Welcome back to Local AI Weekly. I promised a weekly, and then took a month to send the second one. So let's call the first one the pilot episode. From here, the show will run weekly 🤘

This issue leans heavily toward agents, because that's where most of the interesting local AI tooling is landing right now.

Before we see all that, let me thank Atomic Chat, who is supporting this edition of Local AI Weekly.

SPONSORED

Atomic Chat lets you run open-source LLMs privately on your own device, fully offline. It has Turboquant built-in providing faster local inference with longer context windows. Which means you can run bigger models smoothly, right on your device.

Explore Atomic Chat

🧪 On my bench: Unsloth

I downloaded Unsloth this week. It's an open source framework for fine-tuning and running open models faster and on far less VRAM than the usual setup. It's probably best known for its Dynamic GGUF quants, the ones that let people run 27B-class models in around 17GB of RAM, with 1-bit builds that squeeze onto 8GB.

Yeah...that's the kind of claim I don't want to repeat until I've tested it myself. So consider this a heads-up, not a review or recommendation. I'll try it on my ZimaCube (with Nvidia RTX Ada 2000) and give you the full verdict in a coming issue.

🔍 Discover AI tools

Three of this week's finds are about making your local agents "behave". tehy are smart, but they have to be accountable, too. All of these tools are open source.

The first is agent-inspect, a local-first debugger for TypeScript AI agents. It turns an agent run into a readable execution tree, so you can see every tool call, model call, and the exact point where things went wrong. It can also fail a CI check when the agent takes the wrong path, and bundle up a redacted trace to share with someone else.

Next is AutoMem, a persistent memory layer that plugs into agents over MCP. Most agents start every session from zero. AutoMem stores what matters in a graph for relationships and a vector index for meaning, so recall gives you the answer and the thread it belongs to. You can run the whole thing locally in Docker, offline, with your data staying on your machine.

The third is BrowserSkill from Tencent. It lets an AI agent use your real, already logged-in broswer instead of a fresh sandbox with throwaway accounts. The interesting part is that it runs tasks in a separate, visible Agent Window and only borrows a tab you have open when it explicitly asks, so you can keep working while it does. When it hits a captcha or a login, it hands control back to you and picks up after. Agents can still not tick all the traffic lights, it seems 😉

🖥 Which model can your hardware actually run?

This is the question I get asked the most. You see a model you want to try, but you have no idea if it can run on your system. Downloading and trying all of them is a waste of time and bandwidth.

llmfit helps you there. It's a terminal tool that detects your CPU, RAM, and GPU, then ranks open models by how well they'll actually run on your box, scoring them on size, speed, quality, and context. If you have Ollama running, you can even pull a model straight from its interface. It is open source and written in Rust (if that matters to you).

🧩 Open WebUI goes mainstream

A quick sign of the times. Open WebUI has become so popular that managed hosting providers are now offering it as a one-click service. PikaPods, for instance, now runs it for you.

Just so you know, I work with PikaPods as its growth consultant, so take the mention for what it is. But the point is that self-hosted AI front-ends have moved from hobbyists' Raspberry Pi to something people want running full time without managing a server.

🎙 Voice transcription lands on Linux

Back in issue #1 I said voice AI is shaping up to be the next big thing. LM Studio's Bionic release, v1.1.3 proves my point, as it now brings voice transcription to Linux, including on machines with NVIDIA GPUs.

LM Studio is free but closed source, so keep that in mind. If you are on Linux, you will find that the Bionic AppImage is about 1 GB. That's the biggest AppImage I ever downloaded.

📡 Open Model News

The open model space has kept up the pace since last time. Three releases stand out, and all three are MoE models, which is now clearly the direction the open world is heading.

Qwen3.8-Flash-Next from Alibaba's Qwen team. It's a 125B mixture-of-experts model that only activates 6B parameters per token, and it's an early preview of the architecture behind the upcoming Qwen4 family. Needs serious hardware power, though.

DeepSeek V4.1 Flash is here, too. On paper it's a huge 552B MoE, but it activates only 8B parameters on input and 16B on output, with a new architecture built around an aggressively compressed KV cache, roughly 890 bytes per token. Waiting for community quantization on this one.

Those two are heavyweight so the next one is "relatively affordable" by homelabbers. Ornith-1.5 from DeepReinforce comes in three sizes: a 397B MoE flagship, a 35B MoE that activates just 3B per token and runs on a single 24GB card, and a 9B dense model with a quantized build small enough for a phone. It's trained with a self-improvement loop, where the model writes its own tasks and grades its own attempts.

👀 Big Tech Watch

Two things worth your attention.

First, OpenRouter is being acquired by Stripe. OpenRouter is the big model-routing gateway, moving over 10 trillion tokens a day across 400 or so models. It says the product, name, and neutral routing won't change. Worth watching, because a lot of the local and open tooling world leans on that gateway staying neutral.

Second, the industry had a rare moment of agreement. Anthropic's Dario Amodei published an essay called "We Must Pace the Frontier," arguing that AI labs should deliberately slow down how fast they push model capabilities.

Within a couple of days, Sam Altman, Elon Musk, and Google DeepMind's Demis Hassabis all said they agreed with the core idea. The notable holdouts weren't the labs at all. The US President Trump rejected the idea flatly, saying "whoever wins AI wins."

If the frontier slows and open weights keep improving, that's a good thing for us Local AI enthusiasts.

🗂 AI Jargon: Harness

You'll have noticed the word "harness" turning up everywhere lately. Here's what it means.

Think of the LLM as the brain and the harness as the actual body. A model on its own is just a brain sitting in a jar. It reads text and predicts text. A harness is what turns it into an agent that can do things. The harness runs the loop i.e. it sends your request to the model, reads back the tool calls the model wants to make, actually runs those tools, feeds the results back in, and repeats until the job is done. It also manages memory, context, and what the model is and isn't allowed to touch.

The reason everyone is building their own harness is that the model has become the easy part. The value now sits in the harness you wrap around it: your data, your tools, your workflows, your guardrails.

Using an LLM as a glorified chat tool won't help much. Plugging the LLM into a harness tuned to your work or your company's work will let you achieve the actual AI automation.

⚡ Quick Tip: Keep your Ollama models warm

If you run Ollama and notice the first answer after a break is slow, here's why. By default, Ollama unloads a model from GPU memory after 5 minutes of no requests. I noticed this on my Zima setup. The next request then has to reload the whole model from disk into VRAM, which will delay the next response.

You can change that. Set the OLLAMA_KEEP_ALIVE environment variable on the Ollama service to a longer duration like 30m, or to -1 to keep the model pinned in memory until you restart.

My point is that free GPU memory is only useful if you're going to use it. On a single-user machine where you're the only one using the model, there is little point in freeing up the GPU every five minutes and getting slow responses as a result. However, if you use several large models on one card, the default timeout actually helps.

🎫 Events worth attending

If agents are your thing, there are two conferences coming up, and It's FOSS readers get a discount on both.

AGNTCon + MCPCon Europe is happening on September 17 and 18 in Amsterdam. Use the code ITSFOSS_50 at checkout, which brings registration to $375. It's valid right up to the event date, which is tomorrow.

AGNTCon + MCPCon North America on October 22 to 23 in San Jose, California. Use the code ITSFOSS for 20% or more off current pricing; it should be around $375 again. This offer expires on October 7.

In the end...

