.

And then you can either unmount from the file explorer or use the umount command for both partitions:

sudo umount /run/media/$USER/rootfs
sudo umount /run/media/$USER/bootfs

You can now safely remove the SD card.

Step 6. Verifying the connection

Insert the card back into the Raspberry Pi and power it on. After it boots, check the devices connected on your network to find the Raspberry Pi's IP address.

The simplest way would be to log in to your router and see the connected devices.

Otherwise, Angry IP Scanner is a good graphical tool for scanning your network. Command line tool nmap also does the job:

sudo nmap -sn 192.168.0.0/24

Normally, it should show your Raspberry Pi's hostname. You can also try to run scan before turning on Pi and a few minutes after turning it on. This should show the new devices that turn up between the two scans.

Finally, check the connectivity with:

ping RASPBERRY_PI_IP

Or, if SSH is enabled on your Raspberry Pi, connect to it:

ssh username@RASPBERRY_PI_IP

By the way, if you have forgotten the user password of your Raspberry Pi, that can be easily reset too.

Pironman 5 Case With Tower Cooler and Fan

This dope Raspberry Pi 5 case has a tower cooler and dual RGB fans to keep the device cool. It also extends your Pi 5 with M.2 SSD slot and 2 standard HDMI ports.

Explore Pironman 5

Enjoy your Raspberry Pi 😄

My history with Vanilla OS goes back to 2022, when I covered its debut stable release, just a month after I landed this gig at It's FOSS (yapping about open source and Linux).

Since then, my focus has moved to other distros and governance-focused developments, though I did watch it swap its Ubuntu base for a Debian Sid one and, from a distance, then go through an Orchid-themed overhaul in 2024.

I never got around to actually running it later on (or even reviewing it), but that's how things go, I guess.

If you haven't come across it before, Vanilla OS is an immutable, atomic distro built around ABRoot, with Apx facilitating package installation inside containers instead of touching the base system directly.

What's new in Reunion?

Well, if you are someone who is running an ARM64-based setup, be it a single-board computer, laptop, or a fully-fledged workstation that boots UEFI, then you will be glad to know that Vanilla OS now has proper support for the ARM64 platform.

Every package now builds for both ARM64 and x86_64, and the team isn't treating ARM64 as the lesser platform; functionality-wise, it's supposed to match the other one for one.

And, with the exception of NVIDIA builds, most images of Vanilla OS are now bit-for-bit reproducible. That means you can rebuild an image yourself and check it matches what the project ships instead of just blindly trusting what their build infrastructure pumped out.

The desktop-facing upgrades

Powering the release is Linux kernel 7.1.3, which is accompanied by GNOME 50. While we have already talked a fair bit about this desktop environment release across other distro releases, here's a refresher for you.

GDM now runs Wayland only, following through on the X11 removal GNOME backed away earlier. Fractional scaling also graduates out of its experimental flag this time around.

You get a bunch of new default apps too. Ptyxis takes over from Black Box, and its container support is the reason why you should keep an eye out for it.

Papers steps in for Evince with a noticeably more modern PDF reading experience, and Resources replaces GNOME System Monitor, giving you a clean look at what your apps and hardware are actually doing.

Underneath, Apx and VSO now share the same SDK, so config handling, permissions, and logging finally work the same way across both tools instead of each doing its own thing.

Another new addition is Vanilla Continuity, a snapshot-based backup and restore tool that works alongside ABRoot. If configured properly, it can save your home directory, whatever Flatpaks you have installed, and ABRoot's own metadata.

Where those snapshots land is up to you. A local directory works by default, but you can also point Continuity at an external drive it encrypts with LUKS2, or send backups to a remote server over SFTP, FTP, or NFS.

There's also the push for community-built images through a Vanilla-Flavors org on GitHub, which currently hosts the china-image flavor, which caters to people living in mainland China.

If you want to build and maintain your own flavor, the Vanilla OS devs are open to hearing from you.

Download Vanilla OS 3

Images for Vanilla OS 3 'Reunion' are live on the official website right now. Though keep in mind that the download page will make you pick an architecture, then a donation amount.

You can enter "0" as the amount to get the free download link, or you could enter a donation amount greater than zero to directly support the project and then download the ISO.

At the time of writing, donations were not working via the PayPal integration the project has implemented.

