LINUX Unplugged - 681: Ain’t Nothing But a Syncthing

Episode Date: August 23, 2026

Syncthing saves the day in an unexpected way, and we dig into what’s new in Linux 7.2.Sponsored By:Jupiter Party Annual Membership: Put your support on automatic with our annual plan, and get one month of membership for free!Managed Nebula: Meet Managed Nebula from Defined Networking. A decentralized VPN built on the open-source Nebula platform that we love.Support LINUX UnpluggedLinks:Web Boost — Send us a boost via sats or USD💥 Gets Sats Quick and Easy with Strike📻 LINUX Unplugged on Fountain.FMJupiter Garage SWAGAppleTalk 1985-2026 Memorial StickerSorry, I only open regular files StickerLongtime Linux CIFS/SMB3 Maintainer Passes AwayLinux 7.2 Kernel NewbiesCache Aware Scheduling Merged For Linux 7.2 For Boosting Modern Intel & AMD CPUsLinux 7.2 To Revert Back To The FIFO DRM Scheduler Policy Due To "Fair" Regressions[REGRESSION] drm/sched: FAIR policy causes serious performance degradation at max GPU load on 9070XTreduce pipe->mutex contention by pre-allocating outside the lock — While profiling Meta's caching code, I found pipe->mutex contention on the hot path. anon_pipe_write() currently calls alloc_page() once per page while holding pipe->mutex. The allocation can sleep doing direct reclaim and runs memcg charging, which extends the critical section and stalls any concurrent reader on the same mutex.Linux Finally Eliminates The strncpy API After Six Years Of Work, 360+ PatchesSpecially Crafted NTFS File-System Image Allows Root Access On Linux With NTFS3 DriverThe beginning of the 7.3 merge windowLinux 7.3 SMP Improvement To Help Reduce Latency, Improve Real-Time PerformanceLinux 7.3 Expected To "Flatten The Pick" For Better Scheduling While Gaming & MoreLinux 7.3 x86/mm Lands Patches To Greatly Improve Latency-Sensitive WorkloadsSched_ext's Sub-Scheduler Support Now "Feature Complete" With Linux 7.3Btrfs Ready With More Performance Improvements For Linux 7.3: Some ~3x To ~5x WinsLinux 7.3 binfmt_misc To Allow BPF Programs To Dynamically Choose Execution EnvironmentsLinux 7.3 drops support for two ancient file systemsLinux 7.3 Network Changes Merged But Developers "Completely Overwhelmed" Due To AI/LLMsTwo Very Exciting Memory Management Optimizations Going Into Linux 7.3Syncthing — Syncthing is a continuous file synchronization program. It synchronizes files between two or more computers in real time, safely protected from prying eyesSyncthing GitHubRclone — Feature-rich alternative to cloud storage vendors' web storage interfaces.LocalComet — A private, local-first Windows AI assistant built and validated with Codex.A look at the Quickshell desktop-component toolkitMCP-NixOS - Model Context Protocol for NixOS — Because your AI shouldn't hallucinate package names.dataiku/kiji-proxy — Privacy proxy for your OpenAI requestsRockbox - Free Music Player FirmwarePick: ssh-clipboard — Peer-to-peer clipboard sync over SSH. Copy anything on one machine, paste on another. Open source, and Written in Rust ackchyually.

friends and welcome back to your weekly Linux talk show. My name is Chris. My name is Wes. And my name is Brent. Hello, gentlemen. Well, coming up on the episode this week, the kernel release that deprecates our childhood as landed. There's a lot to like. There's one thing that really stings.

But 7.3 is looking like such a banger. I'm not sure if I'm even going to care. We'll tell you more about that. Then, how sink things saved the day this week. Plus, we have some great boosts and picks that you're going to love. And a lot more. So before we get into all of that, let's say time-appropriate greetings to that virtual lug. Hello, Mubaru!

Hey, please. Hello, Brent. Hello, and hell up there in the quiet listening. Nice to see you up there. You guys looking good up there, handsome as ever. Go say good morning to our friends over at Define Networking as well, define.net slash unplug. Go meet Manage Nebula. Decentralized VPN built on the open source Nebula platform that we love and use.

Nebula is so awesome. Oh, I'm so, so grateful for Nebula, especially as. I have been having issues with other mesh networks, it really is so superior and simple in its way and it's in its design too, yet extremely secure. And I love that I have my own lighthouse, so I completely control the infrastructure. I own everything. But like friends and family, I can hook them up with defined networking's managed nebula. They automatically provision a lighthouse now for people, but you can also run your own.

It's really great. It's a great company. It's a great product. It's open source, independently audited, and it's built with modern encryption. So go get started up to 100 hosts absolutely free. No credit card required. Go to Defined.net slash unplugged. Redefine your VPN experience.

See why I love it. Own your infrastructure. Defined. Dot net slash unplugged. Thank you to Define for sponsoring the unplug program. Well, with the release of Linux 7-2. It is the end of Apple Talk and it is the end of our sticker. We'll have to remove it from the store this week.

So you want to get your commemoral sticker. Now is the time. It's like we're making a coin for Apple Talk. it's a sticker. I'll not dwell on this. The audience knows how I feel about this. But get him while you can't. Also, we wanted to make a mention of the passing of a longtime Samba contributor, Steve French. He passed away recently after dealing with an illness. He's been around forever. He's been contributing since 2002. He worked at Microsoft. He worked at IBM. He's worked in Azure. He worked on my favorite project, KSMBD. And so he had to step down for health reasons this week.

and then he did pass away. Arkin Dolan says. Yeah, really. A long time contributor. So our thoughts are with his family. And they do have a new maintainer stepping up for SIFs. So there will be.

And I guess this individual has a history with NTFS, the rewrite. Yeah, we have Paulo Alcantara and someone we've mentioned multiple times on the show, Namjjjian. Those are now, those folks are now co-maintaining CIFS, SIFs, which is NamJ's third file system, the NTFS rewrite, XFAT, and now this. So it sounds like it's in pretty good hands. It does indeed. But we did want to note because, you know, it's part of the community.

And I think it's worth observing and our thoughts are with the family out there. And it is good to see the project continue to get support as well. This is how open source gets made. Mm-hmm. Mm-hmm. All right. Well, then, let's get in to Linux. Speaking of the kernel, Linux 7-2.

So just as we finished up the live stream, I think Linus probably got done listening to the show, clearly. Obviously. And then decided to release. Oh, better get some work done. Yeah, pushed out Linux 7 too. I don't think he wants to push it out before the show's over because he doesn't want to be distracted. I understand that.

It might distract us, too. And this was classified as a very busy kernel cycle release due to increased patch and reporting turn due to AI assisted development. We do have a quote from Linus. You want to take it? Well, this last week of the release was, once again, bigger than I would have wished for. But hey, with the whole quote-unquote new normal thing, if I delayed releases for that reason, we'd probably never have a release at all. There are a number of fairly late reverts. The DRM scheduling reverts stand out, but there's a few other ones in here, too.

It may not be pretty, but it's the correct way to deal with, oh, that code wasn't ready and caused problems. People will be trying that again later. But while the DRM reverts stand out and are some of the bigger patches in here, there's a lot of small fixes all over, mostly in drivers, but networking remained fairly active this last week too, and then there are a few random patches

to architecture files and the Perf core. With this, the merge window for 7.3 obviously opens tomorrow, and I already have 40 pending poll requests. I appreciate all the timely people. You know who you are. Yeah, we'll get into 7.3 here in just a moment. You know, when Linus calls out a subsystem,

it's probably worth paying attention when he calls out networking specifically, it's probably worth paying attention. So I think we'll get to that. But the thing that you and I have been watching and something that I think we talked about before on the show, so we'll just give a brief mention, is there's been some real improvements to, quote-unquote,

schedulers, CPU schedulers, GPU schedulers. The scheduling system and all kinds of stuff. Scheduling got a big work, a lot of work in 7-2. So we're having now more cash-aware CPU scheduler. Yeah. This is a pretty big deal. And it's almost surprising in a way that we haven't had it already, right? Like, caches are a huge part of CPU and computer performance, of course.

And these days, especially on systems with like multiple dyes, right? It's not just one giant CPU, right? It's like multiple CpUs, each with their own last level cache right next to them. And so understanding, like, where is a process or a thread running, and then what caches is it near? and then what data is in those caches can be a big deal. So if you think about something like a database, right,

maybe you're doing a query where you have a lot of working memory that is, the whole point is you want a query optimizer that can parallelize things, right? So you're not waiting on trying to seriously do stuff. So maybe you have a list of things you're trying to look up. Well, both sub queries

are going to need to be accessing that data, right? So if one of the subquiry threads is running on a different CPU core, suddenly it's got to do the inter-core communication to go get the data from the other cache. So you don't really want to do that. So now the goal is upgrade the scheduling system so that it can figure out, like, it starts tracking stuff in a multiple way. So one is just obviously sort of threads of the same thing, right?

But it can figure out when things that are running need to share the same cache and then try to schedule them on the same cores appropriately. That's a nice one. It's taken, there's multiple steps to it. They're learning a lot, or they're taking advantage of a lot of neat stuff that happens. Like the hardware actually has cash monitoring stuff built in. So then, like, the first level is basic structural stuff. It understands where the, you know, what the system has and the basic sort of stuff, basic, like, schedule threads together.

But then you can actually go check to see, like, what the layout the hardware tells it is doing in real time. And it can even go sample the misses to figure out, like, oh, we tried to hit something in the cache, but that's actually on the other system. Really? We might need to start considering if this continues actually moving it to the other core. Wow. Now you don't want to do that too much, right? Because then you could just be bounced in a amount around all the time.

