Just to stir thought, I note the TiddlyWiki[1] community (wiki as a self-modifying single html file; 20+ years old) has of course been exploring AI tooling... though not necessarily as an agentic environment. There's a markdown plugin, and others to make the file executable, or into a self-serving web app. Git is more problematic. So hypothetically, one could have a single-file agentic wiki wandering around and self-editing.
Niri introduced scroll-based window management to me and it instantly clicked. I'm very happy to see a full-on Niri per-workspace emulation mode in OmniWM[1] for the Mac, recently and thankfully made compatible with Sequoia. It immediately became my main window manager.
We're off grid and have 7kw of panels, and 40kwh of 48v lithium batteries, with a generator for backup, which is rarely used since we are frugal with electricity and switch everything off when not in use.
I set it all up myself, and while it is not trivial, it's not difficult either.
Learning to put connectors on properly, size cables and put lugs on properly, learn about earthing and breakers...just one bit at a time.
I'm about to set up another system on the roof of an outbuilding to supply power for a water pump and irrigation where we grow food. This will be much easier and simpler since it will have only one 48V lithium battery, but I'll still use Victron stuff and connect it to a Cerbo so it can be monitored.
If I sold this place and bought somewhere on the grid, the first thing I'd do is cut the cord and set up my own system again.
There are infinite combinations of CPU/GPU capable of running LLMs locally. What most people do is buy the system they can afford and roughly meets their goals and then ball-park VRAM usage by looking at the model size and quantization.
If you have a huggingface account, you can set your system configuration and then you get little icons next to each quant in the sidebar. (Green: will likely fit, Yellow: Tight fit, Red: will not fit)
Further, t/s depends greatly on a lot of different factors, the best you might get is a guess based on context size.
One thing about running local LLMs right now, is that there are tradeoffs literally everywhere and you have to choose what to optimize for down to the individual task.
Support brands with values and local manufacturing. For example: American Giant, Origin, Crye Precision, Randolph Engineering, American Optical, and many more.
Early benchmarks show tremendous improvement over Kimi K2 Thinking, which didn't perform well on our benchmarks (and we do use best available quantization).
Kimi K2.6 is currently the top open weights model in one-shot coding reasoning, a little better than GLM 5.1, and still a strong contender against SOTA models from ~3 months ago (comparable to Gemini 3.1 Pro Preview).
Agentic tests are still running, check back tomorrow. Open weights models typically struggle with longer contexts in agentic workflows, but GLM 5.1 still handled them very well, so I'm curious how Kimi ends up. Both the old Kimi and the new model are on the slower side, so that's a consideration that makes them probably less usable for agentic coding work, regardless. The old Kimi K2 model was severely benchmaxxed, and was only really interesting in the context of generating more variation and temperature, not for solving hard problems. The new one is a much stronger generalist.
Overall, the field of open weights models is looking fantastic. A new near-frontier release every week, it seems.
Hi! I actually have, and have been using as my main device, an MNT Pocket Reform, and at one point was using an MNT Reform.
MNT's devices are honestly kinda incredible. I can't recommend them for everyone yet, though that will change soon. Both of them are a kind of "laptop of theseus"; you can open and change and repair them, and honestly I have. Both device's guts are dramatically different than where they started, but changes happened piecemeal.
The Pocket Reform is an incredibly cute device. I can't pull it out anywhere without people fawning over it. Not even just hackers! It's an open hardware cyberdeck you can use as your main device. What's not to love?
The MNT Reform Next will be closer to what many people want out of a laptop. It'll still be chonkier than a normal laptop. But again, these things are incredibly upgradeable and hackable.
Now for the caveats: for most people, I would wait until the MNT Quasar module comes out. The reason being is that while the current "best" module, the RK3588, is honestly pretty good with the 32gb version, it lacks one critical thing for most people and one other critical thing for me in particular. The first thing it lacks is support for suspend. Honestly, it does make working with a tiny computer like this a bit less appealing than the Pocket Reform's form factor could be, since what you really want to do is just be putting it to sleep and taking it out everywhere. The other thing is that Blender doesn't really run on the rk3588 either. You can kind of get a patched version working based on Lucie's patches, and I did, but it doesn't support the Eevee renderer, which is a must-have for me personally.
But the MNT Quasar board will be apparently fixing both of those above issues, and yes, at that point this will be a device that I can recommend generally. And I'll also note that I got the very first MNT Reform when it came out, and holy moly the state of the hardware now vs when it originally launched half a decade ago... it's hugely far between, but the amazing thing is that to get it up to the current state, I didn't need to throw things away, I could just open and tinker with things bit by bit.
In many ways, the MNT Pocket Reform reminds me of the book the main character has in the solarpunk book A Psalm for the Wild Built; a computer that is issued to you at the age of 16 and that which you carry with you for life. You can upgrade and repair it easily, but you don't need to throw it away.
So yeah, it's not for everyone. But if the idea of supporting repairable, upgradeable open hardware made by a lovely bunch of queers in Berlin sounds great? That you can hack on, that has a neat little community, that will be a conversation point amongst fellow hackers for its quirkiness? It's appealing to some, but not all.