If you're already on Vanilla OS 2 Orchid (v1.4.0), this release will arrive as a normal system upgrade. If it's taking too much time, you can run this to get the release:

abroot upgrade

Auke Kok spent eight years working on Clear Linux OS at Intel. Now he thinks he knows exactly what broke it, and he's trying to fix that by himself.

In a LinkedIn post this week, Auke, now a Linux Distribution Architect at Verge.io, introduced The Ur Project. It's a from-scratch Linux distribution build system he's been putting together in whatever spare time he has.

"This project is not a corporate effort," he writes, and as far as I can see, no company, or even foundation for that matter, is backing it. Besides, as you might remember, Intel discontinued Clear Linux OS last year.

Where Clear Linux fell short

Auke has been writing regularly on The Ur Project website about what he thinks went wrong the first time. The biggest one, in his own words, was that they "made it way too hard for the community to contribute."

Simple things, like testing his own code change or carrying out regular incremental development, were needlessly painful, he says.

The release process didn't help either. Mismatched dashboards and a combination of old and new tooling were bolted together just to track daily updates.

Keeping pace with an assortment of daily updates meant stitching together a growing pile of disconnected tools and interfaces, all running on CDN and hardware resources a solo project at this stage will never have access to.

A build system, not a distro

The Ur Project isn't meant to be some shiny new atomic distro, at least not yet. The priority right now is to work on the machinery behind it, not offer a polished product.

Everything runs on Rust, on both the part that builds packages and the part that installs them, sharing the same underlying code.

Packages are defined through what Auke calls recipes, declarative TOML files that describe what a package needs without prescribing exactly how to build it, an idea he compares to "your grandmother's collection of cooking recipes."

Build steps aren't scripted by hand. Instead, a set of probes inspects the source and decides how to compile it. Repositories run on TUF to keep updates signed and verifiable, with content-addressed storage handling deduplication so identical files across packages only get stored once.

Every package carries provenance data documenting how it was built, and Auke says the system is designed so builds can be reproduced bit for bit. The same underlying system can output an OCI container, a live USB image, a cloud image, or a plain installed system.

AI is involved

LLMs are only tasked with writing test suites and build definitions for specific package types.

The build system itself stays deterministic. It runs without any AI agent present, and MCP integrations were deliberately kept out of routine packaging work.

For now, what's available is the website, Auke's regular blog posts on The Ur Project, and the Forgejo instance that hosts the code.

He is currently in the process of figuring out the hosting, letting people actually pull images and repositories. That will take a few more days before it's ready.

Once that's sorted, he plans to publish bootable images for both cloud deployments and live USB use.

In a very low-key manner, Firefox 155 has arrived as the latest stable release for the popular privacy-focused web browser.

While there are a range of refinements with this release, we will be focusing on two technical ones that change how Firefox handles webpage loading. Happy Eyeballs v3 and QUIC v2 support for HTTP/3.

Happy Eyeballs v3

Instead of testing one path and hoping it works, Firefox now tries a handful of ways to reach a website side by side and goes with whichever one answers first. There's nothing to turn on and nothing to configure. You'll just notice fewer pages that sit there loading while Firefox figures out how to actually get through to the site.

You see, every time Firefox loads a page, it first has to decide how to reach the server behind it.

And there's often more than one option. An IPv4 address, an IPv6 address, sometimes both, and depending on the site, a choice between HTTP/2 and the newer HTTP/3. Older Firefox versions picked one option and waited to see if it worked before trying the next, so a slow or broken first guess meant the page just sat there without loading.

Happy Eyeballs v3 is what changes that. Firefox fires off connection attempts down several of these paths at once and runs with whichever one lands first, instead of committing to a single guess at the beginning.

Officially, Happy Eyeballs v3 is still an IETF draft rather than a finished RFC, authored by engineers at Apple and Google. Mozilla's own developer notes for this release describe the feature as currently supported on some platforms only, without naming which ones.

QUIC v2 for HTTP/3

QUIC is one of the pieces of tech behind how fast a website loads today, and Firefox 155 adds support for a slightly newer version of it.

What actually matters isn't the new version itself. It's that Firefox can switch to it instantly, skipping the back and forth that normally happens when two computers don't immediately agree on which version to use.

That back-and-forth used to cost an extra trip across the internet and back. Now it doesn't.