That could be a problem. So it's a delicate balance, as is everything in scheduling. But this is a huge upgrade for just it being really smart. Now, it will matter more on, you know, big systems with lots of CPU cores and that kind of thing. So you might not see as much of an impact on your single core laptop or whatever. Yeah, yeah. But it's a big upgrade for our friend the Colonel. Linus had made a comment that it may not be pretty, but it's the correct way to deal with,

oh, that code wasn't ready and caused problems. People will probably try that again later. Is he referring to the GPU scheduler stuff that got reverted? Yes. Yeah, okay. 7-2 was meant to switch the DRM scheduler from first in, first out, to a new CFS-style fare policy. So we already have this on the CPU side. This was bringing it over to the GPU side.

But then an RX-90-70-XT owner hit a sustained 100% GPU load. Uh-oh. Yeah, games dropped to 10 FPS. The Wayland session froze. Had to reboot to recover. And then here's the part that maybe wasn't as good of an idea. When this happened, it wasn't just that the default was swapped. It was the only schedule.

The other stuff was ripped out, and all you were left with is fair. So the first in, first out, and the RR policy, those weren't there. There was no runtime escape patch. So five days before the release, someone from Agalia shipped a 20-patch revert, smoke-tested on a steam deck, and it ships first-in-first-out as the default again, sort of unrips things out, but now it still adds fair as an opt-in experiment. So instead of doing a big change to making this default, now we're still in sort of test mode. All right.

That makes sense. Thank you. Thank you. What is nice, though, is the reporter AB tested some of the fixes himself to sort of, you know, this is a very active collaboration process, including, quote, the patch Claude gave me. Ah. Also, I think MongoDB users will be happy to hear this. There was some work in Linux's memory management.

The multi-gen LRU memory reclaim system has been cleaned up. And as a result, kernel developers have reported performance improvements approaching, so, you know, results are very, but approaching 30% in some MongoDB workloads, along with fewer unexpected out-of-memory situations. Ooh, I like that one. Some of us have a lot of tabs and it gets dicey. My bet is not everybody's going to see that 30%, but heck, even a 10% is pretty great,

just by a better memory reclaim. And Brent, I know you're going to love to see this one. Faster, leaner swap in Linux now. Delicious. Are you guys still using the swap file? Well, I have one. I don't know that it gets used a ton. That's what I mean, really, is are you actually,

like when you install your Linux box, are you deploying a swap file? Not are you actively using it necessarily, But I mean, any system that runs for a really long time generally swaps a few things out. Yeah. But yeah, I have 3.7 gigs used 12 gigs free in my swap right now. My swap is zero currently, which is, I think, a little unusual. But I use it heavily for hibernation.

Like sometimes when I'm traveling with the laptop and stuff, I want to maintain as much battery power as possible, depending on what I'm doing. Maybe I'm going on a train at some point of the trip or something like that. And so having hibernate available for me is quite valuable. So that's the main reason I have it. Yeah, that's one of the reasons I have it. Yeah, that's one of the reasons I have. have one too. I also find it better for crash debugging and also, you know, some of my systems only have 16 gigs. Right. So I do actually use the swap file. So it's good to see that there

is ongoing rewrite of the swap table infrastructure in Linux 7-2. The improvements reduce the amount of static metadata Linux needs to keep around for swap and improves how that metadata itself is managed. Nice. Yeah, one example here, a system with a one-terabyte swap device can save roughly 512 megs of RAM just from the reduced metadata overhead. What? Which is, you know, nice. It'll add up over time.

That's a massive savings with how expensive RAM is right now. That's no joke. So again, a system with one terabyte swap device, which is a pretty big swap device, will save roughly 512 megabytes of RAM just from the reduced metadata overhead. Does anybody out there in the audience have a terabyte swap device, boost.orgia broadcast?

Yeah, right. You got to tell them. And why? How'd you get there? Why? Mm-hmm. Mm-hmm. Mm-hmm.

All right, we've got a couple other items. There's one I really want to talk about. Is there a couple before we get to the file system stuff that you want to talk about? I know we have faster pipes. Yeah. Yeah, we do have faster pipes. We covered a bit about this in 6-7-2. Yeah.

There was basically going to be improvements to make piping things between processes faster, kind of all around, especially on the right path. And now we have a few numbers under, memory pressure in particular, throughput is up 21 to 48% and rate latency is down 17 to 33%. Whoa, this is a...

Yeah. Yeah. And I mean, like, pipes and you know, FIFOs are used all over the place. Yeah, yeah. I find it so crazy how, you know, Linux isn't just a baby anymore and yet we're getting insane numbers of performance upgrades

in seemingly every single version. It's just bonkers. It really is like, I don't like, was this all just there for the taking all along? And you just, you work through it, I suppose. Brent, I'd like to see you test this next one somehow. USB 4 stream. So in 7-2, you now, two machines can send raw streams of data

directly over USB4 or a Thunderbolt cable. Well, who needs Ethernet anymore, I say? Well, it is actually pretty fast. It is quite fast, actually. It's neat, too, because, like, before, you know, you could obviously do this and then, like, stand up your own little local network or whatever and set up IPs and, you know, get it going.

You could do it. But this, you just get, like, slash dev slash TB stream zero device. And you can just cat and read to write or DD or whatever and it just shoots it across the USB cable. It's going to be, you know, because you don't have the overhead of the protocol of Ethernet. Nice. Oh my goodness. And it can also still coexist with normal Thunderbolt networking.

So it's not like it's the only thing you can do on the cable. Okay. All right. Well, while we're talking about performance, I think the thing that I'm probably the most excited about because it'll have the biggest practical impact for all of us is butter. FS is getting faster in this release. Butter FS is getting faster. We talked about large folios in the last episode.

They're enabled by default now. There's also experimental support for huge folios up to two megs. But I think the big wins come from better direct I.O. performance, faster sequential rights and a new interface that exposes the raw checksums to user space, so that way you can build tooling around that and all of that. Yeah, including things like perhaps deduplication bread. Oh, hello. So, you know, this is the beginning of an API.

right? That's a beginning of an API. But the direct I.O. performance and the sequential rights are resulting in real performance improvements for ButterfS. I think we'll probably see Larval benchmark that pretty soon. And I'm going to keep an eye out for that. So I think that's probably the thing in 7-2 I'm the most excited about.

So people upgrading soon on rolling releases, you know, do they, I don't know. You probably have to do something to upgrade your ButterfS format. I don't know. But it's something to keep an eye on. Something I'll probably play with and I'll update you. There's one here I'm excited about. Inline storage encryption. There's a new device mapper target called DM inline crypt.

It's designed to take advantage of hardware that can perform inline block device encryption rather than doing all of that encryption through the traditional DMCrip path. Based partly on work that happened in Android of all things and could eventually become a useful alternative to DMCrypt on hardware with dedicated encryption support. You know how I like that encryption. On mobile, you probably do have opportunity to have custom stuff. and not as much CPU as you might want. And they have been tested in the fire.

Android did not start super strong in this area and they've made a lot of progress. And hey, look, it turns out that having Linux on your phone actually pays dividends to like that the whole kernel could use, proving out the entire model that we've been saying all along. It's also worth pointing out, not that you guys don't know this, but it's worth just reminding us all that these tools like this, if they're being actively supported and contributed to by a large, well-financed group of people, it's probably, you know, something to consider just because you want tools in this category, that are being actively supported and maintained.

There's a few other things in here, but the one that really jumped out to me is faster prock. Like, when we're talking about, this is really the performance kernel. Yeah, it really is, right? This is silly, but like, what? Faster slash prok?

I didn't know that was an issue. Yeah. Proc exposes a whole bunch of metadata and info about, like, the running kernel and all that. And there's a bunch of programs, especially command line programs that want to go read that stuff to actually go, right?

Because it's a way that you don't have to have, use syscalls to try to get the same information. You can get it right out of the file system. It's super handy. Like a file. It's standard Unix. Yeah, right? So there's a bunch of stuff that uses it.

And so if you make that faster, you make all those things faster. Yeah, absolutely. It's like, well, it's like Brent said. It's like, we're just seeing these optimizations and areas after all these years. But who thinks Stephen go down there and be like, huh, what could I do here? A clanker? I don't know. Maybe a clanker?

I don't know. Some autist, I don't know, maybe. Like somebody who's really passionate? I've got a few paths and slash, There you go. Well, we should probably mention we now have the safer open-at command because we have a sticker based on it. Right. Or we only open it if it's a regular file.

That's right. Only regular files here. Only regular files. And this has shown up in a few different vulnerabilities. This prevents an application from unexpectedly being redirected to something like a pipe or a device node when what you want is like a text file. Yeah. I think that's a good thing to call that. There's so many things we could get in.

to here with this. It's like this kernel isn't a headliner banger feature. This is the kernel feature. It's all these things from like CPU schedulers to GPU schedulers to file systems to drivers and networking that have seen improvements. Or just new little features like the fact that if you had Thunderbolt or USB4, you can now plug machines directly into them. That just kind of got a little bit better. So it's, I mean, 72 is a really good release. Is there any word if there's LTS talk around this one? I don't know. That's a good question. I've seen discussion around a few possibilities from this one or maybe 7-3 or 7-4, perhaps. Well, 7-3 would be a wild one to make an LTS. There's a lot going on. Let's just talk about this really quick because it's in early development. The merge window just opened last Sunday, obviously.

But there's a lot more scheduler work in here. A lot more scheduler work. Yeah. Right now, there's some focus particularly on how scheduling interacts with C-groups, because, of course, the control group mechanism is involved in maintaining a lot of access to different resources, which impacts how you want to schedule things. So instead of having to traverse a bunch of hierarchical run cues,

it's moving to a single run queue with new methods for distributing C group weights. There's also a new C group mode knob. You can do SMP, Up, Max, Conquer, Tasks, concurs the new default. And this is basically a mode you can choose so that you pick how much CPU weight, right? Like how much your share of the CPU is going to be

gets divided between the different groups. Okay, that seems great. Especially if you care a lot about, you know, scheduling things, and especially C-groups are used a lot in our buddy containers. Yes, in our friend. Yeah, the hierarchical version had a lot of long-standing oddities, including how nice, you know, the nice command priorities, got handled across groups.