I hope it was not too long. Not that I was trying to compensate for the long gap between the two editions...or perhaps I was ;)

As always, some of this will be useful to you and some won't, depending on what you're building or what your interests are. Reply and tell me what you're running locally these days. I read every response.

See you next week (for real this time).

Linux gamers are used to being treated as second-class citizens, and Ubisoft recently gave everyone a fresh reminder when they pulled Linux and Steam Deck support from For Honor.

They blamed anti-cheat limitations, even though the game's anti-cheat software had supported Linux for years.

Meanwhile, cloud gaming has been slowly building up as an alternative, even for people who can't afford to build a gaming rig thanks to inflated RAM and storage prices. While growth hasn't been anything remarkable, people who can afford a subscription have been signing up for such services.

Geforce NOW is one such name that has managed to carve out its slice of the market, recently moving its native Linux app out of beta and into general availability, officially supporting Ubuntu at launch.

Geforce NOW on Linux

Instead of storing games locally, like when using a game launcher like Steam or Heroic, GeForce NOW streams them from NVIDIA's RTX GPUs in the cloud. This way, a laptop or desktop gets GeForce RTX-level rendering without needing to have any of that hardware itself.

Games come from your existing library, not NVIDIA's own store. Link Steam, Epic, GOG, or PC Game Pass, and you can launch anything you already own. All updates and installs are handled on the cloud gaming rig you get access to via the service.

Setting it up

Contrary to what the official system requirements say about using the Xorg session on NVIDIA GPUs, I managed to get the GeForce NOW client running on the Wayland session of Ubuntu 26.04 LTS.

My laptop has a GeForce RTX 3050 with 6GB of VRAM, and the system was running with NVIDIA's proprietary drivers rather than the open source Nouveau. Don't worry if you have a less powerful GPU on board, as the app only needs a local GPU to decode the incoming video stream, not render the game itself.

While I kept my testing limited to Ubuntu, I have a feeling this app can run on other Linux distros too.

💬 Why not test it and let us know in the comments below?

Before I could start playing, I ran a network test via the settings menu to see if everything was good to go. The app reported that I could face stuttering or high latency as I was connected to a wireless network, suggesting I connect an Ethernet wire instead.

You will see me facing those issues very soon. 😅

Initially, I tried using the "Free" tier to see what the native app could do on my Ubuntu setup. It was a big fail, as I was stuck in a 300+ waiting queue for access to a machine, and when I eventually did come to the front, I was kicked out because I had left my laptop idle.

I then decided to get in touch with NVIDIA, and thanks to some helpful folks, I was set with a 14-day promotional key for GeForce NOW Ultimate. Next, I had to test some games that could take advantage of a GeForce RTX cloud rig.

The titles I went for were Fortnite, as the multiplayer pick, and Ghostrunner 2, as the singleplayer pick.

Two games in the spotlight

Fortnite

The first game I want to show you is Fortnite, the infamous battle royale game that doesn't seem to be going out of trend anytime soon. Seeing that GeForce NOW supports directly linking one's Epic Games account to the service, I was able to quickly get into the game.

I left the graphical settings untouched, noting that the display name was "NVIDIA VGX," the FPS limit set to "Unlimited," and the rendering mode was "DirectX 12."

Even the matchmaking region was correctly set, and I was getting pings in the 100-120ms range. In the first battle royale match, I noticed that the image quality was not that sharp, and the game stuttered, with GeForce NOW warning of a spotty connection.

That was a given since I was connected to my ISP's weak wireless router. 🫠

Across sessions, things did improve, and I even managed to win a match. That is when I realized that I could use "Ctrl+G" to launch the GeForce NOW overlay, which gave me access to many options, such as the remaining session time, the ability to enable real-time statistics, and a few others.

Equipped with the expanded metrics, I got to know that I was on a GeForce RTX 4080-powered rig that was pushing the in-game FPS up to 160, as advertised for the Ultimate plan, while the stream was running at ~55 FPS and the server latency was at ~45ms.

I was also losing considerable amounts of frames and packets due to my network setup, which had been quite inconsistent at providing the required bandwidth to drive such a demanding multiplayer game mode that hosts up to 100 players in a single match.

Ghostrunner 2

Next up is the singleplayer pick for this review, Ghostrunner 2, which continues the story of Jack, a clank..., erm, cyborg, rebuilt by a group of human rebels after they found him in a near-destroyed state.

Before I could start playing this, I had to again link my Epic Games account to validate access to the game. This is something NVIDIA needs to work on. If they could give gamers a one-click way of validating all their owned games that are available on GeForce NOW via a specific platform, that would go a long way in ensuring a friction-less experience.

Coming back to the game, the last time I played it was when I tested the demo version in the third-party GeForce Infinity app, where it performed awfully. While there were a bunch of factors that caused those problems, this time around, it was a much better experience.

Like Fortnite earlier, I didn't change the pre-set graphical settings, keeping them at a relatively high preset without enabling any kind of upscaling via AMD FSR or NVIDIA DLSS.

Again, I faced some microstutters, but the game itself looked really good and was very responsive throughout my playthrough. Without the lag, the game felt as good as running on a similarly equipped gaming rig locally.

I can now confidently say that using Geforce NOW on a wired setup or with a very capable wireless router is the way to go. As doing so ensures a consistent flow of bandwidth that is required to stream such amounts of data quickly.

Being located closer to your nearest GeForce NOW server also helps. 😉

During play, I also noticed that my laptop going into Sleep mode would automatically disconnect me from GeForce NOW, with the app notifying me of the occurrence. All the while, the gaming rig itself remains active, and I can quickly jump back in by hitting Play again.

Some closing thoughts

At the end of my testing, I was left thinking that NVIDIA could easily bring the native client to platforms like Arch Linux and Fedora. This would help them cover a large chunk of the Linux gamer base, seeing that there are many derivatives of these distros that are quite popular among gamers.

Coming back to the question I asked in the title of this review: Is GeForce NOW gameable on Linux? On Ubuntu 26.04 LTS, yes, even without the Xorg session NVIDIA recommends. Wayland handled things fine.

While its closed-source nature and paid tiers will be dealbreakers for many of you out there, the thing it offers is convenience. You get easy access to a gaming rig without having to invest in one, and the latter's a far cry (gaming pun hehe) right now thanks to inflated prices.

Game on!

If you are looking for instructions to install GeForce NOW on your Ubuntu PC, then continue reading. First, go to the official website to fetch the .bin file; look for it under the "Linux" category.

Once the file is downloaded, go to it in Files and right-click on it. Now click on "Properties," and enable "Executable as Program."

Launch the executable by either double-clicking on it or clicking "Run" from the right-click context menu. Wait for it to complete the download and installation.

Next, you will have to agree to NVIDIA's Privacy Policy and GeForce NOW's Terms of Use. Be sure to go through them, as they contain many clauses.

The app will then ask you to sign in with your NVIDIA account, for which the authentication will be carried out via the default browser on your system. And that's about it; just select a plan and start playing.

Ever found yourself in a situation where you have a document with plenty of images and you need to save images from it? Maybe a selected few or all of them?

Now, manually going to each image and then right click might be the no-brainer but an extremely tedious way to achieve this.

Let me tell you the quickest way to extract all the images from a work document.

Access the menu and go to the Save As option. What you have to do here is to save it as HTML. It will create several web files inside the same folder where your document is, but it will also extract all the images. That's what you want, right?

I'll show the details steps for LibreOffice in this quick tutorial.