HTTP/3 is the newest version of the protocol that moves data between your browser and a website, and runs on QUIC rather than the older TCP. Setting up a secure connection over QUIC takes one round trip instead of several, which is a big part of why HTTP/3 feels quicker on shaky connections.

This is the right time to switch

Might I add, this is the best time to switch to Firefox. uBlock Origin's full version has been broken on Chrome since Google disabled Manifest V2 back in July 2025, and yesterday Google kicked it out from the Chrome Web Store entirely.

Now, only uBlock Origin Lite remains.

Of course Firefox was never really affected by what Google did, consistently supporting the full uBlock Origin experience without any nerfs in its ad-blocking abilities. You can get its latest release via the Firefox Add-ons portal.

That is not all; the 155 release also has many other new additions, like the Smart Window rollout (limited to users residing in the US, Canada, or France), blocked tracker counts in the address bar, and the ability to reorder containers.

Any of which could cajole you further into switching away from Chrome or any other Chromium-based browser.

Get Firefox 155

Binaries for Firefox 155 are available already, though the official download page hasn't caught up with the new build yet; it should get there shortly. If you can't wait and would rather get this release now, although I would advise waiting.

Mozilla's FTP mirror already has the 155.0 builds for Linux, Windows, and macOS. Remember to select the correct platform and language options to get the package relevant for your computer's specs.

On my Ubuntu setup, I went with: linux-x86_64 > en-US > firefox-155.0.deb

Zack Bartel is a developer who switched to Linux full time and hasn't really looked back at his earlier daily driver, macOS, as something he would return to. There is, however, an exception.

Continuity.

That is Apple's catch-all term for how iPhone, iPad, and Mac talk to each other. It covers a wide range of small conveniences. Copy something on the iPhone and paste it on the Mac. Start an email on one device and finish it on another. Take a call from the Mac even when the phone is across the room.

But, to no one's surprise, there's a gap in its support. It doesn't officially work on Linux. So Zack built Tether, an open source project that replicates pieces of that experience on Linux without requiring a Mac device.

What does it handle?

Considering an app whose first release was just a few months ago in May, Tether covers a lot of ground already. Clipboard sync, file transfer, OTP autofill, iMessage, SMS, notifications, and contact sync are all live right now, with some of them being in the beta stage of availability.

A background daemon called tetherd handles inter-device communication on Linux, talking to a CLI and a native GTK4 app for pairing. A SwiftUI iPhone app finds the daemon automatically over Bonjour.

Linux has no supported way into iMessage or SMS, and Zack ruled out the common workaround of proxying everything through an actual Mac.

ancs4linux and BlueFerry supplied enough Bluetooth groundwork for him to build his own path, though both are GPL-licensed, so a clean room C++ rewrite kept Tether licensed under MIT.

To secure it all, he built the pairing process on mTLS, so the iPhone and the Linux daemon each have to prove who they are before anything connects. Security sweeps run regularly on the codebase too via Anthropic's Opus and Fable AI models.

Before you say that KDE Connect already does a lot of what Tether does, know that Zack is aware of its existence and calls it great, but points out that the app doesn't cater to the use cases he had in mind.

Get it running

For Arch Linux users, Tether is available via the AUR; you can get it by running the following command:

yay -S tether

If you don't know what Yay is, it is an AUR helper that makes the process of getting packages from the Arch User Repository easy. Though you do have to take note of the risks associated with using the community-run platform.

For other distros, the project's GitHub repo carries .deb and .rpm packages built for recent releases.

On your Apple iPhone, you will have to download the Tether - Linux Companion app from the App Store, which needs your device to be running at least iOS 26.1 or later.

There's also two extensions that you will need if you want OTP autofill to work properly.

The Firefox extension drops codes into login forms the moment they arrive, and the Thunderbird extension watches your inbox for anything that looks like a verification code and hands it off to the Tether app running on your Linux machine.

If the past few weeks are any indication, more and more open source projects are drawing lines around how much AI they will let into their workflow and codebases before it's too much.

Debian just voted to allow generative AI in project contributions, while Rust adopted a tiered policy that keeps AI mostly out of the actual code. Both of these projects tackled the same underlying question.

And now, Linux, one of the biggest open source projects around, is getting hammered by scrapers, mostly AI-powered ones, sending the same repetitive requests over and over and burning through compute capacity.