So that might improve from some of this work, too. Oh, okay. Yeah. Early testing on an old Sandy Bridge machine paired with an RX580 showed enough improvement that the work has been pitched partly as a win for gaming on lower end hardware. Nice. So it might not start thinking about gaming a low and hardware, but that's where it ends up with the kernel sometimes. Well, I mean, a lot of those handheld devices, those little purpose built, purpose built? Or no, that makes sense.

Hmm. Okay, now, this next one I can't even believe. So I'm going to say it out loud, but I still don't believe it. Okay? In Linux 7.3, in speaking of schedulers, there appears to be up to a 90%, I'm going to say that figure again, around 90% reduction in a nasty source of X86 scheduling latency. Here's what strikes me about this. Number one, 90%'s a big number. Number two, X86 is a big platform.

This isn't arm. This isn't like some sort of maybe risk chip that nobody's using. This is X86. What? A 90% reduction in a nasty source of scheduling latency? Now, I realize this must be an edge case here or something, right? Because this is massive. Yeah, and part of it is just sort of how we've been doing things. So, like, back to, you know, you have multiple cores and you have multiple caches.

And one of those things is the page cache, right? Sort of like, how do you have the files, what you've copied off the disk to memory for working with? Well, those different page caches can get out of sync. And so you need to maintain coherence across all of these things. So you do what's called a TLB shootdown, which is where the one that figures out that something needs to be invalidated, sends out interprocessor interrupts to all the other cores to say, like, hey, you got to take this out of your cache and go get it fresh because it's no longer valid. Okay. Well, the calling CPU can wind up sitting with preemption disabled while waiting for the remote CPUs to finish that operation. And so ByteDance actually observed weights causing up to 16 milliseconds of scheduling latency on a 16 core system just from this.

So Linux 73 changes that mechanism, so CPUs can now be preempted while waiting in these situations. And in bite-dance further testing, the P99 maximum preemption-disabled interval fell to roughly 1.5 milliseconds. And which the patch series described is producing as much as a 90% latency reduction. So there's your 90% in the affected workload. Wow. With one particular workload, which may not be yours, right? But it is aimed at, in particular, at latency-sensitive and real-time style workloads, which you know Linux is doing more and more of,

and it removes an architectural source of scheduler stalls that can affect ordinary machines too. So you heard Wes there, too, he mentioned. This is patchwork from ByteDance, and this is now merged for Linux 7.3 after 10 rounds of patch reviews. But I don't think this is the only set of patches that ByteDance has committed recently.

It seems like their contribution to the kernel is kicking up. And is ByteDance? Are they the TikTok company? Yeah. Right? The TikTok company? company? What?

I mean, you know, we've said it in the cliche way, cliche way of, right, like, I don't know, meta runs Linux servers or whatever. But I think there's a real component where, like, if you're trying to do media distribution at scale, you really push what computers can do. And that means you might need to interface with the kernel. Whether you like the content or not, it's up to you. So this is the world we live in now. And, I mean, that's a serious contribution.

That's a real contribution. Something we've talked about on and off for a little while now, but really seems like it's coming together in Linux 7-3, is this subscheduler idea. So cool. So cool. So I think this could be one of the more consequential long-term changes because you're going to see vendors take advantage of this.

Network vendors, hardware vendors. So they have now, I guess, like a subschedular system where you can build your own sub-schedular for Linux. Yeah, so you have a root scheduler, and that can now give a C-group sub-tree, back to the C-groups, to another scheduler, including with revocable access to individual CPUs,

you can control what it's scheduling over. So in practical terms, you could imagine different scheduling policies for a few different workloads on the same machine, right? So maybe you're trying to do latency-sensitive audio capturing. Yeah.

But you've also got a background job doing some transcription in bulk, right? Or syncing or whatever you just have running. Or you're trying to do games, but you've got containers that are running that are doing image processing at the same time. As one does. Or I can see this a lot, right?

You might have, you're running a, a web server, right? And you might want a particular scheduler for your database that's different from how you want to run it for the EngineX reverse proxy. Or, you know, to bring it to the desktop, my kid, right, constantly he's got the Discord going with like 15 people in the call. He's got the game going. He's got OBS recording and streaming to the call. And you know, and you want the game to perform well at the same time. And like that, that seriously taxes the system when you're doing all those things. So you could see a world where there's a SteamOS scheduler, sub-schuler. And this sort of scales with how much investment you want to put, right?

Like with this Sched X stuff, that, of course, is built on being able to write like BPF programs that can do scheduling, which is furthermore means you can take a standard Linux kernel. You don't have to build your own modular. You don't have to do potentially dangerous and, you know, require lots of testing changes on general scheduler algorithm changes. You get to implement your own, just, you know, adding it in from user space, and then get that attached to particular programs and workloads that you need to run. Less powerful. What other OS is doing that?

All right. So Brent, you know, how much would you pay? We told you're going to get performance updates to ButterFS in Linux 7-2. Yeah, delicious. But how much would you pay for a whole other round of large performance wins in ButterFs in Linux 7-3? Because, buddy, I'm going to give it to you for the price of free. Yeah. Another round of ButterFs optimizations will be coming in Linux 7-3.

And some of them are kind of a big deal. I mean, I'm sure you went through this, Wes, but DirectIO again, get some attention. Indeed. ButterFS can now use IOMAP to bounce buffers instead of failing back to a buffered IO. Don't know what that means at all, but it seems good.

Yeah, well, it's, you know, buffered I know that. I know what a buffer is. Well, and so DirectI. I.O. is unbuffered, right? And so now it's basically a setup where in some situations, direct I.O. didn't really work.

You had to fall back to using the slower buffer. stuff. Now there's that, in that same situation, Butterfest is internally using an I.O. Map bounce buffer of its own that lets it stay on the more direct Iopath. But gives like direct I a little buff it if it needs it? I think so. Huh. Okay, thank you. All right. See, now I get it. And that's why that, uh, the case that was sort of the study for this. Yeah. Report reportedly goes from roughly 50% of theoretical throughput up to about 95. What? I'm sorry? That's massive again. So a 45% increase. Are they using the same math? that we're using.

This is really something. I mean, just in the next couple releases, there are so many performance wins. Yeah, really. I mean, there's more removing an unnecessary one jiffy delay in a non-sSD logging path. That's like a 5x throughput improvement

in one test. Really? And then another is interesting. It replaces the use of an X-array data structure with a local, least recently used, LRU list for inhibited extent buffers. That one's a 3X improvement. All right.

I mean, it's funny that a, three, it says three to five X by the way. But it's funny that a three X improvement seems kind of tame at this point. I know. It's just so great still. Ah, I mean like, where you at, Butter Fs, Hater? It's time to update. Time to update.

All right. There's a several other things in here we could talk about. Yeah. How do you like the name? Fail F.S. I love it. Seems like one of our challenges is coming up. I feel an episode already.

I hear the name and I'm already picturing the episode. Yeah, it's great, right? So you know what you love about five. file systems, they let you open and write to files and do stuff with files, right? Very handy. Yeah, well, Failfess is a file system where every possible operation fails. Okay. All right, all right, cool.

So what's cool about this is it's a security idea, right? So what you do is you open a file and you get a file descriptor, which is basically like a handle to the file, right? And after that, like, you've done the security checks, the colonel seen if you have permissions, all of that stuff has already happened. And then you have a handle you can use to read and write from the file. what failfess lets you do for a security system perspective is you can, if you know all of the files

that something needs to access, you pre-open them, and then you launch something and you give it a fail-fs file system, and that means there's no way to ever escape its sandbox. All it can ever use is whatever FDs you pass it when you spawn it. Oh. So it's an idea to like pre-restrict it.

So it's not in a true route, it's not stuck in some folder, it's not in a weird mount. It just, it, the only file system it knows about is not the root file system at all. It's fail. F.S. That's so funny. It's such a funny way

to solve the problem. It's interesting because it builds on a few, I think in 617, maybe we talked about null FS, which was something cleaning up sort of early pivot route, switch route stuff, where you had just a super minimal file system that can be the root of the whole tree of file systems. Right. So FALFS is kind

of a similar like, not really quite a, it's like more of an abstract file system, but they play important roles in the kernel. That's a good one. We've been talking about file systems a lot this year because there's been a lot a really great news. And I gotta say failifest seems like the easiest file system to implement.

You just make a lot of mistakes for every single implementation. Easy testing. Exactly. Yeah. In this case, if you don't implement a bug, it's itself a bug. I see BPF, BPF is getting some new fancy BPFMT

MISC. Is this a mount? No. Okay, so it's bin format MISC, which is maybe you're familiar, bin format is what lets you basically register different types of binary files. Like you can do this with an app image, right? Or you can do it with a wine, or you do it with like a jar file to launch a Java application.

And so this is where the kernel looks at the magic bytes at the start of a file. That helps it determine what that is. And then you can register executions or, you know, runtimes for that. Well, so there's two pieces. There's actually a NixOS angle here. Oh, yeah.

So right now it's all fixed, right? And maybe you know with NixOS, you have to embed like the dynamic library interpreter, the LD, in basically it has to hard code that in all of the binaries it produces. And this is where the patch elf stuff comes in and all that kind of stuff. And it means you can't really relocate that stuff. It needs this fixed, hard-coded path. What's coming in 7-3 is a bunch of that's being made more dynamic.

And in particular, you could implement an algorithm that was like, hey, wherever I am, go look up in like dot-dot lib from me and just find the, interpreter right there. So then you could start shipping the interpreter that you need to run sort of hermetically sealed in with the rest of your Nix OS closure. Oh boy. And then, so that's one level of dynamic stuff. And then on top of that, you're now

also going to be able to attach BPF programs to do that role of figuring out what interpreter to run. So you could make your own interpreter? Yeah, you can have your own interpreter and you can choose very dynamically at runtime, which interpreter is selected. Wow.