Extracting all images from a document in LibreOffice

First thing first, create a new folder and move your desired document file in there. Now open the document in LibreOffice.

Click on the top file menu and then click on the Save As option. The same can be achieved with Ctrl+Shift+S keyboard shortcut.

When it gives you the option to save, make sure to change the file type to HTML like this:

There are actually a number of file format options when you click that. You will have to look for HTML here.

Now that you have ensured that it is saved as HTML, click the Save button.

And then you should see all the images and media files from the document along with a single HTML file in the folder where you saved it. The default location is the same as your original document.

As you can see in the screenshot above, it also saved GIF along with PNG images. There is HTML file and the original doc, too.

This is much better than individually right clicking images and saving them.

And what if you wanted to select only a few images, not all? Well, in that case, you could either manually save the ones you want or extract them all and then only keep the ones you want and remove the rest. It is really up to you.

Another tip, since we are here, if your doc has way too many images and you want to extract a few from the selected chapters only, you can create a new doc and copy paste the chapter in it and save it as an HTML.

Basically, you can utilize this "save as HTML" option in a variety of situations.

Works with other platforms, too

This little trick actually works with most office suites, from MS Office to ONLYOFFICE. It works in Google Docs, too. Consider it a "universal trick" for extracting all images from a document.

I hope you liked this quick little LibreOffice trick. A quick thanks, if you liked it, is always welcome 😄

Ubuntu has managed to do away with GNU Core Utilities in its default stack. The last three holdouts, cp, mv and rm, have moved to uutils' coreutils; the Rust reimplementation Canonical has been feeding into the distro since 2025.

They had been held back from 26.04 LTS over flaws in the uutils versions. Everything else, from ls and cat to chmod and du, made that jump in earlier releases.

This change, while big, sits hidden away in an obscure mention in Canonical's work-in-progress release notes for Ubuntu 26.10.

It's been a long road

Canonical started oxidising Ubuntu last year, and Ubuntu 25.10 became the first release to ship coreutils as the default. That release also made sudo-rs the default privilege tool, replacing a command that had been in place for decades.

26.04 was the release where the plan did slow down quite a bit, as Canonical kept cp, mv, and rm on their GNU versions due to a bunch of TOCTOU issues that were blocking the full implementation.

These were caught during an audit, when Canonical commissioned Zellic for two rounds between December 2025 and March 2026, focusing on the most security-sensitive utilities first.

Across both rounds, Zellic raised 113 issues, and 44 of them were assigned CVEs. Canonical says the vast majority have been resolved.

Getting here has had its ups and downs, and the last stretch was not clean. In July, uutils cp went back into the archive and came straight out again after it broke live image builds.

The fix was quick; as the developers marked it "Critical," the fix went upstream, and the migration landed in time for 26.10.

What changes for you?

When typing commands, nothing changes for you on the surface. uutils coreutils is designed to be a drop-in replacement for essential GNU tools, and the project treats any divergence from GNU as a bug, further pointing out that some options may still be missing or behave differently.

So if you prefer staying on the GNU version, you have the option to install the coreutils-from-gnu package that houses all the required components.

The next stage

Coreutils is one piece of a broader campaign. Earlier this year, Canonical became a Gold Sponsor of the Trifecta Tech Foundation, pitching in €40,000 a year to fund memory-safe system software.

Under this, their current target is ntpd-rs, a Rust rewrite of the tools Ubuntu uses to keep its clock in sync. While work is still ongoing, it has already arrived for testing.

Its transition to being default is targeted for Ubuntu 27.04.

What Canonical is gradually building up towards is the completion of their oxidation vision for Ubuntu, and it's not about blindly including new components. Rather, it looks like a measured approach that's being worked out a few steps at a time.

Suggested Read 📖: Linux is getting ready to purge old 32-bit Arm platform code.

Rune has left behind its proprietary old self for a new open source identity under the GPLv3, letting you clone, read, or build off it from GitHub. The relicensing came via a string of moves, one of which was the commit that marked the project's license transition.

The developers have also decided to keep commit history mostly intact, which runs back to 2017, with some gaps here and there. I point that out because projects that go open often squash the history first, so old secrets and sensitive commits never land in the public repo.

To give you some context, Rune is an IDE for Linux and macOS that can be used for writing and debugging code, running your terminals and CLI tools in the same window, and letting a coding agent take on parts of your tasks.

It is based on Go, and the team defends its use by showcasing vtebench runs that show the Go terminal keeping pace with Alacritty, Ghostty, and Kitty on a set of terminal workloads.

Do keep in mind that these were not language benchmarks; these only show how far the IDE has come.

The Rune team is further looking to expand language support beyond Go and Python, with Rust and Zig already in beta on main and an open call for developers to steward their own language.

While the IDE itself is free to use, people can opt for paid plans to get access to the P2P Rune Network, premium support, and a separate tailored plan for enterprises with extra bells and whistles.

Going the open source route

The license, GPLV3, allows contributors to keep the copyright in what they submit, and the company (Unstable Build LLC) takes no special right to relicense their work. Earlier, the project asked contributors to sign a CLA; that's not the case anymore.

Rune's creator, Ernest Romero Climent, is not happy with what has happened elsewhere in open source, where companies build in the open, lean on the community for years, then change the license once the project is worth something.

Instead, they are looking to pay contributors via an opt-in credits program, where accepted work earns credits and a slice of eligible service revenue funds a pool, which is then split by the credits each contributor holds.

All awarded credits will be tracked in a public ledger, showing the revenue behind the pool, any deductions, the credits awarded, and how each payout is worked out.

As I said, all of that is optional. You can still send patches without joining the program.

Suggested Read 📖: openEuler and the race to build Asia's AI infrastructure.

When it comes to open source and AI, we almost always look West. While a lot happens in the United States and some in Europe, the East is not lagging either.

In the recently concluded LEAP East conference, several startups and enterprises came together to showcase their innovations and share what they were building.

Out of them, openEuler caught our attention for its focus on Open Source AI.

The SuperPoD

At LEAP East 2026, openEuler showcased its SuperPoD OS, which extends operating-system support for UnifiedBus-based SuperPoD computing systems.

Think of a SuperPoD as a cluster of several powerful computers that need to act like one giant machine. Running AI workloads takes computing power and doing it across hundreds of such machines is messy as they need to share memory, pass data around quickly, and not conflict with each other.

UB Service Core is the open-source software layer, part of the openEuler ecosystem, that makes coordination across those machines actually work. It provides five cluster-level system services for UnifiedBus-based SuperPoDs: Engine, Memory, Communication, I/O, and Virtualization.

The Engine pools memory and DPU resources across the cluster and schedules them dynamically. It also handles automatic failover so the system keeps running if a node goes down.

The Memory component lets machines within a UnifiedBus-based SuperPoD share and pool memory using unified-memory programming, meaning software can treat distributed memory as a single addressable space.

The Communication piece provides high-performance networking over UnifiedBus while staying compatible with standard Socket and Verbs interfaces, so existing applications don't need to be rewritten to benefit from it.

The I/O service handles global data read/write caching at a high level, which matters for workloads like AI inference that repeatedly access large shared datasets.

The Virt component covers virtualization pooling, decides live-migration policy (when and how to move a running VM to another machine), supports disaster recovery, and improves communication efficiency between VMs and containers.

Together, these five services give a SuperPoD cluster the coordination it needs to perform like a single well-organized system, with a claimed 30-50% performance gain over conventional architectures.