Most of it has nothing to do with anyone actually writing code, btw. 🙃

This is absurd

Konstantin Ryabitsev of the Linux Foundation has put out numbers behind a complaint he has been voicing for a while now. Fourteen of git.kernel.org's 90 CPU cores, spread across five nodes, spend every second of every day turning commits into HTML pages.

On the outside, you might think, What's the problem with this? Every commit in Linux's history sits out in the open, is free to clone, and predates the wave of AI tools now scraping it.

You see, all of those characteristics are precisely what makes the repository a "goldmine of learning data." It houses the mainline kernel tree, every stable release branch going back years, dozens of subsystem maintainer trees, and even pre-git history from the BitKeeper era.

What's actually absurd is the way these clankers are going about the scraping business.

Konstantin ran the numbers on it and found out that a regular clone of linux.git, going through its whole commit history locally, takes about 200 CPU-seconds of server time. Scraping the same 1.48 million commits through cgit's individual pages instead eats up 280 CPU-hours.

Do that same scrape across every one of the 922 forks hosted on the server, and the total balloons to 258,160 CPU-hours, something like 4.6 million times more expensive than a single clone.

All that for a worse copy of data that was already free to grab in the first place. 🤭

And that's before counting the separate URLs cgit hands out for every patch, diff, and plain-text view of each commit too, which is what pushes the number of pages a single fork exposes into the quadrillions.

Blocking them turned into an arms race. Fail2Ban and IP bans worked until bots spread across whole subnets. ASN blocks worked too, until millions of residential and mobile IPs took over instead.

Anubis came next, being erected as a proof-of-work wall bots had to solve before getting through. It worked for a while until the clankers started solving increasing levels of challenge difficulty.

Wrapping up the writeup, Konstantin notes that:

However, you should know that out of the total of the 90 cores across 5 geo distributed nodes, there are 14-16 cores that are constantly doing nothing but rendering commits for scrapers.

On average, that's 20% of our entire capacity — except the swarms descend in waves and the actual graph is a lot more spiky than a 20% flatline.

Further speculating that when the AI bubble bursts, the project will quickly see a substantial decrease in the amount of "entities" (his labeling of the scrapers), trying to feed git.kernel.org to their models.

Though he also hopes they "smarten up" and stop scraping their data in the "dumbest way possible."

If you did not know who Konstantin is, he is the director of IT infrastructure security for the Linux Foundation and one of the sysadmins who keeps kernel.org up and running.

AI is inherently greedy

None of this compute waste is unique to git.kernel.org. It is just the clearest example we have got so far.

Feeding a model means burning through cycles on tasks that could be done in a fraction of the time, and right now nobody building these systems seems particularly bothered by that math.

The only approach these systems seem to know is the one your average billionaire runs on, say a fictional one like Carter Pewterschmidt. Already has more than enough. Still wants more. Does not particularly care how it gets there.

Stay on that path long enough, and the greed stops paying for itself. Plenty of companies are already finding out the hard way, and the numbers back it up too.

If you are a content creator in 2026, then I am sure you end up juggling through a stack of apps and browser-based tools for the same handful of jobs. One resizes an image, another handles format conversion, a third takes care of watermarks, and a website somewhere tells you the exact pixel dimensions Instagram, X, or YouTube expects.

None of those steps take long on their own, but doing them one after another for every single post does add up.

Image Toolbox is built to make all that easier. It's an Android app that offers resizing, format conversion, watermarking, and platform-accurate exports under one roof.

What can it do for you?

Working under the name "T8RIN," Malik Mukhametzyanov has come up with Image Toolbox, offering it for free without any advertisements inside the app.

For you, as a content creator, the tools that matter the most are batch processing, format conversion, the 61 built-in export presets sized for platforms like Instagram, X, and YouTube, background removal, and watermarking.

The latter, background removal, comes in both automatic and manual modes, with a choice of models including MlKit, U2Net, RMBG, and BiRefNet.

It's built in Kotlin, uses Image Toolbox Libs for shared components, and Coil for image loading. The whole project is released under the Apache License 2.0.

I used it for a while

When you first launch it, Image Toolbox asks you for the usual set of notification and file access-related permissions, along with a dialog that lets you opt out of update checks.