Dang. It's pretty early. And like this is going to, This is like, the kernel opens the features. It doesn't actually add the stuff you'd use. So this will be downstream dependent on distros and users and stuff to actually register the interpreters and like write them and all of that. But I certainly expect we'll see an XOS at least. This is the stuff that makes the Linux kernel the king.

Oh, yeah. It is the king. And the other kernels, they can even with this stuff. It's really something. I suppose in order for the king to maintain its reign, it must shed some older projects from time to time. And in the upcoming Linux 7-3, 7-2 that was just released, but 7-3, which is coming out in October,

32-bit arm support has been deprecated. Whoa. Yeah. So I believe it will survive through 7-3, but it's officially deprecated, which doesn't mean it's gone, but it means it's now marked deprecated. So there's that. One thing we maybe should mention, too, is, which I don't know if it will be fixed in 7-3 or not, but presumably it would be.

There's NTFS vulnerability going around right now. Okay. Essentially, the old, not the new, NTFS, but the older Paragon NTFS. This seems to be live now. Essentially, you can set basically a version of Unix security model permissions,

you know, Chamad stuff, on NTFS files. And that includes the set UID bit. And the old driver and a lot of configurations for desktops will mount, well, like, auto mount an NTFS thumb drive with that set. No way. Yes, so you can have a specially crafted

NTFS thumb drive that gets you root just by plugging it in. Well, I mean, that could be handy sometimes. You know, that's really something. But I think the largest change maybe, it's not a feature that's being added to Lytics 7-3. A third to a half of the patches look like they are AI-generated low-priority fixes that are queued up in Linus's next next patch set. And going through that, they say, quote, we are completely overwhelmed. Yeah, there's a networking maintainer.

Yeah, the networking maintainers. say we're completely overwhelmed. Their counter move is AI reviewing AI with Meta funding a multimodal review for every networking patch, but the practical result seems to be really just a lot of trimming from the kernel of older network

drivers that are not really maintained because they're getting poked at by Clankers. And so they have a whole AI reviewing AI submissions process now thanks to just a ton of compute contributed by meta. Yeah, they describe it as managing patchwork, automating common process complaints, editing commit messages,

and maybe applying patches, which already got reviewed by tags from people we trust. So they're experimenting, actually. Yeah, I suppose they have to with the volume they're going to get, right? Because the Linux kernel has to be the one everybody thinks of when they all of a sudden,

I can build software. I'm going to go to a Linux kernel. Right? A lot of people think that. But it's interesting that 7-2 was widely impacted by AI-assisted development, and 7-3 is both in contributions

and, again, in pull requests and things like that. So it's essentially continuing to escalate, and so they're responding by adding more compute on their side. I think that's going to be a story worth following. We also saw Andrew Morton talking about using Gemini to not write the code, but write some of his memory management subsystem patch summaries this cycle. Look at that.

Well, for years, we've had the project of, well, thinking about first, and then also just tightening up our production infrastructure behind the shows. Big dreams. Some of it's actually happening recently. And Chris, Dropbox has been a big part of that, but I know. We've had issues with cloud sinking in the last, what, several years. And nobody's excited about cloud stuff anymore, and you're making some changes. What did you replace?

Well, so, you know, I've been a, I'm a fan of sync thing. I want to be clear, I've used it for years. When it first launched, I started using it. I was initially with BitTorrent sync, and I was sad when that went away, but I happily migrated. It was. It was great. But Sync Thing has served me well. And I've mostly used it for, like, back-in-large file transfer media. of stuff like that.

And I've held on to Dropbox because we got this sweet 15 terabyte deal. Yeah, yeah, yeah. That's crazy cheap. And how do you say no to 15 terabytes these days? But I hate Dropbox client. I don't really like Dropbox the company. So I kind of compensated and got by for a while

by using the Mistral Open Source Syncline that was really lightweight. Mine, you could run it headless, you could declaratively configure it. Ooh. And it gave me Dropbox, like, back in the OG days of Dropbox again. And it just ran headless in the background and synced my damn files. That's all you want.

And used their storage, right? And that worked well for a couple of years. And then Mistral announced that they're shutting down. And I thought, okay, I'm going to have to migrate soon. Right. So either you have to go back to the standard clients or... Yeah. And then, as sort of like to really poke me to get going,

I don't know, like an old machine joined the Dropbox, I have my suspicions, but all of the the files got a sync conflict in my drop box. No. That doesn't sound good. A little bug in how the exclusion folders work with mistral, and it started syncing to all my systems. Like, you know, terabytes of data. Terabytes of data.

So all my hard drives started filling up, and I realized it. So I first, I shut down the sync client on my, on my workstation upstairs because I'm sitting there. And then it dawns on me. Every other computer is also doing this same thing right now. Oh, my goodness. So then it's like a mad dash before it fills up their hard drive. Right.

Or whatever it might do. Been there. Been there. Yeah. Yeah. That's a nightmare. You've been looking at sync stuff this, or the last couple of weeks, too. Yeah.

I guess a similar problem, actually. I've been trusting cloud syncing less and thought, geez, with these new mesh networks, certainly there's something better we can do, right, and not have to rely about it with an always-on system in the cloud. So you pushed me towards syncing. I've complained for years that I can never wrap my head around. it how it works and how to configure it and stuff like that but uh somehow on my third try around i got over that and it's actually quite simple i think that i don't know what the issue was well the biggest head shift right is um like with next cloud sync or dropbox sync it's user login and then you have

these folders and you sync these folders down and with sync thing there's no user right i mean you can pass-protect the dashboard, but it's these keys that you exchange. And then they discover themselves and they sync and you share these keys and that's how you set it up, which is actually great because it's accountless, which means anybody that has sync thing can just grab that, you can share that key with them and they can sync. They don't have to create an account anywhere. But it's a total flip of the traditional file sync model where you have an account, you log in and it just sinks down your files. But I think both of you and I have found recently that Next Cloud isn't working for us. That might have been the direction I would

have gone for these files because I do really like having a web interface. I think that's really nice. Yeah, I share a bunch of stuff with other people too. Yeah, sharing is also a big one. Yeah, that's a tricky one. The other thing that got me with Sync thing, I think, was the configs for each client. You have to kind of put your head into different perspectives when you're configuring stuff. And also there's this concept of like, I forget what they call it now.

Maybe West, you'll remember, but of like the quote unquote class. but it's really just the amalgamation of all the files of all of the devices into this one like a virtual file system almost that you can think of it that way. And you get to pick and choose which folders you bring down into this particular device. So it's interesting. Very powerful when you start learning the config. It just sort of slips into the background too. It does have a web UI. There are some desktop clients.

But it just sort of slips into the background. So you're not really thinking about it. It's just like you just have these files that persist. across all your machines, and it just sort of does it. It's best when it's at the layer of infrastructure magic. So I had this problem happening, right, where all my systems are getting blasted with files to the point where it was going to affect production. So I shut it all down and figured, I got to redo this, and I wanted to replace it at that point

with sync thing. I just felt like it's been the most reliable, and it was probably my safest bet now that I'm also not using NextCloud. Sort of the way to go. That's a big change, though, right? Because you kind of get, it's a very, it's a very dangerous. different architecture. And in particular, if you don't have a cloud relay, you're going to need a way for these things to talk to each other. That is a thing. That is a thing. I will say, I think, at first I thought

it was going to be a downgrade because of that, because I don't have a web UI, I thought, this is not going to be as functional. But what I realized is that with Sync Thing, because it has an API, there's actually a lot more visibility into it than I have with Dropbox. So with Sync thing, I can see which peers are connected, what folders are caught up, what's still outstanding in the sync, and I can see does this particular host have that file or not? And I can expose that in a way that an agent can read, or a script can check, or a Prometheus endpoint can monitor.

Yeah. And so then it starts to become a lot more like, kind of like this black box of sync and like this production monitor thing. Which is even better than I had with Dropbox. So then it becomes like, oh, I'm not just replacing Dropbox. I'm like improving the functionality here. So I leaned in on that. I really wanted to lean in on making sure my Hermes agents could understand where files were at and where they're needed to be reviewed.

Because it's something my Hermes agents can do because I put this on that system as well. They can write a file to a folder for them. And then they know when the workstation I'm at has received that file. That's really powerful. Yeah, yeah, yeah, yeah. Especially when you want, you really need to be, you know, You know, the number of times where we've tried to sync something at the studio, we've assumed that it synced.

And then we discover later after we've left the studio that, in fact, it did not fully sync. Or like one file out of the seven we needed wasn't there or whatever. Yeah. You just don't want that. Yeah. It's come up. I mean, it's come up recently, too. And, you know, you need to be able to change these things.

You need to be able to make it flexible, too, and all of that. And Dropbox didn't really meet that requirement. It didn't really do it for me. So that part feels like a huge upgrade now. So I think the trickier part, really, was all of the systems. are different. I mean, I have my NIC systems, but then my Aurora host here is running, you know, U-Blue, Silver Blue. The soundboard machine over here is an old Mac, you know, so that's a totally different system. So that part... So it's not one platform that you have to build across. Yeah.

You got to make this work across a disparate set of platforms with different hardware. Yeah. Did that mean you also had like disparate folder structures across those systems as well? Like, how did you implement this in a way that was quote unquote smooth? Yes, it was, you know, like, especially like on the Mac. Oh, yeah, like slash users slash username. Like the whole path is totally different. But then on some of my systems, I have those scary raids. And so I put the sinks on the scary raid.

Several of them. And then like the laptops, there's just in the home folder, because I don't have more than one hard drive in a laptop usually. Usually. So it's just in my home folder. So like literally almost, and then you have my Hermes agent's host, which stores that in a completely different location too. So literally all my hosts.

