you need to have developer doc in src, is the only way to have the full picture in the future. The main point is to keep it in sync, the agent harness must enforce that
> A year or two ago my iPhone started showing ads for Kentucky Friend Chicken on the home screen.
Please, elaborate. Because this sounds like a very thin slice of the whole story. Where in the Home Screen? How did it show up? Notification? Live activity? Automatically installed app? SMS?
iOS, regrettably, features ads in some places, but I have never seen (or heard about) any of it showing up on the Home Screen by itself and I cannot even fathom HOW that would play out since the only for information to show up in the Home Screen is through widgets, and those have to be put there, or they are shown “intelligently” (notice the quotes there) on smart stacks IF you have the “suggest widgets” thingie enabled—but then again this would mean you have the KFC app installed on your phone and you use it rather frequently so I don’t think it would classify as an ad in any valid way.
this fixes a real annoyance on older workstation boards where above 4g decoding is a requirement but the option isn't even in the bios menu. saw a build recently on an x79 board where it had to be exposed with a hand patched bios before anything like this or a modern gpu would work at all.
Because the same module runs in three places: inside the Electron IDE, in the shipped app, and in the test suite under jsdom, with no native build per platform. The other reason is the boundary itself: wasm exports can't re-enter, which forced the design where every side effect is yielded to the host and the VM is never on the stack while a dialog is up. That's what makes MESSAGEBOX not freeze the window. Perf isn't where these apps hurt, they're I/O and UI bound, and VFP itself was a p-code interpreter. The crate is plain Rust, the CLI runner is native, so a native build is a cargo flag away if it ever matters
>Instead of discrete, fully independent cores, AMD grouped processing units into "Modules" where each module contained two integer execution units, but they had to share critical resources like one FPU, the instruction fetch/decode pipeline, and the L2 cache so when both "cores" inside a module were heavily taxed especially with math or physics-heavy calculations (like in videogames), they choked fighting over shared hardware.
They did just fine in parallel workloads, so I think this is not accurate. The design scaled just fine. The problem was that each core was weak.
Thanks, this is the most useful comment here. FoxDev reads the DBC the same way VFP does, so today it inherits the hole exactly. The runtime is actually the one place it can be fixed. I would put in a hash of the stored procedure text into the built executable and refuse to run a container whose procs don't match. Adding it to the list. Would you mind if I quoted your comment in the issue?
Pricing is largely based on how many users we think we can fit per our baseline modset, and how much the cost of that modset is determines where we can work back from in price. From there, we would scale on a per modset basis.
It may not turn out to be a perfect science, but in the name of shipping something and getting feedback, it's out there
It is very new (launched a couple days ago), so still trying to figure out how to get it out there. I'm a software nerd not a marketing genius unfortunately.
As for how it works, its a pretty standard inference provider, where we offer a chatbot and an API, at unlimited usage for a fixed cost. Since its very new, we're still wondering if there's something we're missing as for why fixed cost AI is not the norm.
Hoping it is similar to the music industry, where you once had to pay $1.99 to download a song, and now you can stream all you want for $12.99. We want to begin that new trend.
Funny enough the customer is one of my dads friends that has been running the same shop for 20 years, wanted bigger tables and keep milking that for the foreseable future. I wanted something simple, no jit stuff, no gc complexity. Just a stack based interpreter. Yes it was LLM assisted like most stuff nowadays.
I see what you’re getting at, but let’s not pretend like “load-bearing” is something that AIs made up. It was a turn of phrase long before AIs became mainstream and people are still allowed to use it.
I have never understood the per token pricing model, so I built WavexAI to see what I was missing. Maybe I'll find something at scale, but turns out you can provide fixed cost inference without the fear of running out of tokens.
MicroVMs make the most sense to me, as they are the simplest and cleanest abstraction. I personally use VMs with a mounted github repo and found it to be the simplest mental model. I block all network access except for the domains that the agent needs to talk to, and I remove sudo. Since its all scripted, the VM is easy to drop and recreate and I know exactly whats installed on it.
Drop is interesting because it is thinking about the approach differently - basically an extension of the user's existing userspace. This prevents files from being inadvertently deleted by the agent, but it doesn't prevent the agent from uploading files that it shouldn't be.
It will be near zero when we get frontier like performance in models running on laptops. Basically the cost of electricity + one-time cost of hardware.
Astrophyzix Observatory. (2026). Astrophyzix Risk Index - A Heuristic Scoring System for Near-Earth Object Close Approach Visualisation (Version 2.0). Zenodo. https://doi.org/10.5281/zenodo.22240223
Astrophyzix Observatory. (2026). Astrophyzix Digital Observatory Daily Close Approach Report: Methods, Provenance and Limitations Documentation (Version 1). Astrophyzix Digital Observatory. https://doi.org/10.5281/zenodo.22868834
I love the tools that exist these days that help me quickly capture thoughts and ideas from the things that I see online. I hate the fact that if I stop paying for those tools, I lose the easy access to my library and the ability to peruse my library in a convenient manner.
To combat this, I did create my own app that I call Tenoa. And after some positive feedback, I did put it out there to see if I can gain some traction.
But my question to everyone is how do you handle the lock-in with your own personal knowledge bases against subscription-based tools?
people are forgetting how many copilot licenses are sold coupled with gpt models, adoption is pretty low and they are making ton of money on that, "allocation of unused tokens"
https://youtu.be/PfHu-UJaK0Q?is=bp0HygfjqBJBDpoO