According to openEuler, UB Service Core can improve application performance by 30–50% by taking advantage of the peer-to-peer UnifiedBus interconnect architecture.

This SuperPod or UB Service Core is fully open source under the openEuler umbrella.

Agentic AI in the OS

openEuler is evolving toward an AgentOS model. In openEuler 24.03 LTS SP4, DevStation (desktop variant of openEuler) includes openEuler Intelligence, a desktop AI assistant that uses an agent runtime and MCP-based tools to support development and system operations.

Pre-built agent harnessing in the operating system is not uncommon these days. openEuler 24.03's built-in AI agent can be used to control the Linux system.

So, instead of doing things manually, by mouse clicks, you could just ask the agent to dim the brightness, lower the volume, or change the wallpaper. The built-in AI agent has custom skills to control the system settings in openEuler's UKUI desktop environment.

This is a work in progress. But the direction is clear. AI agent should soon be able to let you control the system instead of doing things manually.

Expanding into a community-based ecosystem

openEuler has launched openEuler Hong Kong User Group to bring together developers, researchers, and organizations in Hong Kong to connect, share experiences, and explore open-source innovation together.

openEuler is an open-source project incubated and operated by the OpenAtom Foundation. The community is expanding international collaboration through local user groups, industry partnerships, and participation in global open-source events. The Hong Kong User Group represents another step in openEuler’s efforts to encourage local participation and international collaboration.

E-mail is one of those things that get really unruly before you realize they have. You sign up everywhere with your e-mail, get lots of spam from services that you almost forget you have even heard of, and all the important things, consequently, get lost in the noise.

This is the problem Paperweight aims to solve. It is an open source application that helps you see what accounts or newsletters you subscribed to so that you can take a decision on them.

Paperweight helps you reduce your digital footprint locally

You probably already know that a digital footprint is the trail of information you leave behind when you use the internet. While it is not easy (nearly impossible) to erase the entire digital footprint, we sure can reduce it.

There are premium services like Incogni from Surfshark that does this job for a fee. They look into all the data broker databases and also send the data deletion requests on behalf of you.

Paperwight aims to help you do some of those tasks. The idea is this. You plugin your email into Paperweight. It scans all the emails and gives you:

• Accounts you signed up for or made a purchase
• Email lists (newsletters) you are on
• Personal data (like phone number, payment details) in the messages

While you can bulk unsubscribe, rest of the action needs manual action. You will have to approach the respective websites by sending them emails individually.

And here's the thing. All this processing happens locally on your system. The inbox is scanned on the system, there is no server in between, so no data collection. The source code is available so you can verify these claims.

Let's see paperweight features in detail.

Bulk Unsubscribe

All the services that send you e-mails are listed under "Mailing lists" right on the main window, and you can go in and check what e-mails you've been sent, how many, and the dates of the recent ones.

Here's the exciting part: you can select multiple services at once and unsubscribe from their e-mails without having to individually open every single service's e-mail and going through the unsubscription process, which usually takes you through a few browser hoops.

After selecting the services you wish to unsubscribe from, it becomes a simple matter of clicking the dumpster icon and confirming:

Note that many email services, including the privacy-focused ProtonMail has this feature built-in.

Account Inventory

An entire "Accounts" section shows you exactly all the places where your e-mail is being used to log into other services. All the services where you did "login with Google" will also show up here if you're using a Gmail account.

There, you can see the kind of access it has to your data, including financial, identity or health information, and gives you an unsubscribe as well as a "delete all e-mails" button.

Again, the automated unsubscribe request may not always be respected by the service at the other end.

Breach Alerts

Under the banner of the same accounts inventory, Paperweight also makes an exclusive list of services on that roster that have been known to be at risk or breached. It gets the security risk data from publicly available Have I Been Pwned, and gives you the options to delete/request your data, and to delete their e-mails on your account.

GDPR Deletion Template

GDPR (General Data Protection Regulation) gives users the "right to be forgotten". As such, mostly companies have to comply with the rule and relinquish any and all data they might have regarding an individual if the individual requests so.

How does Paperweight come in? Well, they give you a template to request your data from any service you are subscribed to. Just click on the "Data requests" option down below in actions, and you'll see an almost fully written e-mail to get the company rid of your data.

It is still not fully automatic but it is still better than doing it all from scratch.

Interface

The user interface is pretty sleek and pretty simple. On the dashboard, three big blocks show the number of e-mails analyzed, mailing list subscriptions, and accounts where your e-mail is used.

On the left, there are small buttons to access pretty much the same things. There are a few "at a glance" type lines showing recent activity and highlights of the analysis of the account.

Finally, a frequency graph shows the number of e-mails you've received in the past few months.

The interface for further options are shown in the screenshots above and pretty much show you all available options right on the face. You don't have to look too far for anything at all.

Data Traffic

Just out of curiosity, I tracked the network traffic going through Paperweight and to my relief, it only seems to be talking to the e-mail service providers' servers or the servers of the services that you're trying to modify. For example, there will be traffic to and from Canva's servers if you're trying to unsubscribe from their mailing list.

Free vs Lifetime

Paperweight might be open source, but it is a freemium sort of application. There's a free version that lets you use Paperweight with one email account and 90 days of email history.

And then there is a lifetime license that costs a one-time payment of $69 and gives you the option to add multiple accounts and scans through your entire past e-mail history.

Here's the thing. The source code is available on GitHub, under the MIT license and if you want, you can fork the repo, modify the code to remove the restrictions, and use all the features. There is no constant validation for the license. However, that would be a breach of honor.

Although I feel that $69 is a bit too much, probably because I am a student. Price parity and pay-what-you-want model would have been better. But that's just my opinion.

Getting Paperweight

Paperweight is available for all of the big three operating systems Linux, MacOS and Windows.

On Linux, they provide an AppImage and a .deb package.

Conclusion

For a lot of users, Paperweight probably provides a service that you had no idea you needed. But now that you are, you can use it periodically to maintain a clean and healthy inbox. Once a month, on a weekend should be good enough.

The free version does feel limited, though, especially the 90-day history cap. I would have preferred a Melia like solution that allows full access for a single email account and lifetime license for multiple accounts.

If you give Paperweight a try, do share your experience in the comments.

Most of us carry a phone every day, but very few of us could say what's actually going on inside one. That gap is exactly what CircuitMess is trying to fill or should I say, has been trying to fill.

The Croatian company CircuitMess, founded by Albert Gajšak, ran its first MAKERphone campaign back in 2018, a build-it-yourself phone kit that raised over $324,000 and eventually shipped to more than 6,000 people as "Ringo."

Now there's a sequel. MAKERphone 2.0 is live on Kickstarter (partner link) and it has already sailed past its goal with weeks left to go.

The pitch is the same as before: you assemble a working mobile phone yourself, then learn to code on it. The major difference this time is that you don't need a soldering iron to do it. That's a big relief, for me at least, as I am terrible at soldering.

📝 Key Specifications of MAKERphone 2.0

As you can see, the MAKERphone looks like the normal keypad cellphones. Here are the main details:

• Processor: ESP32-S3, dual-core, 4 MB flash, 4 MB PSRAM
• Cellular: 4G LTE Cat-1 with 2G fallback, Nano SIM
• Display: 1.77-inch color, 160×128
• Input: 12-key numpad, nav joystick, 4 function keys, volume keys
• Audio: DAC, speaker, I²S mic, 3.5 mm headphone jack
• Battery: 1000 mAh, USB-C charging
• Storage: microSD slot
• Connectivity: Wi-Fi, Bluetooth, USB
• Expansion: UMAX modular connector
• Assembly: no soldering required this time
• Coding: MicroPython, Arduino C++, VibeBoy (AI-assisted)
• Price: starts at $119 for early backers, $169 otherwise

The kit starts at $119 for super early backers and climbs from there and goes for a retail price of $169.