Before I started editing images, I first tweaked the app to my liking.

In the sidebar menu, I noticed that the app had pulled the purple straight off my phone's Dynamic color setting which was nice, and I then turned on "Amoled mode" for getting a deep black-themed interface.

That was followed by a prompt switching of the app to "Dark" mode instead of "System."

Of course there's a lot more buried in there, like separate categories for Layout, Confetti, Vibration, Screen, Text, and Behavior. I left most of it untouched, as the default configuration was good enough for me.

I kept the focus of my testing limited to a few tool entries (among 100+ tools), as many of the smaller tools kept appearing in the same editor.

First up was Single Edit, which I used to play around with an image's quality, dropping the quality slider to 7 percent, and watching it turn into an oil painting with no realism whatsoever.

During that, the file size for the 3.8 MB photo went down to ~128 kb. I used the three buttons right below the preview to fully rotate and mirror the image horizontally too.

Exif editing also lives in the same area, with parameters like capture date, compression, camera make, and camera model all being editable without me needing to leave the app. I changed the camera make and model values, and the changes were reflected successfully when I saved the image.

Though I did go a bit overboard with tone curves, where I pushed the RGB curve too far, and a simple picture of a cloudy day turned into a fever dream. 😅

It has AI powers

AI Tools was next. I added an image of a clock tower and went with the default/already downloaded U2NetP model that specializes in background removals. I set the output format to be WEBP and clicked on the save button.

It took some time to process the image, but the end result was an image with its background removed. 👇

As you can see above, Image Toolbox gave me a clean output in the requested file format, without any particularly bad artifacting or overcuts. Zoom in, though, and the edges do lack sharpness. I think that's acceptable for a deep learning model that size.

You also have the freedom to opt for heavier, more powerful models if required.

Finally, I checked out "Presets" under the Resize and Convert tool, which basically showed me the same interface as Single Edit. The horizontally scrollable list had a variety of aspect ratios to fit the various social media platforms.

I picked an image from the gallery and then proceeded to look for the right aspect ratio preset.

I ended up using the 4:5 aspect ratio preset for getting an Instagram-ready image. Though I admit it came out a bit squeezed, the image itself didn't really lose its quality or colors.

Track usage

Image Toolbox ships with a "Usage Statistics" page, accessible via the chart-looking button at the top. Its job is to quietly track things like app opens, which tools you reach for most, and how much data you've saved along the way.

Don't worry, nothing gets sent to some data mining server; everything stays on the device.

My statistics showed me a total of 3 app opens, 19 tool opens, around 14 MB in data savings, and WEBP as my most-used format.

And if you prefer multitasking, then you can queue up a batch of images and just check the notification panel for updates instead of keeping the app window open.

Install it on Android

The sheer number of tools and settings in here might overwhelm you, though that does come down to personal preference.

For me, there's plenty in this app I'll probably never touch. It makes me want a stripped-down version, something lighter on storage and memory; that would help it run better on older devices too.

There are two main app store listings for Image Toolbox. On F-Droid, it is under Malik's own name, and on the Play Store, it is under the "ProgiX" publisher name.

Both are the same app, but the F-Droid version is a "FOSS" build that ditches Google-related trackers and dependencies, while the Play Store version is a "Market" build that keeps them, including analytics and crash reporting.

There's no separate paid plan available here either; the lead dev only asks for donations via the sidebar menu.

You can also download APK files for all the variants of Image Toolbox from the Releases page in the project's GitHub repo.

Multikernel Technologies has put out mklinux v7.0-mk2, the first public release of the multikernel Linux tree it's been building since last year.

The basic premise of the project is that one physical server can run several independent Linux kernels side by side, each with its own dedicated CPUs and memory. So, the same box could run a database on one kernel and a GPU training job on another without either kernel touching the other's resources.

Don't worry, it isn't some obscure copy but rather a direct fork of the Linux kernel with a bunch of patches added that give it such a high level of parallel play.

How it works

One kernel runs as the host, controlling the machine's CPUs, memory, and PCI devices. When it spins up a new kernel instance, that instance gets its own dedicated slice of hardware and boots straight onto it using Linux's kexec mechanism, running natively rather than under a hypervisor.

The host tracks all of this through device tree overlays under /sys/fs/multikernel/, and it can shift hardware between running instances while everything stays up.