And that guy's probably the only one who knows where all the paths are. Well, that's the thing. So I focused on the main folders that I use, not all of them, just the main ones I use today. And then I had the agent figure out where they all need to go and which systems we could just hot swap and take over the existing Dropbox location. Nice, right. Because we're working from like a unified pre-model mostly. Yeah. Why sync everything again if I don't have to? Yeah.

But that could also go very wrong on some systems. So on some systems, I had to make the call not to do that. And I had to create a new folder package. Start from scratch. Yeah, yeah, yeah. Okay, well, that sounds like maybe you ran into some issues. Some of this sounds like you've handled it. You've obviously been using the sync thing for years.

But I'm wondering, did anything go wrong? Did you have any scary moments? What was the process like? I think the scariest moment was I forgot to include the Mac. I don't think about the Mac. But you need a lot of soundboard files. That's the soundboard files. Yeah, exactly.

said, I was like, oh, right. But what is really fantastic with this setup is everything's documented, all the endpoints. I mean, the clanker can even connect in and read the database that needs to. So it is really straightforward for just, you go in there and say, deploy it on a new host.

And it, I have some guidance in there, like it caught, okay, well, this folder is your backups folder. That's 20 gigabytes. Do you want to sync that to the soundboard machine? I do not. I do not want to sync that. So that was kind of nice. So I kind of, I have I'm really only dealing the 10 folders that I actively use.

And I'm considering just keeping everything else on Dropbox, removing all the Dropbox clients, and then using R clone to mount it when I do need something from the archive. Because I'm paying such a cheap price. It's like $50 a year or something for $15. That's a cloud backup. So nuts for 15 terabytes of cloud backup. So R clone could be a good way to interface with that.

Can I get access? The Mac was tricky. Yeah, right? Yeah. The Mac was the scary part. because I almost completely forgot it. I just totally blanked it out. Was it smooth for you going cross-platform?

It wasn't too bad. Yeah, I mean, the nice thing is you got home brew on the Mac. You can use that. I think Aurora, so Aurora is a, like I said, it's a silver blue, you blue host, and I went Podman. I just did a Podman container. Nice. A rootless Podman container.

And I don't think Singthing cares. And then I assume it's just riding back on the back of your existing mesh infrastructure, right? Yeah. So a container plus Mac. and there you go. It's really good. I'm really liking it. I find it to be such an upgrade,

and I thought it was going to be a downgrade. Screw that Dropbox. Are you using Dropbox N-Sync thing looking at the same folders? Because I would assume that's going to cause you some issues. Yeah, yeah, yeah. No, I had to be careful about that. It took me a second to realize

I had to shut it down on one of my hosts. That's all takes. But when I scrambled, I didn't just like stop the sink. I'm like, all right, disable the service, rebuild. Like, this thing's gone. I did not want. So I just removed the client. So I don't have to worry about that except for, I think it was the Mac. In fact, I think Dropbox is actually running right now. The damn, I even disabled it from starting, and it still starts. I hate Mac OS so much. So yeah, it's still, it's such a pain in the butt, and I don't think about the Mac. So the Mac is an issue for sure. I'm just going to uninstall it after the show off of here.

and I don't know, I don't really need to access the archive from this thing. You know what I realized? I think I got an art clone on the Mac. We could put Linux on this. I've been waiting to replace this with a Linux box. Right. An actual...

But we could just put Linux on this because it's an Intel Mac. Yeah. Isn't the whole soundboard thing? Isn't that the reason? Yeah, we have to build a soundboard. But we need to do that anyway. We have to do that anyway. We've been saying it for like, what, 10 years?

But what's the point of starting the project if you don't have any hardware to run? Fair enough. Fair enough. Well, let's do that after the show. Right, right. So that's been an issue. You know, but being able to run in Podman, worked out great,

and then it's just installed natively on the NICS systems. So are you managing your configs through NICS as well? Yeah, for those hosts. Yeah. For all those hosts, everything. That's how I started doing it. It's so nice. I even...

It is very nice. Yeah. And it's positive pressure to use NXOS for more and more things. I would... One recommendation if you're going to throw like a mobile device. device in the mix, which I don't know if you want to at this point, but I found it really nice to have the clanker just like make a config file for the Android system and just suck it in. You can import it?

Yeah, you can import them pretty nicely. And that keeps you from like having to thumb all the config options and stuff on the phone. So that would be one of my recommendations if you get there. I wonder, presumably you could also maybe get some port forwards going on your Android and have it API. You're right. Or just it'd be on the mesh too. Yeah, right. Yeah. So I think the, if you do dabble in the clankers, I really want to make a strong recommendation that if you need to solve a sync problem, you look at sync thing because

my favorite part of this is I built a rapper called Hermes Sync Thing. And so the agents are not doing like raw credential passes over the network and like looking things up in the LM every single time. So the agents aren't using like the raw credentials and connecting the API without instruction every single time. The LMs just kind of figuring out how to do it every time when I want to check on the status of stuff.

I've created this little wrapper, and this wrapper automatically has all the presets to go see if all the host is connected, if the folders are caught up, what might be outstanding from any individual hosts? You know, like, for example, what does Aurora still need before we can go on air because it got to get the clip syncedo?

Ooh. Yeah. I like that. And I can ask it, did this file reach this host? And all the clinker is doing in the background, is calling this Hermes sync thing. That's a wrapper that just has the endpoints you want. Yes, predefined setup.

And so it's not, and the advantage there, it isn't creating the workflow every single time. So you're saving tokens, you're reducing the impact of hallucination, because it's fairly straightforward. You can do like a multi-stage, right? So it's like you have a skill describing this how to work with it and telling it how to use your wrapper,

and then the wrapper can embed all of the implicit knowledge of how to get those core tasks done. And the wrapper can query the destination machine, own syncing database. Nice. And see, okay, is it present? Is it valid? Is it synced?

Was it ignored? Was it not deleted? Whatever. So, like, you get beyond just proof of delivery, you can get proof of removal and all of that. And it's just using the wrapper every single time, which I could go down

on the command line and run myself if I wanted to. Totally. But the nice thing is, is when you do something like that in that wrapper, is you can use a free model, a cheaper model that doesn't have to be like a big, you know, frontier class model. It can be an open source, even eventually self-hosted model,

but it can execute these, queries for me and tell me if everything's synced. There's no debugging, there's no code writing, it's just column tools. Yeah. It's pretty great. So I was going to ask if you were going to try and come up with like a web interface replacement, but it sounds like for the most part, this is good enough.

Yeah, I still toyed with the idea of putting some sort of web file manager in front of the sync thing directly. Just for like upload download or something like that. Yeah, or this one off machine. Right. Or mobile. That could be another way to do on mobile. I don't know what sort of state of the art there. Like, I don't think it's

PHP file manager anymore. So if anybody has recommendations for that, I would definitely like them. Because I do think that could be useful, but I don't think it's as necessary as this. It's been pretty nice. So are you done? I mean, is there anything left that you need to build in this thing

besides the web interface? Are you happy with the deployment? Is it on all your machines? Yeah, everything's deployed. It's gotten really nice. You know, I think the trick was is deploying from one central machine and just having it reach out over SSH

and just blast it onto everything with their various esoteric setups. I think the missing piece, still like I mentioned, might be the R clone mount. Right. And just coming up with a better way to take advantage of that 15 terabytes

while I have it. Not sure. But I think that's the work still to be done. There's no errors. Everything seems to be sinking. And then I do have a couple, like this machine, I've got to clean up drop box on.

Same thing is just, it feels like this sleeper of a project that we don't talk about that we take for granted. I think it's just assumed. that we've all implemented it. Like, that's the feeling I get, is that it's been around forever.

It's been really stable for a long time. It's just kind of, it's like SSH. It's just there when you need it. But for anybody who hasn't tried it, do you have any suggestions or tips or pitfalls to avoid or anything like that? And if you have SIGT thing, do yourself a favor.

Seriously, listen to Chris's hot tip here. Download open code. use the free Ox model that's in there, use any of the free models, point it at your URL for your Sing Thing, and have it tell you what this API can do. Because this is where it really exceeds what Dropbox can do for you, and it becomes a complete system that you have observability over that doesn't have a cloud, doesn't have any big tech login.

And I think so many times we deploy things like Home Assistant, Plex, Jellyfin, and Sync Thing. and then a lot of us, and I'm definitely guilty of this for a long time, just ignore that there is this beautiful API that allows us to do everything we could do through the clanker or through the UI with our clanker or open code or whatever through this API. And so you can start writing little scripts or automations,

or check on the sync status of things. Think of the poor developers who implemented these APIs. They want you to call them. They do. Don't let a matrophy just there on the side. Take advantage of those APIs. is because I stopped thinking of sync thing as just this thing I self-host

and it moves my files and now a full system I can manage. It's a sync architecture. Yeah, it's a framework. Nice. That's what it is. That's what it is. It's way better than any of stuff

you could get with the commercial stuff. And it really is taking advantage of that API to its fullest. And you can craft it to fit in with whatever you need it to be. Well, as usual, we got a little feedback this week. Alex sends in a little note, hey, Chris, Wes, and Brent.

I'm Alex. I'm Alex. That's A-L-E-K-S for, you know, that's a much cooler way to do it. I'm Alex. Alex is a solo. I don't know. An X is pretty cool, dude.

Oh, that's true. X is pretty cool. K is different. I like it because it's unique. Yeah, yeah, yeah. I mean, it's pretty ready to get an X in your name. So Alex with a K says, I'm a solar developer working on Local Comet,

an open source, Windows First Local AI, built with Lama CPP, Tori, and Rust. Oh, you don't know, right, rust. It gives people a simple desktop way to run local models with more privacy and control without depending on a cloud service. I'm actively developing it and would be grateful if you'd consider mentioning this on this here podcast and he sends a little repository. Well, this is right up our alley.