I ran across their trackpad [1] last night; it uses the same Azoteq TPS65 module as the keyboard I just bought. Unfortunately the module's discontinued. [2, 3] I guess people have been going through stockpiles of these given that the last manufacturing run was 2 years ago, but I note the MNT Reform Trackpad is listed as out of stock...
I'm wondering what they'll do to replace it. You can still buy the IQS550 chip it's based on and apparently make a very similar PCB. [4] The black frosted glass seems like more of a challenge for most DIY keyboard makers, maybe not for MNT. The replacement project I saw recommended "2mm thick matte acrylic".
[edit: or maybe the TPS65 manufacturing restarted? mouser apparently has it in stock. [5] although that's the A unit without the surface.]
The LeRobot / SO-ARM101 uses $9 motors [1] has 5 degrees of freedom and is available with all parts for $300 [2] fully assembled. They don't quote a payload, reach, speed or repeatability. Their target market is people who want to play around with things like imitation learning outside of simulation at the lowest possible price, and who don't care about payload, reach, speed or repeatability.
The reBot Arm B601 uses $150 motors [3] has 6 degrees of freedom and a kit with all parts is $1200 [4] not assembled. They claim a 1.5kg payload, 650mm reach, and 0.2mm repeatability - numbers that are good for the price, if true (take that 0.2mm with a pinch of salt). It has no brakes, so don't lift anything you're not willing to drop. Obviously it doesn't compare to a $40,000 industrial arm, but for the price you wouldn't expect it to.
(Q|O)SFP are basically just raw high speed serial interfaces to whatever - you see this a lot in FPGAs, you can use the QSFP interfaces for anything high speed - PCIe, SATA, HDMI…
A very clever immutable Linux distro, and is the basis for the excellent PiCorePlayer, a favourite of mine to run Squeezebox clients (and/or Lyrion music server) on any Pi
Stop using MacOS spaces. Never full screen anything. Throw everything around with hotkeys using OSS rectangle. Use shortcat to automatically bring your cursor to anything on your screen and use enter to click and type.
It's surely not a great book and if you are someone who reads a book every few months i wouldn't recommend it. It's very weird and different and fun, though. I suggest it for people who read a lot of sci-fi and are looking for something that doesn't feel the same as 10 other books they've already read.
Visidata is a fast popular data tool in python https://www.visidata.org/ if you just want data entry.
If I may be permitted a small plug ...
Oleo is a GNU spreadsheet going back absolutely donkey years. I cloned the repo here: https://github.com/blippy/oleo and made minimalist fixes to get the thing to compile.
I wouldn't say that my project is particularly good, but it does have a number of features that I find useful. The next release will include Tcl bindings, so you can customise some of the workings of the app. There's also a module available where you can go to town programmatically.
It opens up a world of possibilities, including stuff like pivot tables.
package require oleo
load-oleo mysheet.oleo
set ws [toMat]
set subset [subRows $ws [mand [mstreq 2 foo] [mstreq 3 bar]]
It has what I think is a nice little query language, too. In the last line I select the rows from the spreadsheet sheet column 2 is equal to foo and column 3 is equal to bar.
I'm kinda dog-fooding it at the moment to put in place features that I need. It doesn't have the full equivalence of SQL-equivalent though.
Never do anything on faith or as a handshake deal. Always ensure you get paid (hint: escrow is kryptonite for weasels). Trust everyone, just not the devil inside them.
The iNaturalist API is an absolute gem. It doesn't require authentication for read-only operations and it has open CORS headers which means it's amazing for demos and tutorials.
I work the email security company xorlab[0], where my colleagues and I did a thorough analysis of real subscription/email bombing waves that we saw at our customers[1].
Here are some interesting additional information from the attacks we analyzed:
* Email bombing as a service is a thing, where you can buy 10,000 credits for $10 and easily bomb target inboxes with over 2000 emails per hour.
* Most all email bombing attacks starts in the morning, between 8-10.
> My solution was to kexec into a new kernel+initramfs which has a DHCP client and cURL in it - that effectively stops any filesystem access while the image is being written over the disk, then to just reboot.
* Set up a private space which will be used for google play services required apps (bank stuff, etc). Install google play and google play services in the private space. Do not install google play services on your main profile. Set the private space to lock after 5 mins of inactivity. Set up google play on a brand new google account. You'll need to provide a phone number during setup. I used my normal phone number, others who are more concerned about deanonymization could use rental phone numbers or other things. Install any apps into the private space.
* Try to install apps on your main profile, ideally open source, privacy respecting stuff. Some recent apps I've found that work great and replace google infested stuff - AntennaPod for podcasts, OrganicMaps for OSM maps, Obsidian for notetaking (google keep), KOReader for ebooks, Molly/Signal for messaging. Vanadium as the default browser works well, except it doesn't have adblock plus for youtube (it does some other ad blocking though and works fine).
Things I still don't have a great solution for:
* Android auto - I don't think it works from a private space due to auto locking. Still figuring this out
* Spotify - since it also needs to run in the background and I haven't found a better music replacement.
Overall graphene has been a far better experience and I like it much more, and feel more in control of my hardware.
Here's a little script to download all the publicly available scans (135) as GLBs and stick the metadata in a JSON. The scans are all CC0 (public domain)
[1] https://tiddlywiki.com/