That's a different setup from both VMs and containers. A VM still puts a hypervisor and a host kernel between the app and the hardware. A container skips the hypervisor, but every tenant on the machine shares one kernel.

Multikernel avoids both problems. Each instance is a full, independent kernel running on its own dedicated cores. Modern NICs and SSDs can hand out separate hardware queues too, so an instance gets exclusive access to a slice of the device, not just a slice of the CPU.

This app-kernel and device-kernel split is the pattern the company documents for the architecture generally, dedicating one instance purely to drivers and I/O so the app-facing kernels never see a device interrupt at all.

This release only supports x86_64 hardware, though Multikernel Technologies says the architecture-specific code is already modular enough for other ports to follow later.

Who's this for?

Multikernel Technologies already sells three products based on this kernel tree, and what they're pitched as says plenty about who this is actually for. Private Cloud is sold as a way to consolidate workloads on bare metal without a hypervisor.

Sandbox is marketed for running AI agents in their own kernel with direct GPU access, and LiveUpdate promises to patch a running kernel without taking the system down.

Every one of those pitches targets infrastructure and SRE teams running their own hardware, not desktop users.

The company's own site names even more targeted use-cases beyond those three, ranging from high core count ARM servers to RISC-V systems built for one kernel per cluster on heterogeneous SoCs.

OEMs looking to bake Multikernel into an embedded device are also welcome.

If you remember, Cong Wang, Multikernel's founder and CEO, first pitched this architecture in an RFC posted to the Linux kernel mailing list back in September 2025. It's great to see how far this project has come.

There's more to it than what I have covered so far. You can read the release announcement to dive into the comprehensive benchmark numbers shared by Cong.

And here's the source code for your perusal. 👇

TDF and Collabora have been at odds since April, when TDF's Membership Committee kicked out every Collabora staffer and partner in one move, with more than 30 people dropped.

That didn't stop Collabora from shipping a range of new releases, with Collabora Office 26.04 being the most recent one, where AI was introduced as an opt-in feature, letting users pick a model and bring their own credentials.

Now, LibreOffice, and by extension, The Document Foundation, has put its rejection of AI at the very forefront with the LibreOffice 26.8 release, going as far as calling it out in the release announcement title itself.

🆕 LibreOffice 26.8: What's New?

There's no AI in LibreOffice 26.8. Nothing about the suite talks to a remote server, and none of it needs an internet connection to work.

This isn't an oversight; it's a promise. TDF says that if your work can't leave the building for legal or privacy reasons, the only way to prove that later (during an audit, for example) is if the software never gave it anywhere to go.

Anyhow, coming back to the editors. 👇

Writer now detects paragraph direction automatically when you open or paste a document, end-of-line spacing follows the paragraph's direction instead of the reading direction, and object resize controls play well with RTL and vertical CJK documents.

It also gets a way to see the baseline grid during editing, and the grid itself now aligns correctly inside text frames. A new Draft View for stripping out the headers, footers, and page margins for distraction-free writing is available too.

Calc picks up the same RTL detection we saw earlier. Typing right-to-left text into an empty cell sets its direction automatically instead of requiring a manual toggle.

With additional improvements like Calculated Fields for pivot tables and a "Shuffle" command for scrambling cell order within a selected range.

Similarly, Math gets new named functions and operators built for two west African writing systems, N'Ko and Adlam, which cover the Manding languages and Fulani, plus left-pointing vector arrows for formulas written right to left.

Chart has basic support for reading and writing back the newer chart formats (chartex) that show up in Microsoft's OOXML files, covering the box-and-whisker, funnel, Pareto, sunburst, treemap, and waterfall types.

Impress and Draw can mix page sizes within one file instead of forcing every slide or page to match, and Impress also picks up presentation sections, letting you group slides under a name instead of scrolling through one long list.

And generally speaking, Calc, Writer, Impress, and Draw now each carry their own background tint in the Notebookbar (the ribbon-style interface on top), so you can tell which app you're in without reading the title bar.

There's also a new Paragraph Composer that balances word spacing across a whole paragraph instead of judging each line by itself, fixing the uneven look that comes from justifying one line at a time.

⚙️ How to Install/Upgrade?

You will usually find LibreOffice already installed on popular Linux distributions, and it is only a matter of time before this release shows up as part of a routine system upgrade.