This is a DIY gadget, the phone that arrives as a box of parts and a build guide, and you put it together into something that makes real calls and sends real texts over 4G LTE. Pop in a SIM and it behaves like an actual phone.

Now 4G seems like a step down from the usual 5G but the device is not supposed to replace your smartphone. If you are not going to use all those internet connected social media and apps, you don't really need 5G. I explain the concept later.

At the core, there is ESP32-S3, the same family of low-power microcontroller you'll find in countless hobbyist boards. So no, this is not going to compare or compete with Snapdragon or Apple Bionic chip and they don't even pretend to do so.

The original MAKERphone asked you to solder the whole board together. Version 2.0 drops that requirement, which lowers the barrier a lot if you've never touched an soldering iron. CircuitMess says it's built to be assembled by kids from around 11 up, so a functional adult should manage fine.

From the official assembly video, it seems like just putting the pieces together.

There's a 1.77-inch color display, a headphone jack (yes, in 2026), and a modular design that lets you bolt on extra hardware. More on that below.

Modules, coding, and a smart band

The part that makes this more than a nostalgia kit is the module system. MAKERphone 2.0 takes add-on modules that change what the phone can do. Several of these modules come in the box.

There's an infrared module for controlling an old TV, a camera module, a 2W speaker, an LED module, and a temperature and humidity sensor. There's also a perfboard so you can wire in your own sensors.

Once it's built, you code it. You can write for it in MicroPython or Arduino C++. CircuitMess is also pushing an AI-assisted coding tool it calls VibeBoy, so the "vibe coding" fad has officially reached the DIY phone aisle.

And that's not a bad thing entirely. Not everyone can code or make a custom app in MicroPython. But LLMs have removed that barrier. You have the hardware; you use a LLM and let it generate the kind of software you want. That's the reason that Seeed Studio and many other hobbyist hardware makers are promoting their own AI-assisted platforms. It makes hardware more enjoyable for bigger group of people.

CircuitMess is selling a companion too: a programmable smart band that pairs with the phone over Bluetooth. Think of it as a semi-smartwatch as it can show you time, let you play games and use timer and stopwatch functionality. There are no sensors here to track your health and activities, as far as I can see.

It's a separate accessory rather than part of the base kit, so treat it as an add-on if the idea appeals.

The idea behind this DIY phone

A 4G phone seems like an outdated device in 2026. After all, 5G is everywhere and even the most basic smartphones support 5G these days. But the premise of MAKERphone is entirely different. It's not a smartphone by choice.

We have seen such devices in the 'maker space' in the past. There is Flipper Zero that has an entire different purpose, then there are handheld retro gaming devices, pocket computers, etc.

This is a similar idea where kids in STEM or hobbyist adults can build the device as per the liking, by adding the desired modules.

To give it more functionality, there is the scope of a 4G sim card. This gives you the ability to make calls and text messages.

With that, it becomes a bar phone that you assemble and design new games and apps for your own self.

And the device is still useful as a phone. If you want kids to stay connected with you, not the social media, it works. The same goes for adults who deliberately want to stay away from the internet (except when using a proper computer).

🛒 Pricing and Availability

MAKERphone 2.0 is live on Kickstarter now and has already hit roughly 790% of its funding goal, with about 27 days left on the campaign as of writing.

Pricing starts at $199 for the early bird tier, with higher tiers as the discounts sell out. There are way too many tiers to fit in all here.

US, EU, UK and Canada residents do not need to pay any import fee. VAT is likely to be extera depending upon your location. Shipping cost are not clear to me but the device should start shipping in July-August 2027.

If you just want an inexpensive (smart)phone, this isn't the device for you. If you want to understand phones, or hand a kid something that turns screen time into build time, and the budget and wait time is okay for you, you can make a purchase.

A laptop fingerprint sensor unlocks the laptop and usually stops there. Immurok wants to do quite a bit more than that.

It is a tiny wireless box that can unlock your desktop session, approve a sudo command, authenticate via polkit, log you into a server over SSH, generate TOTP codes for 2FA, and put a physical touch in front of an AI coding agent. And it does all of that with the same finger, over Bluetooth.

The good news is that the core features work. The less good news is that on Linux, getting there was not as simple as touching a sensor. During my testing I ran into bugs, missing prompts, confusing behaviour, and one timing issue that made the initial setup a little challenging. But once things were setup, the device worked smooth and that's what matters, right?

Let's take a detailed look at what Immurok gets right, what still needs work, and whether this little box could earn a place on your desk.

I also have a video review of Immurok, if you prefer watching over reading.

What Immurok actually does

At its simplest, Immurok is a wireless fingerprint reader for a desktop or a laptop that does not already have one. But the feature list goes well beyond unlocking a screen.

You can use it to unlock an existing desktop session. Through Linux PAM integration, it can approve sudo and other supported authentication prompts such as polkit. It can generate and store SSH private keys on the device itself, and it can release a TOTP code only after it recognises your fingerprint.

Then there is the more unusual usage: AI-agent approval. The idea is that a coding agent can prepare a privileged command, an SSH operation, or a request for a secret, but it cannot complete the protected action until you physically touch the sensor.

According to Immurok, fingerprint templates and private keys stay on the device. There is no cloud account and no telemetry. Pairing uses P-256 ECDH with HMAC-SHA256, and authentication responses are cryptographically signed.

Those are the company's security claims, and as I said above, this review tests how the device behaves in use rather than auditing that cryptography.

The hardware

The hardware makes a good first impression. Here is what you are getting.

There are controls at the back for power and for actions such as pairing or confirmation. The silver finish, rounded corners, and proportions make it look like a tiny Mac mini. That is not a complaint and even works well for people in the Apple ecosystem. It is neat enough to sit on a desk without looking like a random security dongle.

My review box contained only the device. There was not even a USB cable and I am not complaining. The battery life is pretty good, so charge is needed like once a month and type C cable is in every household, I presume.

Immurok claims more than a month of normal use per charge. When I started testing, the Linux TUI showed 84% battery on the first run. After roughly a week, it was at 78%, so about 6 percentage points in that period. That looks promising for the month-long claim, but one week is not enough to confirm it, especially since battery-percentage readings are not always perfectly linear.

Installing on Linux means building it yourself

Windows and macOS get graphical applications. On Linux, the version I tested used a Rust daemon, a command-line client, and a terminal user interface. There was no equivalent graphical management app and no easy-to-install binary as well.

On my Ubuntu 26.04, I first installed the required development and runtime packages, then Rust through Rustup. The make check-deps command was helpful here, because it listed the missing components instead of making me discover them one build error at a time.

Then came the first Rust build, with around 200 crates to download and compile. It kept the CPU busy and produced plenty of warnings, though these were warnings and not compilation failures.

After that, make install was fairly quick. It installed the daemon, the command-line tools, the PAM module, the authentication helper, and the system integration, and the user-level daemon was already running when the install finished.

For an experienced Linux user, this is all doable. But for a device that presents itself as a simple replacement for typing passwords, it is too much friction. Before this ships to general users, Linux needs straightforward packages for the supported distributions and a much clearer first-run flow.