Yeah, we got Python typescripts, felt, and yes, Rust. Neat, a private, local first Windows AI assistant. Yeah, and you don't see Windows getting a lot of love in this camera. That's what I was going to say. I'm not about to go make myself an EXE, but it's clearly out there. And the people, I mean, if we're going to use in these models to help us with our sync thing, I mean, as we already talked about on the show, do you want to be touched in Windows or you want your model to do it? I mean, honestly, why not, right? And then there's probably somebody out there listening to this. show that has WSL and has all this running and then they use Windows MCP and all of that. But if you're just getting started, something like Local Comet could actually probably help

you get WSL up and going. I mean, I don't know what these things can do on a Windows platform, but it is really nice to see it. And maybe it's an easy way, like if you have a Windows gaming machine that does have a GPU, then maybe it's an easy way to go experiment with some local models too. Yeah, I bet a lot of Windows, you know, that's a great point. Don't lock them away on the Windows world. You can still get to play with the fun tech.

They've invested in those GPUs after all, right? Right. So they might as well use them. Yeah. Local comment. So there you go. We'll put a link to that in the show notes. If you got a Windows box with a goopoo and you want to play around with it, you'll find it at Linux unplug.com slash 681. And now it is time for the boost. Oh, our first boost comes from the one, the only Greg the lawyer. He was in our mumble room for a little bit earlier today. He's often on the live streams and frequently a big supporter of the show. And he comes in with a baller boost of three. 33,333 Sotis. Hey, rich love star! And Greg says, inspired by the release of Oma Archie Atra or Quatro, I have been clanking my own NixOS Senate.

Wait, wait, wait, wait. I got to stop you there. Clanking as a verb. It's a verb. I'm not seen till night. I like it. I do like it. I've converted from KDE to Hyperland and Quickshell, nice, making my own desktop so much more fun. Oh.

I love seeing reports like this, Greg. We are entering a new era of customization on Linux, and Hyperland is just so forth. Speaking of Omaha Archie and Hyperland, an Omaha Archie Foundation of Types was launched with $8 million of backing by Jack Dorsey, the CEO of Dell, and many others. And one of their first expenses as a foundation,

if you can call that, I'm not sure, is their backing Vaxery, the Hyperland developer for the next three to five years. Oh, that's incredible. That's huge. Yeah, at least the next couple years, I mean, I don't know how much Vaxri is going to take as a paycheck, but that's massive, so he can focus on the Hyperland desktop,

which I think is going to be really good. A quick show show, I was looking like quite the interesting toolkit as well. So I put a link in the show notes to a great LWN write-up of that recently. Very good. Thank you, Greg. Appreciate hearing from you. Appreciate the support, too. Tomato Boosin with 22,200.22 Sats.

This old duck still got it. Catching up with the missed episodes from this summer. Loved the backwards episode, and how it broke your brains. Regarding virtual machines in the kernel, I'll just lean into my recent NetBSD trolling here and point out that kernel already has a Lua VM in it. It's honestly pretty fun. That is fun. That's all so fun.

A great show on entropy in the cold card mess. I tried looking into their firmware before moving my BTC to it and found the whole thing overly complicated and confusing, so I decided I didn't trust it. Nice to dodge. Despite having bought one, my coins are still on a sparrow-creative. wallet on an offline machine. Yep, turned out to be a better call.

Wish I'd shared my discomfort publicly, and I hope all the J.B. listeners got their coins out in time, especially in this new era of Clanker-assisted hacking. Those of us in free software who have the expertise to audit code should should really speak up if something smells funny. Yeah, there's something to think about there. That's a great point. Definitely worth checking your own stuff, too. And at least checking the stuff you're going to use, even if you don't go file a Clanker report, although if you find something serious, you probably should consider it.

Yeah. Thank you, sir. Appreciate that boost. Tert Ferguson also sent in a row of McDuck's 22,222 sets. Turg Ferguson! This old duck still got it. I enjoy the peek into the future of IT defense, but you did not answer a key question that I have. Do you think AI security tools are actually useful for regular Linux admins yet, or are they still too unreliable to trust? I think that's a depends, maybe, because you'd experimented with a meta model that did have loose. some of the reporting, which maybe would be like, okay, that's an example of when you can't fully trust.

Yeah, I think it's kind of, like a lot of things, you know, defense in depth. So you probably want to have, that's where you like a diversity of models or harnesses or processes so that you do have some way to have a, like, you know, grounding is very important here. And so if you have a way that you can actually do some fact checks or you have a council of several models each doing it with different unrelated contexts, things like that to try to be able to figure out when that's happening. I think turned the biggest barrier right now to getting actually good results out of these tools would be paying the cost for the higher end uncensored models that you've got to pay kind of per token for. So that's the gating factor right now, right? For them to do anything more advanced, really, I think. And that, yeah, that kind of pattern, too.

You can have a grunt model and then another model that sort of checks on things does its due diligence and has a little more intelligence, but doesn't have to do all of the original work. Nathan from Brissy Boussin, 4,102 cents. I like you. You're a hot ticket. Come on, lions. Hat trick time. Yeah, that probably works. You're getting a few people thinking about it that puts it in the good direction, I think. All right.

Thank you, Nathan. Appreciate you. Right? You want to take Monty there? Yeah, Monty sent in a row of ducks. As for the Frigot System Overview, I have my... I have been running Frigget for several years, and I haven't touched the config in a very long while.

Very interested in some of the newer features, including how to drain your own detections. Yeah, and Frigot has recently added more detection capabilities in the freer stuff. That's something I'll play around with, you know, and maybe we'll do an updated segment, and I'll dig into that. One of the things I have recently done

is I've expanded beyond person identification to animal identification, which again, keeping my up for coyotes and, yeah, the birds and stuff like that. And that's been interesting. So, all right, thank you, Monty. That's good feedback. Quick interruption.

I think you'll love this. You should go pull up our live chat. Or Matrix are going over there right now. Because check out what Greg, the lawyer, just did. While I was listening about Singthing, I clanked a Quickshell Sing Thing widget for my system. That's so great. I clanked it, he says. And look at this lovely little widget, right?

Shows devices connected. It's got some folders there, WebUI, re-scan, refresh stop buttons. QuickShell is so cool. And clearly it is quick. Yeah. And it's not going to break with a hyperland desktop update, which is nice, too. So you can build that. It's going to work for a while.

Well done, Greg. That's why he dropped out of the mumble room. You had to go work on that. Plank that up. I'll take Thomas. Thomas English comes in with 16,345 Satoshies. What are the best methods to get my agent to do my entire NixOS setup for me? because I am clueless.

I've only ever vibeslopped Linux servers together, but going to take the plunge and smash out this Windows box, NixOS for me. Any tips for a sailor that can't swim? I've never seen a line of code in my life. Agent First NixOS setup. What say you? I'm going to paste in your YOLO pie and close my eyes.

Oh my goodness. I'm going to paste it into YOLOPIE. It's like a prompt request vibe. Yeah, it sure is. I mean, you could, I mean, I don't know if you'd want to have it. You could point a clinker at hyper. I was going to say, and we've had some, the whole collection of listeners who have shared in past show notes. So you could perhaps the first prompt is go get it to harvest some links from that.

Yeah. And then from there, it could go build out a config for you. I think that's a good starting spot. I would, I'd start probably in a web chat with one of the free models to just sort of build out a plan, talk about, you know, kind of ramble about your intentions, build out a plan, link it to hyper vibe. I think there's a re-spin of a door that looks really good. You can even point it at OMA Archie and see if there's some design tips you want to pull from there and application ideas because it can look at all that stuff and sort of build you a master list. And then you could have it from there produce a master prompt that you then put into the final clanker on your machine that does the work. And you probably get pretty good results with that.

Yeah. And, you know, the main kind of things too is once you get like the installing NixOS part can be just pretty much like installing a normal any old Linux distribution. It's just getting the once you have it running. and you have a way for a clinker to control it, then you're off to the races. You are. You are indeed. I think I would recommend a very different approach, actually.

I think you should definitely use a clinker to help you, but point it at some of the documentation, and then just have it almost like interview you about what you want this system to do, what kind of things you want in it, and have it explain what it's doing. Because I think the long-term benefit of understanding what's happening under the hood for NixOS is really good, even though you don't have to worry about the implementation details, like all the syntax and stuff like that. It's really good to understand what's possible.

And I think starting with some of the documentation, or at least projects that are well-regarded is a good way to go. Two more tips. When it's done or when you're at a stopping point, have a crate like an agent's MD that explains your system so that the next fresh LM context, when it starts up, it kind of knows your system. So a basic weight layout of where you're configs at where your preferences are, the desktop choices, the things you prefer the clanker do, you put all that into like an agents MD or some sort of description file about your system. And then when you start a new session, you have it pointed that.

So that way it's not rediscovering your Nix config where it's at every single time and all of that. It knows the rebuild command. It knows all the, yeah. Yeah. So I think that's probably maybe the most important tip is just have it build one of those. And like, yeah, the other thing that would maybe be good for you to consider is second tip, you could ask it to document in a human reviewable format, decisions that are being made, and just kind of, it could append and update a markdown file or a text file for you that you could review at the end. You could have another LLM review. Something, you could have it also produce two documents if you want. You could have it produce a technical document for an LLM

with no context, a fresh LLM session with no context, a technical document that explains everything. A fresh LLM with no context needs to get started and work from the beginning to modify, change, upgrade, anything to my system. And then a second document that's for a human audience that needs to understand the decisions that are made and the design choices that were made. And it will just continue to update both of those documents for you as it works in that session. As long as you maintain the consistency and it sees that every session. So I remember that. Yeah, a devlog is one word you can use for some of those concepts. Oh, okay. I have two more quick tips that I think are good for someone starting out. If you are comfortable enough to install things like in open code,

Having an MCP is really helpful here. So there's MCP dash NixOS, which is a really great program just to help, like, query what's even available in NixOS. And it saves a lot of time and mistakes and just, like, querying the internet. So I would recommend that. And then if you are up for it, writing yourself just a little skill for NXOS with, like, common patterns that you particularly like once you get down the road is really, really helpful, at least has been for me. Yeah, and if it's not obvious what they might be, they will become obvious to you as you use it. They will become obvious. If you keep having to tell it the same thing.

