curbstickle

Please add the appropriate tags (see rule 7) and disclosure if required (rule 8)

Thanks!

OP did not make it and can’t do a disclosure, its not required.

You’re also handling bluetooth and wifi functionality with that permission, fyi.

Please tag and disclose as appropriate, refer to rule 7 & 8.

Completely agree

My current decision is between a new Motorola signature 27 with graphene (running under the assumption Linux will be able to make the jump there, which isn’t an unreasonable conclusion to jump to), or getting a mostly basic flip phone that can hotspot and using a umpc with it.

Where Linux phones have failed for me is the limited options for phones and on the actual phone part, so I feel like separation that out with a nice umpc and letting a phone be a phone is a worthwhile option.

We’ll see how things play out though!

The permission structure sucks.

Its a silent drop (no errors are returned), webview breaks connected to WiFi even without a local resource being requested, it lives inside nearby_devices so giving permission there includes this (and gives a broader spectrum of access such as Bluetooth and WiFi).

Its just shuffling the problem around.

Also, fuck Google.

The problem with graphene, unfortunately.

It doesn’t actually get away from google, at best its plugging holes in a (to me) sinking ship.

The solution is to avoid google going forward, for many reasons.

Haven’t been brought to 17 yet thankfully.

My long term solution is to not use android. My short term solution will likely be to use android even less, and be even more pissed at Google.

I wish it would because there’s no way to stop someone from creating an alt to promote their own projects and claim “they didn’t know it was AI” when they knew full well it was.

They will have to wait 30 days every time. This hasn’t been an issue for any post so far.

AGENTS.md in a repository are used to guide development.

That is not entirely true.

It can be. Yes, it provides agents with contexts - rule centralization, commands, conventions - which may be useful for external tooling for a project that runs on code live like this. It isn’t exclusively used for project development, which makes its presence not an indicator.

I see at least one: if the project doesn’t have a clear no-ai policy, then assume it is AIP. But that’s not my proposition at the moment.

Also discussed in the meta thread about this, and what results is projects being called an AI project that are not projects developed with - or even designed to allow interaction with - AI.

Also a non-starter.

I was just asking for a precision, but that’s because I didn’t check archives enough

There is precision in the detailed rules post.

If you want to see the prior meta discussions, you can look at my post history. I don’t often make standalone posts, so they are quick to find.

Its pretty far from perfect actually, and will often result in both false negatives and false positives.

Not a viable solution.

Just like the discussion last time on this - every user will not be able to be a forensically review.

This is demonstrated clearly with how you misunderstood the role of the agents.md file in the other thread, that was for AI to be used with the app, not for the app development.

So if you’d like to propose this, and people would like this, you’re going to need to include how or its simply unworkable as a rule. And I don’t see any option for how this could be realistically implemented.

Just about everything is Proxmox for me, though a few things are just Debian.

I use LXCs, in general I don’t like Docker so I avoid it in most cases. I have a couple VMs for it where I may spin something up to test, but if its around long-term it gets moved to an LXC. Usually because I prefer a bit more control on setup, and find it more manageable than having a bunch of mounts for everything. It also keeps things arranged in a way I prefer, gives me a single interface to access what I want, and makes backup more straightforward.

Whether I want a whole VM or just to run a single app as a service, I only have to go to one spot to do it.

Well thats what I get for skimming and only catching the first two sentences - thanks!

Please add the [CBH] tag per rule 7.

Just use nginx and certbot manually then, its just a pretty front end for them after all.

I’m fairly certain NPM (nginx proxy manager) doesn’t have any llm use inherently within the project, not sure on financial support. Other people have made context providers to use an LLM with NPM, but I don’t believe the core package itself does.

May be what meets your reqs.

Proxmox is based on Debian the way Ubuntu is, but more highly focused and without departing from main Debian strategies.

Proxmox is a kernel and set of tools for managing virtual machines and containers, in this case LXCs. These are different from docker, they share the kernel with the host but have isolation at namespaces (user id’s, process trees, network views, etc). They boot in seconds because they bypass hardware virtualization like you’d have with a VM.