Pairing: connected is not the same as paired

The first pairing attempt immediately showed how early the software was.

I powered on the device and tried pairing from the TUI. It said the device was not connected. The CLI told me to press the device button within 30 seconds, but then failed immediately. It did not wait for 30 seconds at all.

What eventually worked was connecting Immurok manually through Ubuntu's Bluetooth settings first. Once the operating system showed it as connected, I went back to the Immurok TUI, pressed p, and then pressed the button on the device. It exchanged keys and reported that pairing had succeeded.

So there are two separate states here: connected over Bluetooth, and paired inside Immurok. That distinction may be technically reasonable, but the application never explained the required sequence when I actually needed it.

The product should walk you through this step by step, including the LED state, the operating-system connection, the Immurok pairing action, and the button press.

Enrolling fingerprints

Once the device was connected and paired, enrolling the first fingerprint went more smoothly.

In the TUI, I pressed e, placed my finger on the sensor, lifted it, and repeated. My test firmware asked for 12 captures before it enrolled the finger in slot zero.

The device supports up to 5 normal authentication fingerprints, and when you add more after the first one, an existing authorised finger has to approve the new enrolment. That is a sensible security requirement, but again, it needs to be stated clearly on screen.

The command line interface is finiky at times. At one point, trying to enrol again failed because the fingerprint was already present, and a deletion attempt simply failed without explaining why.

The bug that stopped sudo from working

With a fingerprint enrolled, I tested sudo apt update. Nothing happened. Touching the sensor did not authenticate me, even though the fingerprint was clearly there.

The problem turned out to be a race in the Linux daemon. When the Bluetooth session began, the daemon immediately asked the device for its status. But a Python notification helper was still starting up its D-Bus connection. On my machine, that helper took about 5.2 seconds to become ready, while the status request timed out after 5 seconds. That tiny difference left the daemon with no device status for the rest of the session, which showed up as NO_STATUS.

Since I had installed Claude for testing the agent specific features of Immurok, I let it handle the issue and it fixed that by increasing the limit to 15 seconds.

After rebuilding and installing the patched version, sudo authentication worked. I ran the command, touched the sensor, and it continued without asking me to type the account password.

It is a good thing that the code is open enough to inspect and fix. But this was also a basic authentication path failing because one helper started a fraction of a second slower than expected. These things should be fixed before the mass release.

Unlocking Ubuntu, and why fallback matters

With the patched client, Immurok could unlock an Ubuntu session that I had already logged into and then locked. I touched the sensor and the desktop opened.

It did not work at the very first login screen after powering on the computer, though. That is intentional behavior applicable to macOS and Windows too. The user-level service is available for unlocking an existing session, but the initial login still needs the normal account password. This distinction matters, because "screen unlock" can easily sound like "replace your password at every login," and they are not the same thing.

The fallback behaviour was good. When I disabled Bluetooth or disconnected Immurok, sudo and login did not leave me waiting for a fingerprint device that was no longer there. They went straight to the usual password field. That is essential. A convenience device should never turn a temporary Bluetooth glitch or a flat battery into a lockout.

SSH keys are the highlight

SSH is where Immurok starts to become more interesting than an ordinary fingerprint reader.

It can generate an ECDSA P-256 SSH key on the device. The private key stays there, and you export only the public key to place in the remote server's authorized_keys file.

ECDSA is a practical choice here because it can deliver security comparable to RSA using much smaller keys. That means less storage, less data to transfer, and lower processing overhead, making it suitable for a small battery-powered device.

The first time I generated a key through the CLI (not the TUI), it appeared to just stop. The missing piece of information was that Immurok was waiting for my fingerprint authorization. There was no prompt telling me to touch the sensor, and no useful feedback when I did. I only knew what to try because I had already been digging into the behaviour. Once I touched the sensor, the key was generated.

If an action requires physical authentication, the interface has to say so. The TUI was better here. When I generated another key there, it explicitly told me to touch the sensor to authorise the action, and the key then appeared in the next slot.

I exported the public key, added it to my Raspberry Pi Pironman Max, and enabled Immurok's SSH takeover. This adds the Immurok agent socket as the SSH IdentityAgent, which gives the device-backed keys priority.

When I connected to the Pi, Immurok asked for my fingerprint, I touched the sensor, and the SSH login succeeded.

Importantly, my normal SSH setup was not destroyed. Existing keys on the computer stayed available as a fallback, so when I connected to a server that did not have the Immurok public key, SSH just continued using my existing key.

That is the right approach. You can introduce Immurok gradually instead of forcing every server to change on day one.

You can delete keys through the TUI after fingerprint approval, or through the CLI, though the CLI again failed to show the required fingerprint instruction during deletion. You can also import an existing private key onto the device, but with two limitations.

Import was only available through the CLI in my tested version, and it accepted only ECDSA P-256 keys. If your existing key is Ed25519, RSA, or another format, you cannot simply move it over to Immurok.

TOTP just works

Immurok can also hold TOTP secrets, and this was one of the smoother experiences.

I added a GitHub TOTP entry, went into Immurok TUI, and then requested a code. Fetching the code required another fingerprint touch. I entered that code into GitHub, and it worked.

This is a simple feature, but a useful one. The TOTP seed is not sitting in a general desktop authenticator, and someone using your unlocked computer still needs your fingerprint before Immurok will release a code. As always, recovery codes still matter. If the device is lost, damaged, discharged, or reset, you need another way into the account.

Putting a fingerprint in front of an AI agent

The most distinctive pitch is using Immurok as a human-in-the-loop control for AI coding agents.

I installed the Immurok plugin for Claude and activated it for the project. Immurok also provides an imk wrapper that can launch a command in agent mode. In one test, the agent tried to connect to my Raspberry Pi over SSH, Immurok requested fingerprint verification, and after I touched the sensor, the agent connected to the Pi.

The idea is that even if the computer is unlocked, the agent cannot complete a protected wrapped action while you are away from the desk.

But there is an important limitation: this is only a gate if the action actually goes through the gate. Project instructions may tell an agent to use imk, and a plugin may make that tool available, but an ordinary command can still take the normal route if the wrapper is bypassed. My earlier Git push did exactly that, because it was not wrapped.

So this is not automatic enforcement just because a file tells the agent what to do. A serious setup has to arrange for sensitive operations to always run through imk run --agent, or to enforce the protected path at a lower level. Immurok can provide a valuable physical approval mechanism. It cannot secure a workflow that still leaves an unprotected path open right beside it.

Things that could be better

Even with the core features working smoothly after the setup, a few rough edges are worth calling out clearly. None of these are dealbreakers on their own, but together they are the difference between "promising" and "ready."

The TUI and CLI need serious polish

The Linux TUI brings together the dashboard, keys, PAM integration, logs, firmware updates, and settings. It shows connection state and battery percentage, and for most tasks it communicates better than the CLI does.

The CLI version lacks tab completion, which made nested commands harder to discover. Pairing claimed it would wait and then failed. Failed enrolment and deletion did not give useful explanations, and SSH import lived only in the CLI while other actions were easier to follow in the TUI.

Documentation is not optional for a tool like this. Immurok touches PAM, Bluetooth, SSH configuration, private keys, TOTP, etc. The GitHub README has details, but buyers need a stable setup guide, clear in-app prompts.