Yeah. There you go. If you tell it three times, then you should put it in a skill. Moonenite comes in with two. Thank you very much for the boost. Good luck. Let us know how it goes. Moonenite comes in with a boost, 2,500 sats.

B-O-O-O-S-T. And he says to all the new Fiaters, you now have boost power. You got boost power. Nice. I like that. That's true. Thank you, Moona-N-N-A-N-N-E. Thanks for the sound clip, too.

A pabby boost in with 7,108 sets. I don't understand what the heck is going on here. Time Travel Boost, Episode 671. Love hearing from the time travelers. Thanks for the Ulanzi Smart Pixel Watch tip. I've been looking for one to integrate into home assistance since I'm not getting around to building the Autrix kit I've had for years. Also, on 1.5 speed, Forskine sure sounded a lot like Forskine, which had me quite confused for the segment. Had to look at the show notes.

great name. It's the Ford Scanner. With the 1.6x. I think it's what could else, yeah, that's right. What else could they do? Ford would come after them if they put the D in there. They can't put the D in there, so they just had to go with four scan. Forkscan?

No, that doesn't help. Yeah, it doesn't work either. Yeah, it doesn't work. I think it though. Appreciate it, Pabby. Well, Zach had Hack sent in 6,000 sets. Oh my God, this drawer is filled with broolopes. Love the episode on Hacking Yourself.

I did dislike the episode where everything was in Reverf, verse, though. Need to still watch your clanker therapy video. Yes, that's linked at every episode recently, right at the top of the links. For now. Almost done linking that, too. Something you could take a look at is Rockbox. I'm running it on a SureFans F20 for podcast and music. It has better audio than the phone and increased privacy, too. That's true. I haven't heard of Rockbox in 100 years. That's really cool. This sounds vaguely familiar. Is that a piece of soft. where you can throw on some MP3 devices. I think I might have tried to play with this a long time ago.

Long long time. It's still around. They go. Zach, thanks for the memories. Appreciate that. Appreciate that. Thank you very much. Sir, I'maublet's here with a row of ducks. It's pretty fitting. Thanks for breaking our brains with the reverse episode. I was about to

blame my podcast out. But how dare you skip the hello friends and welcome back intro? Anyways, here's some value for the show. Oh, thanks. Hey, if it got a boost, maybe it was worth it after all. But I think we're done doing that. I think we're done doing that. Our pal Magnolia ma'am comes in with a double row at ducks.

Hey-oh. Affleck! Two of the same message, but it says, just going off of the title here on 680 GoHack Yourself, but I just got started with syncing myself after a Git workflow of Mind failed. Can't wait to see what the show's about. Ah, that must have been to the pending. Yeah, the live one.

Live boost. Well done. Well done. Boosting one before we even started a show. Mm-hmm, we're doing. Yeah, let us know how it goes with your sync thing set up. That should be very interesting. Blackhost comes in with a Fiat boost of $1,000?

What? What? What? When did that get in there? I don't know. I'm just seeing it. When did that come in? 10.21 a.m. this morning.

While we were already on the air. That is crazy. Wow, Black host. Wow, that's really, that's, I got to pick up my jaw here. Yeah, thank you. No ads placeholder. That's so sweet.

That is really amazing. Wow, wow, wow. Really do appreciate that and it definitely will make a difference. Thank you. David from Burlington comes in with $10 fiat. Hey, you know what? We have a, we have a boost sound for that now, right? Should we roll out?

Let's try. What do you think? You're ready? See if it feels good. You got boost power. That's not bad. It's not bad. Hey, folks, it occurred to me that while you all have extensive tech backgrounds,

I don't know if I've ever heard where it all came from. Oh, boy. Can you guys give a quick summary of your past work history? Oh. Stay a while and listen. Oh, boy. It's basically a virus that infected all of us. Could we do like some special on that?

Maybe. Like in a members we put in the show, maybe? I don't know. I don't know. We'll have to think about that. Thank you for asking, though. Yeah. And for the support.

The reason why we don't have time to. do it now is because we're old. And so there's a lot to it. It starts... We only remember so much. I remember before Linux. I was alive. Yeah. So yeah, that's a great that could be a great topic. We haven't really recapped that in a long time. I suppose some people have some questions on that. Thank you, David.

From Burlington? My Burlington? Yeah, which Burlington? This one in Ontario. Maybe a zip code boost in the... Yeah. I'm not getting beers with them then. But all right, we have some member boost that we'd like to get to. Our buddy Anonymous is back. Do you guys really use Firefox? You talk about a record number of tabs.

Have we mentioned that? I don't know that. But I don't believe it. I've used it on and off for years, and I keep having to bail back to Chrome trying again, but after a day two, it just hangs on me. I have to kill all Firefox. I got 90 gigs of RAM, a 22-core Zon.

50 tabs really shouldn't hang. Wow. I never have it hang. And it's probably open the entire time my laptop is online. I have in the past had issues, but it's been years. Yeah, interesting. For the question of, if we, you know, do we actually use Firefox? I'd say it's my permanent first choice. I mostly only use Chromium when something goes wrong and the website's not built for Firefox or for the show, because I need multiple browsers. And I would say I probably have the most tabs and windows open out of all of us, probably out of you guys combined. So there's that. But I only really used to have issues on like low end hardware. where I didn't have enough RAM

but never experienced the slowness or anything like that unless I was filling up the RAMs. So I'm not sure what your issue is. Hopefully you can figure it out. I love this. On my home desktop, I do about Garuda and Arch for Daily Driver in Windows 11

for occasional work applications that, well, I couldn't board to whine. My 15-year-old plays video games, Minecraft Steam on Garuda. Yesterday, my 6-year-old came up to me while I was working and said, Dad, can you boot into Linux? Windows is boring.

That's amazing Nice Aww He says with all of the Success with the Reverse episode How about a randomness episode You could do the segments In random order

Good luck Yeah that won't be harder at all I don't know You really want to melt our brains It'd be fun too If we just literally like rolled a dice or something And we didn't even know It's totally random

There's something to do that You know Or each MP3 that gets generated Is a unique Different order Boy if we could pull that off Why do this to ourselves Anton member boosts in.

For the Web Boost Anonymous username issue, the quickest fix might be a small confirmation pop-up when the username field is left blank or default, and the user clicks the member boost or checkout button. The pop-up could confirm that they meant to send it anonymously. I don't love a pop-up, but that is not a bad idea. Not a bad way of some of it. Thank you for thinking there, Anton.

Well, Jimmy's here with a free members boost, which you could do at boost. jupiter broadcasting.com, if you're so interested. How do you reconcile the privacy considerations of AI? It can be amazingly helpful, of course, but they get a lot of sensitive personal info. That's what's holding me back. A local model would fix that, but of course the V-RAM required for good models is quite prohibitive.

How can you be running graphene, but pumping configs into the AI machine? I guess it, I mean, do you consider Nix configs to be confidential information? I generally don't. Especially for something I might consider putting on GitHub. You guys don't keep your bank account information in your Nexus config? None anymore.

No, now I don't know. Declarative banking. So, like, if I were putting in personal information, I suppose that would be one thing. Yeah, you know, health stuff, addresses account number, you know. Right. So there can be ways you want to have concerns about that. And some information will leak, right,

if you ask it questions about how do I set up X thing or I'm trying to solve this problem, then, but the question is what risks, how does that fit into your model of? Right. In my model, while that is a risk, it's actually less of a risk than like, you know, data brokers that are monitoring all of my transactions or, you know, government surveillance, or Google surveillance or, you know, like those are actually more prevalent in my day-to-day life that I already live with. I actually wouldn't mind if they did train on my name. Nix configs, and they got better at it. So in a way, I kind of want them to, I don't think the models I'm using are using my

data for training, but if they wanted to become better at NICs, I really would not have any problem with that. I think of it, too, as sort of like a controller agency thing. Like, for me, I mean, yes, the Graphene OS sort of privacy stuff is great, but I really like that it makes me actually feel like I control the Android phone. And that's where it's like with the AI, I might make mistakes or whatever, or be lazy, but it's ultimately my choice of like, do I engage with it? What do I put into that prompt before I hit enter?

And stock Android didn't really feel like that. I have, yeah, various categories, like you say, Chris. NixOS config's totally good to send those out, I think. They're not private, really, at all. But I do hesitate on some topics. Like there's a couple topics that I have that, like, accounting automations and stuff that I have sitting in the queue for when, I feel like there's some great option for, like, privacy, local AI stuff.

So I can't wait. But the other thing I would say is there are a couple things you could do about that. Like, I don't trust myself completely to keep myself safe. So I put a couple layers. Like there's an MCP that you can add. I'm trying to see it here that I've added to my open code just to, like, pull out secrets and bank out and stuff. of credit card number. So it just goes through and sees if you've accidentally put anything like that in there.

And so there's a couple things you can do just for peace of mind maybe. Before it hits the model. Yeah. Wes has talked about also like at the injector level, if you have an injector sitting between your client and the cloud service, you can have something that looks like it's the, it looks like the API to the client, but it's actually sitting between the cloud model and the client and it's looking for sensitive information. Yeah, we've talked about a few of these.

Key G-Proxies, another one. There's a few options out there now. And you could also consider, this is a lot more technical setup, but if you do entertain something like renting a cloud GPU and then running your own self-hosted models, that has a lot of nice privacy implications too. And maybe if you only need it for certain batch work

or you want to build a certain thing and you're not going on day-to-day stuff, you're not running an agent or something like that, you're just using it for that type of work. A rented GPU with a self-hosted model, well, then you've got full control. You've got out of it. So that's a good question.