I’d say gaming is better off as its own project effort. You could do things like GPU pass through and a dedicated VM for gaming, but you would be adding a pretty hefty layer of complication. If you dont need to convert av1 to something else, I’d just go with a 7th+ gen iGPU, and dont worry about it. Mine are 7th-10th gen, I have dozens of TBs of media, and they just chug along without issue handling somewhere between 4 and 6 streams at typical top use, mostly streamed without transcoding anyway. AV1 for the source file, you’ll be transcoding almost everything because most devices dont support native decode anyway, so I don’t personally find it worth it yet.

Just to mention, your Mac (if its an M series) can handle LLMs reasonably. I have an M2 Mac mini that runs some models, but my web interface for accessing those models is hosted on an LXC on one of my proxmox nodes.

It mostly gets used for local audio transcription and summaries, though occasional I’ll let it do other things like autocompletes for code (but not generating code, I don’t consider generated code to be useful).

I’d still say no.

You don’t really want a desktop task to crash and take down all of your services in one shot. As a VM with GPU pass through maybe, but not really in that beginner territory.

A solid call either way, easier to manage and safer imo

No, not really.

I have a whole complex setup including unique profiles and auth with unique limitations per user on services, like what books are available on the eBook server, what shows/movies they can see, even which block list is applied for DNS.

But things like app restrictions on a tablet are entirely manual. I kiosked off and will manually install via adb.

There isnt really an equivalent single tool I’m aware of.

For mdadm, it stores the metadata on the drives themselves. As long as your is is separate, yes, it should be just fine and detect automatically. After install it should just be a mdadm –assemble –scan and off to the races.

That said, I will always recommend backups. You never know what will happen.

In terms of docker, yes you can back things up and move them over. That said I recommend having your docker contents that need to be persistent as part of a mount somewhere, so that instead of needing to copy them around, you just have a compose that points to a mount.

Makes backups easier too.

Distro kind of matters.

Its a server, stability is the top priority. I’d avoid Ubuntu personally, and recommend Debian and Debian variants. For your first server (and your 4th+), I’d suggest proxmox. Its Debian with virtualization and a web interface, making it easy to manage services.

I’d also mention that you do not want a DE for your server. Its a lot of overhead for no real benefit, instead you should think of it as a headless machine you can remotely access - which is also what will make proxmox handy for you.

Intel wins in transcoding in most scenarios, so an intel-based system with an iGPU is a solid target. Full transcoding of AV1 starts with 14th gen, but that includes to AV1. For decode of AV1 for transcoding, 11th+. For anything else, 7th+.

With ram you can do more with more, but leveraging virt will make the load a lot lighter. Ive got systems with 32GB, but they are also hosting whole ass VMs and not just lxc’s.

PiGallery/PiGallery2 , Photoview, HomeGallery, lychee, or thumbsup might be up your alley.

Personally I use Immich and ImmichFrame.

MNT Reform.

Like framework, but without the support for DHH/Omarchy.

Please tag and disclose (if applicable) per rules 7 & 8

You can, but you can also run into some issues if you have git LFS, if one rejects for whatever reason, or sometimes branch issues.

Thats why I prefer to go to one repo and push out from there, but it is absolutely doable with git by itself.

Be aware that you may need to temporarily use reolink’s app to set it up first - turn on the ports, for rtsp/onvif/http(s) then you can trash the app.

I think the HA integration will use the proprietary port/api for setup and you can avoid that way too, but don’t quote me on that one, you may want to confirm first.

I’m with you, I don’t like having to do the same thing more than once

I set up an “about my repos” repo, which has a readme detailing the process, and explanations of what is where and what they are all for, which is what I point clients to. That also has a link to my personal forgejo to demo local git as well (which can also sync, btw) along with a note that it may be down for various reasons like updates or changes I’m making, which is why I also sync elsewhere.

Depending on what services you’re selling to clients, it makes for a neat and simple addition to the list of skills to demo!

And you’re kind of making me want to use a mini rack as an excuse to my wife for a 3d printer, but she already knows I’m not getting rid of my 12-15RU racks under my desks, so I’m going to have to wait until we find the property we want so I can build my workshop….