The log file that never stops growing

There is one smaller issue that gets bigger with time. The daemon continuously writes to ~/.immurok/logs.txt. After about a week, mine had reached roughly 4 MB. After 2 weeks, it is 15 MB.

At a steady rate, that is closer to 200 MB a year than multiple gigabytes. But the real problem is that the file looks unbounded, and a persistent reconnect loop or a recurring failure could make it grow much faster. The application needs log rotation with a sensible size and retention limit.

On Linux, sending routine logs to the journal would make far more sense than quietly maintaining an ever-growing private text file.

Is Immurok worth it?

After more than a week, I am more than happy to use Immurok as a daily driver. The trouble was mostly at the initial setup stage, and I am way past that now.

I liked the fingerprint reader on my Dell XPS in the past. My present laptop, Asus Zenbook, doesn't have one, so the Immurok fingerprint reader complements it very well.

The hardware is attractive, compact, and useful, and the core features work well. I used a fingerprint to approve sudo, unlock an existing Ubuntu session, authenticate an SSH connection to a Raspberry Pi, retrieve a working GitHub TOTP code, and approve an agent-run SSH action.

I also like the fallback design. A disconnected device did not block password login, and servers that were not configured for the Immurok key kept working with my existing SSH keys.

The hardware works great and the software on Linux needs upgrade. Source code build should not be forced onto users in 2026. The CLI needs to be better at showing the appropriate feedback message.

I have already seen two firmware updates and there is a software update already on their repo. So, it's just a matter for software updates and the device would become suitable for even the beginners.

So, if you think that the device suits your workflow and it doesn't dent your budget, you can opt for Immurok.

Getting Immurok fingerprint key

At the time of writing this article, Immurok is still in crowdfunding. It is available through Kickstarter for $59 US, with a stated retail price of $69 and estimated delivery in November 2026.

Please readhe their terms and conditions on shipping, custom fee and return policy.

I tested the pre-release review unit and the software is already changing quickly. The version shipping to backers is still a few months away, so there is time for the Linux experience to improve before customers receive the device. I do hope the installation, pairing, prompts, documentation, and reliability are all much better by November.

For now, I am happy to have Immurok on my desk.

Mirko Brombin, the creator of Bottles, has got Microsoft 365 running on Linux. His posts on X showcase an installation that completes on a Windows 10 or 11 Wine prefix, signs in with his existing license, and launches at least one editor.

He achieved this using two things, an unreleased version of Soda and Bottles 67.4. If you remember, Soda is a Wine-based runner that facilitates easy running of Windows games and applications.

On first glance, what he posted might look like a developer-focused log, but these changes are the foundation of what full Microsoft 365 support via Bottles will one day hopefully look like.

The results of his work

Mirko was able to login into the MS 365 app using his credentials and was able to launch MS Word and use it. Of course that's not the only thing he got working; since two-factor authentication passed, the suite detected his license, and OneDrive integration was functional.

The memory footprint of the office suite was less than 500MB, so this should be light on your Linux system.

The build he is currently testing is 2608 (20326.20144), which he says is the last build the installer will offer for Windows 10 and 11. And before you get all hyped up, wanting to try this out on your computer, know that you will have to wait.

Soda 11.0-11 (error 404 for now) has no release yet, and the component catalog that ships with Bottles 67.4 stops at Soda 11.0-10. Even the long-standing community request for a Microsoft Office installer in Bottles is still open.

Suggested Read 📖: Bottles offers hope for running Windows-only software on ARM devices.

Do we need Microsoft 365 on Linux?

Till date, there's not been an official Linux desktop build of Microsoft 365, and seeing how Microsoft currently operates, I don't expect one to appear anytime soon.

What's been achieved here is remarkable. Once fully implemented, this will help draw in people into Linux, as many shudder at the thought of switching to the platform because their preferred office suite doesn't work here.

For context, CodeWeavers sells a paid Wine product, CrossOver, and its database still lists Microsoft Office 365 on Linux as "Installs, Will Not Run," last tested against CrossOver 26.3.0.

Which is why I hope the fix travels further than Bottles. Wine sits beneath every one of these projects, including Soda. So a patch that lands upstream reaches all of them at once, and you would not need a particular runner to get Office working.

💬 If Mirko gets it all working, do you think other projects will start integrating his work?

A few days ago, thirteen patches were sent to the kernel mailing lists that are looking to delete some of the 32-bit Arm platform code that the upcoming Linux 7.3 release cycle has marked as deprecated.

While none of them have been merged yet, Arnd Bergmann, who has been trying to retire this code for years now, has finally made some good progress.

His suggestion is to do these removals in Linux 7.4, which is supposed to land sometime between December 2026 and January 2027, with the CPU support getting dropped a release after that.

The plan has been around for some time; the earlier proposition was removal in 2025 or 2026, after the matching LTS kernel shipped. Arnd opened his July thread by apologizing for letting the earlier attempt remain idle for almost two years.

What he calls cleaning out the deprecated platform code now runs to about 300 patches. This first series removes just the arch/arm/mach-* code and the Kconfig bits that depend on it.

What's going away

These will sound like a collection of 1990s and 2000s hardware that used to be relevant but aren't anymore.

The platforms losing support are the SA1100 boards, the Acorn RiscPC, Footbridge, the legacy PXA board files, and the Orion, Dove, and MV78xx0 boards.

The same series also takes out the OMAP24xx, i.MX31, and no-MMU i.MX chips, LPC18xx and Axxia, along with the STM32F4, F7 and H7 microcontrollers, Versatile MPS2, and AT91 SAMV7.

Of course, this doesn't mean that these boards will suddenly stop working (stating the obvious, I know). Any one of the devices listed above just won't be able to run a mainline kernel from 7.4 or 7.5 onwards since the code that lets them boot is slated for deletion.

Another thing worth knowing is that in the same 7.3 cycle, Linux has gained support for other Arm-powered chips like Apple's M3 Pro, Max, and Ultra, Samsung's Exynos 1580, and Qualcomm's Shikra.

This shows the continuously shifting nature of the kernel, which has to regularly adapt to new hardware while also ditching old support that isn't used by many.

Linus has been here before

If you think about it, Linus Torvalds has been a strong proponent of removing such old code.

Back in 2022, he asked whether it was time for the i486 series of CPUs to be dropped from Linux, seeing that i386 was already gone in 2012. By 2025, Linus was saying that there was no real reason for anybody to waste their time developing for the platform.

So I anticipate that he will accept these removals, unless there are any shortcomings that have gone by unchecked.

Even so, old hardware has its advocates. People and enterprises do keep these machines running, and Arnd has left the window open for anyone who wants to argue against their removal before the drivers go.

Broadcom has taken down the public download pages for the VMware Virtual Disk Development Kit (VDDK). The pages for VDDK 8 and 9, along with the landing page, now return a generic error message.

So far, there has been no formal announcement from Broadcom.

The removal was carried out sometime last month, as people on Reddit and vendor forums started reporting dead links in mid-August. ShapeBlue, an Apache CloudStack services firm, documented the change on August 25. Red Hat followed days later with a support note for its Migration Toolkit for Virtualization.

While existing copies will still be functional, meaning implementations that currently use VDDK will continue working as usual, those who want fresh copies are out of luck.

What's the fuss about?

The VDDK is a library that lets backup and migration tools read VMware virtual disks from outside the hypervisor. VMware launched it in 2008 as an easily accessible SDK for customers building disk utilities, and it soon became the backbone of image-level backups and agentless migrations.