If you can't wait, then downloading the DEB or RPM package is the quickest way to get this release installed. Alternatively, you could also opt for the Flatpak, but know that it receives updates a bit late.

Refer to the release notes if you want to see the new changes in action.

Linux 7.3's merge window is still open, but what's already confirmed looks like the foundation for one of the more performance-heavy releases in a while. Btrfs direct I/O is pushing close to 95% of theoretical throughput, EXT4 is hitting 2.19 million IOPS on fast NVMe, and a KSM memory bug that caused 700ms freezes is fixed down to under 2ms.

Canonical is co-funding a three-year PhD at the University of Bristol to build a system that can translate large C codebases into Rust automatically. The test cases are AppArmor and snap-confine, both already running in production on Ubuntu, picked specifically because they're complex.

While not FOSS, TeamSpeak has been consistent in supporting Linux as a platform. It is now getting the ability to make direct calls between users without requiring a server.

Proton Mail's new Categories feature is live and works similar to how Gmail tabs do, sorting incoming mail into Primary, Social, Promotions, Newsletters, and two optional categories. The difference is that Proton sorts by metadata only since it can't read your email contents.

A one-line kernel bug in the Intel Xe driver was causing GDM to restart endlessly on Battlemage G21 cards. Tracking it down took Linus Torvalds 18 kernel boots and 24 debug patches.

There seems to be an official Flatpak package in early development for the Chromium browser.

This edition of FOSS Weekly is supported by Linux systems vendor, Star Labs.

🧠 What We’re Thinking About

You might have already heard about Omarchy's recent cash infusion; here's a quick run-down:

Ten tech founders have each written a million-dollar check for the Omacom Foundation, a nonprofit that DHH is in the middle of setting up to fund Omarchy and the open-source ecosystem around it.m

The first beneficiary is Vaxry from Hyprland, who will receive funding under an exclusive 3-year sponsorship.

The entire episode has (once again) divided the Linux community, as some people dislike DHH for his (right-leaning) political views, while the supporters don't agree with the usual (left-leaning) political views of most Linux-related organizations. Guess politics spoils a lot of things.

🧮 Linux Tips, Tutorials, and Learnings

You might have come across the term "Linux is just a kernel". Here's what it means.

Another basic but essential concept you should know about the root user, as a desktop Linux user.

Serious about perfect document formatting? Why not try your hand at AsciiDoc? It's a plain text format that converts cleanly to HTML or PDF, and this guide covers everything from installation and basic syntax to customizing your output with CSS and XSLT.

GNOME's default look is clean, but it doesn't have to stay that way. Our list covers 17 extensions for changing how the desktop actually looks, from blur effects and animated window close transitions to wallpaper slideshows and a proper desktop clock widget.

👷 AI, Homelab and Hardware Corner

Would you be interested in a newly announced $42 USB dongle that can turn your phone into a keyboard, mouse, and SSH terminal?

If that's not your thing, why not check out MNT Research's mini PC case designed to take the same mainboard as the MNT Reform laptop?

✨ Apps and Projects Highlights

weathr is a Rust CLI tool that shows your current weather as animated ASCII art, complete with a house, clouds, birds during the day, stars and fireflies at night, and weather-appropriate animations for rain, snow, fog, and thunderstorms.

📽️ Videos for You

Learn to tweak the KDE file manager in this video on our YouTube channel.

💡 Quick Handy Tip

In GNOME, File Roller lets you open up compressed archives to check out what is inside before actually unzipping anything. If you only need a single item, you can just grab and extract that specific file or folder instead of unpacking the whole thing.

Here's the unofficial Flathub listing if you want to get it installed quickly.

🎋 Fun in the FOSSverse

Confident that your container knowledge is enough to beat this crossword?

I laughed harder than I should have on "Microslop".

🗓️ Tech Trivia: On August 29, 1990, the UK's Computer Misuse Act took effect, one of the world's first laws targeting computer fraud. It followed the 1984 Prestel hack, where two men accessed Prince Philip's mailbox but couldn't be prosecuted under existing forgery law.

🧑‍🤝‍🧑 From the Community: Pro FOSSer Neville is asking fellow FOSSers whether they do anything to customize their desktop environment. Go on, share your riced-up setups in that thread!