Thank you, Jimmy. Abby comes in with a free time travel member boost. So great, this is now possible, Pabby writes. Thank you, Pabby, and I love the time travel boost. But you got to say what episode you're at when you time travel. Otherwise, we don't know from what time you're at. Digital Jolt comes in with a member boost. All right.

I'm a bit late, but just wanted to say, I really enjoyed the Reverso episode. Okay. It took me a minute to remember that you'd mentioned it a few weeks ago, but I smiled every time I heard one of the jingles appear in the wrong place. Good, all right. I'm glad. Love that.

Well, Paul member boosted in. Where are you on the ad block is privacy topic? I don't like giving money to corporate overlords. This is the main thing holding me back from YouTube Premium right now, for instance. I prefer to support a few selected creators. My willingness to spend money is limited, leaving many creators out. Yeah, that's a tricky one because the ad blocking, at least in the browser, is it's not just about ads.

It's also about performance and readability, obnoxiousness with the pop-ups. So there's other things now that are also a factor there. So it's not just like a statement on trying to get the content for free. And I do tend to agree, I think. I kind of align with you, Paul. Like I think Foronix is a great membership. LWN in our space is a great membership.

We try to make the Jupiter signal and unplugged core contribution, you know, really as much high signal as we can, worth it membership. But I have over the years decided, because I have a family that watches YouTube, I did subscribe for YouTube premium to avoid that. And the YouTube experience is notably

better. It's for sure. When you sit down on an account that doesn't have it, I cannot believe how much ads they're slamming in there now. I just don't want my kids seeing all that. As for the sort of, is it piracy or not? I think it's one of those, like, you know, it's sort of a spectrum. The world exists where both people

are dependent on ads and one, local control of what your computer does is important and you know ad blockers are a known technology that exists so i think it is it is true that choosing that perhaps has a moral dimension but the idea that you know you only need to make one choice for every single creator that doesn't seem right to me yeah it is a nuanced problem i think for me this is a really good one i think about this all the time i've kind landed on ad blocking everything because I feel like very naively actually I feel like we could find a better way to fund things instead of using ads so if they're not making yeah if they're not making

you know money on ads maybe they can find a better way but I'm sure it would just be more draconian the problem is it's always it's always kind of eventually a race to the bottom you always have to add more ads you know it's just the way it goes yeah I would say one option is you you know let's say you're using ad block plus or something like that, you could disable it on websites that you want to get the revenue. Like I do that on Photopi. Yeah. Yeah.

But, you know, they ask, and I think it's worth it. PhotoPee, they deserve my ad revenue. That's another one I should probably. It does give you a chance to find out, too, like, how far did they go? How are these, like, horrible ads, or is it a reasonable sort of ad placement that isn't actually that bad in practice? Good question. The last thing I would add is, like, the internet is completely unusable without it.

So there's that, right? Yeah, it's bad. All right. I'll wrap us up. Senior Smile comes in with a free member boost. Papa John was correct. I forgot to put my username last time. I said a USDA boost referencing my previous anonymous boost about K-8s.

But you read them in reverse order. Oh, sorry. We don't always get that right. Yeah, sorry, senior smile. And then Bearded Zero wraps us up. I enjoyed the backwards episode, and I got to say, I think doing the boost and feedback may have been a good option. It lets it bring the feedback into the show and kind of feels like a bookend for the last episode.

Just a thought. I also felt it was kind of nice to warm up with the feedback in the boost. And I think if I were going to always do the show in the order that felt the best, that might be the order I'd do it in. But if you think about it, you see that, you know, it's like if you're new to the show and it takes us 30 minutes or whatever to get to the actual content. Not everyone loves that. Not a good impression. But I liked it too, Bearded Zero. I liked it too.

Thank you everybody and everybody who boosted below the 2000 set cutoff and all of that. We really appreciate you. We really had a banger. So we had $124,000 sat sent in, but then when you bring in that Fiat boost, I mean, we should take a moment again. I mean, that was really something. That was really something. A thousand dollars came into the show today from one single contributor, and we needed it. It has been tight behind the scenes.

It was... That works out to be 1.29 million Satoshis. Wow. Look at you doing the math. No, it wasn't. It was hybrid sarcasm. Thank you, hybrid sarcasm. I mean, that's got to put Blackhost on the leaderboard for the year.

Oh, there you go, right there. Yeah, maybe we'll have to do like a boosties preview sometime soon. Yeah. So when you bring it all together and you take the Fiat boosts and you convert them to the current Satoshi value and you just smash it all to one line item, we had 16 of your stream stats. That's 27,485 Satoshes. When you combine it all together, it gives us, I mean, a very impressive grand total of 1,458,305 Satoshes. That's crazy. What?

If you'd like to support the show, you can become a member, put it on autopilot. You get the bootleg or the ad-free version of the show, and you get a free boost every single week. Or support us one-off. Like your particular show, want to get your thought on here? Just support us. Go to boost.jubterbroadcasting.com. You can set a boost with Fiat or Sats. Crazy.

Wow. All right. And thank you everyone who supported the show. We really appreciate it all. With that, it's time for some screaming picks. All right, buckle up, boy. because this week we're talking about SSH clipboard, and this could change your life.

A peer-to-peer clipboard sync server over SSH. Copy anything on one machine, paste on another. A picture, a block of code, it doesn't matter. You copy here, you paste there. Rich text content doesn't matter. It keeps the native formats intact. Native formats, not just text preserving images, rich content, things like screenshots,

something you might want to put into a ticket maybe. You want to take it from one machine and put it on the other. Maybe you've got a clanker that's working on something and doesn't have to save it to the file system anymore. It can actually just put it right in your freaking clipboard. Or if you put it on multiple machines, the clanker could put it on the machines clipboard you tell it to. You're saying I should be using this instead of netcading things around. This is actually an option to reduce your net cat dependency. It looks pretty new, huh?

It is very new. I should mention. From Justin Schroeder and Andrew Boyer, the contributor so far. Yeah, yeah, yeah. It is very new. It is very new. But I think there's been several attempts at various tools like these over the years.

I think when I was first getting into Linux, remember trying to roll something myself. It's nice to see something that looks like a structured, well-thought-out approach. Yeah, I went through it. I dug through it, and I was very impressed. It's early days, but very, very impressed with what they've done so far.

And, of course, if you've got SSH key set up, just you reuses those, so it handles all of that for you. It's cross-platforms. You can move stuff between Mac and Linux. use the reuse the copy-paste mechanism there. It won't break your other clipboard managers. That's good.

Right? It's using Rust and Ratatoui, which is just turning out to be such a great library. Yes. I think they'll be packaged in multiple ways right now. Like they're kind of the recommended path as NPM. I did look into it. You could probably,

probably package this. It's pretty straightforward. So it is very doable. And it is MIT licensed. MIT licensed. Now that has been Black House pick of the week. Yeah. Brought to you by Blackhost.

I had another one. Oh, a sneaky pick. You didn't even put it in the dog. I didn't put it in the dock, so I don't remember what it was, though. So you're saying you have a big for next week. It's in his clipboard. It's been a long time since we've done. Yeah, you should have said it up and probably isn't my clipboard.

Oh, my God, that's hilarious. It's been a while since we've had a single pick picks. I know. Well, that is a banger. That is a banger of a pick. I just can't remember what my other. other pick was. I can't remember what it was. It's probably sing-thing.

Yeah, I'll have air. No, I was pretty proud of it. It's funny that I can't remember it now. That's what I get for editing editing a text editor first. So we'll sneak in the next episode. And if you've got a banger pick, you can always let us know, too. We love it when we feature an audience pick. We'll put in the feedback, or we might even put it in the main pick segment. So you can boost out jubiterbredcasting.com that or go to

Linuxunplug.com slash contact for that. All right, well, that wraps us up. Links to everything we talked about today over at the website, Linuxunplug.com slash 681, or jupiterbroadcasting.com for the links and all the shows. And if you want to just, you know, chew away at some extra stuff, you want to know what we talked about, anything like that. Tell them, Wes. Tell them about these fancy metadata that we got. Yeah, well, the first level is cloud chapters.

What? Yeah, that's right, a JSON file with all the information about where our segments are and what they are. And, you know what? If you look in that, you get custom Drew-made chapters. They're great. He's on all that editor-clemerness. They are great. But if you need more details, check out the VTT file, or if you must, the SRT file.

We have subtitles, transcripts, right, available in the RSS feed itself. And if you want to see our ugly mugs, there might even be an MP4. There could be with screencast stuff and all that of what we're talking about. I'll just say again, thank you to Blackhost. You know, it strikes me. I was thinking about it this week, like all four of us, the three of us here on the show and also Drew, you know, we really have shaped our lives around this. the show. It's that level of dedication. It really drives our life. It's the defining clock and

the heartbeat of our day-to-day life. And so when you see support by folks, like we saw this week, it really does validate putting that kind of effort and energy and building a life around the show to make it possible so that we were here every single week going on like, what, 13 or something years? Crazy. Not missing a beat. So we really do appreciate the support. Thank you, everybody. Like I said, links at the website. And of course, we'd love it if you join us live, making a live vibe on a Sunday. See you next week. Same bad time. Same bad station.

Get past the Sunday blues by making it a Tuesday on a Sunday over at JBL Live.tv in your local time at jupiterbroadcasting.com slash calendar or stream live in your podcasting 2.0 app of choice. And of course, the Mumble Room's going. We've got details on the website. The Mumble Room gets a low latency, high, quality, open stream. It's pretty nice. Shout out to our members for making the show possible and keeping us going to everybody who supports us at boost.juporbroadcasting.com. But that's it for this week's episode. Thank you so much for joining us, and we'll see you back here next Tuesday, as in Sunday.