Unfortunately, its license makes things complicated. The library cannot be redistributed, so no Linux distribution provides it, and no open source project can bundle it for users. Everyone has to point to Broadcom's download page, and that page is now gone.

Convenient, not for users but for VMware's owner, Broadcom, who has managed to cut off public access to a tool many relied on to move away from the platform.

The ones who are stuck

Microsoft's Azure Migrate documentation is one example of the disarray caused by the silence. It tells admins to download VDDK 8.0 or 9.0 from the Broadcom portal for agentless migration, and the links embedded there point to a page that isn't accessible.

The docs suggest falling back to agent-based migration if VDDK is unreachable.

Red Hat is in the same boat. Its support note reports customers hitting "not found" or "access denied" errors while trying to access the public download links for VDDK, further noting that it cannot host or redistribute the kit.

They are directing their customers to contact Broadcom support while the Red Hat Engineering team hunts for an alternative, long-term path out of this dilemma.

On Reddit, a user, bobdrakken, already did that, and what they received as a response ought to show you where Broadcom stands on this. 👇

Hi ****,

Greetings from Broadcom Customer Care.

To ensure the highest standard of security, reliability, and product features, the Virtual Disk Development Kit (VDDK) is no longer available for use or download.

Broadcom continues to actively maintain a variety of APIs and SDKs to enable authorized technology alliance partners to build backup and recovery software solutions that are validated and authorized to integrate, interoperate, or be compatible with Broadcom products.

For all backup and recovery requirements, customers should rely on such solutions from Broadcom authorized technology alliance partners."

I hope this helps you.

Please get in touch with your Partner for better assistance, since there is no pending action from our end hence this case will be archived now.

I would be glad to assist you if you have additional support requests.

A pattern unfolds

This takedown fits a pattern running since Broadcom closed its $69 billion acquisition of VMware in November 2023. The subscription-only overhaul followed within weeks, ending perpetual licenses and folding standalone products into bundles.

European customers later reported renewal prices climbing massively, enough to draw formal complaints from industry bodies.

It's not entirely been about extracting the maximum out of VMware; Broadcom has thrown in some goodwill along the way too. Workstation Pro and Fusion Pro went free for everyone in November 2024, and a free ESXi build returned in limited form in 2025.

Each move can be defended on its own; licensing overhauls and price changes are normal business; so is tightening download access.

But taken together, these paint a picture that while VMware itself is running just fine, the ones trying to move off VMware need to jump a few more hoops before they can achieve that.

Via: Virtualization Howto

Seeing all the hype surrounding AI in the Linux ecosystem, you might've asked yourself, What makes a Linux distro "AI-focused"? If you go looking for an answer, you will get all kinds of results depending on which project you look at.

One wires an AI agent into first boot, right next to picking a desktop environment. Another keeps every model running air-gapped with no network access.

There's no established approach here; every project is going at it differently. The distros listed below are each building toward their own idea of what an AI-native desktop should look like.

1. Omarchy

While Omarchy started off as a simple script to get Hyprland running on a stock Arch install, it has since grown into a full-blown, AI-agent-first Linux distribution with its own foundation and eight-figure funding behind it.

The project markets itself as "the malleable OS for the age of agents," and that framing shows up the moment you boot it for the first time. Setup asks you to pick a default AI agent from a list that spans Claude Code, Codex, Copilot CLI, and several others, then signs you in so it can start working on the system right away.

Once configured, the agent can then help you troubleshoot crashed programs, build out plugins and themes, and put together apps from nothing more than a description of what you want.

And underneath all of that is Arch Linux, so you know you are working on a proven base.

2. Shadowfetch Linux

Shadowfetch Linux is built around something it calls "Mission Control," a system for giving an AI agent a task and then actually checking its work.

Nothing from an agent locks in without your sign-off. Once a task is completed, Shadowfetch lays out everything that happened, what changed, which tests ran, what the output looked like, and waits for you to approve it.

It comes with two modes, "Fire," which keeps the system connected to the internet, handing over tasks to cloud agents. While "Ice" works offline, relying on local files and standard media export tools like FFmpeg instead, though it doesn't run local AI models yet.

This distro also limits what an agent can actually touch with a feature called "Firebreak" which keeps the agent inside a defined boundary where it only ever sees the project it's working on and whatever system files that task calls for.

3. Gnoppix AI Linux

Gnoppix AI Linux, or Gnoppix Linux as it used to be known, traces back to 2002, when its founder Andreas Müller built it on Debian 3.0 "Woody." That work got him hired by Canonical as one of Ubuntu's earliest developers.

Anyhow, Gnoppix keeps AI local and off the network by default. Its homepage showcases a Mistral 7B model running locally, with it being "AIR-GAPPED," meaning that the model has no network access.

That same privacy-focused approach carries through the rest of the OS, with no telemetry, a built-in encrypted messenger, a VPN, and a search tool that strips identifying details before a query goes anywhere.

There's also a space for testing and fine-tuning models like Mistral, Llama, and Qwen on Gnoppix's own servers without needing the hardware to run them yourself, and the project itself is Debian-based.

Official material describes the project being run out of Panama with an AI lab in Japan, and there are many pricing tiers for this one, so choose the one that suits your use case.

4. MakuluLinux

MakuluLinux comes with Electra, an AI assistant you control from a floating bar at the top of the screen. Type what you want in plain English, and it figures out whether that means chatting, writing code, drafting an article, or running a system command.

Electra isn't locked to MakuluLinux either. It installs on Ubuntu, Debian, Arch, Fedora, and a few other distros too; MakuluLinux just ships it by default. The developers say that they built and operate the AI backend instead of wrapping someone else's API, and there's no key needed to use it.

If you lose your internet connection, it switches to a local model automatically and can even remember past conversations.

Though the company's pages don't agree on how many models are actually available. The number ranges from the low seventies to over ninety, depending on which page you check.

5. Deepin

This distro can be regarded as one of the early adopters of AI in a Linux distro, while most of this list only started moving in this direction over the past year. deepin weaves its AI layer, called UOS AI, directly into the browser, mail client, and system-wide search instead of keeping it in a separate app.

That search piece, branded "Grand Search", is the most visible desktop-focused part of it, and deepin's most recent stable release, 25.2.1, pushed it further with something called Intelligent Search.

Instead of remembering an exact filename, you can type a phrase like "financial spreadsheets modified yesterday" or "videos larger than 1 GB," and the system parses that into filters for time, file type, location, and size on its own.

Honarable Mention

Ubuntu isn't trying to be an AI-native distro yet, but it's taking the first step toward being one. Ubuntu 26.10, which is scheduled for an October release, is going to feature its first AI-powered feature.

Myna is a dictation tool that runs entirely on the machine itself, triggered via a keyboard shortcut, dropping transcribed text into whatever you were typing into without needing an internet connection.

This comes after Canonical's VP of Engineering, Jon Seager, revealed their plan for splitting AI work across Ubuntu into two categories earlier this year.

Thanks to Joey Sneddon of OMG! Ubuntu, we now know that local speech models for Myna will ship as separate, removable snaps matched to your hardware, some built for NVIDIA GPUs specifically, others CPU-only, with broader language support planned for later.

If you aren't a big fan of AI in your Linux distro, then I recommend going through ZDNet's roundup of the distros that plan to stay that way. There are six of them: AerynOS, MX Linux, Debian, Arch Linux, Slackware, and Void Linux.