I may be skeptical, but I’ll certainly be happy to get some insight on alternative ways of doing Camera2 api work!
I think it’s actually the other way around. People are using LLMs for the instant gratification of getting what they want but are not actually willing to maintain projects or the communities around them…?
Here is the proof of concept: GitHub - Logic-gate/sfos-jp2-camera2-bridge: An experimental bridge for Camera 2 API · GitHub
A parallel capture stack that bypasses the GStreamer camera path for still photos. Built against Android NDK locally so that the path looks like this.
App → Camera2Preview(GUI only) → sfos-camera2-probe → libsfoscamera2.so → Android Camera2 NDK → camera service → HAL → RAW/JPEG/preview
The breakthrough is getting sailfish to talk to Camera2 stack directly, without going through GStreamer | Camera1. libsfoscamera2.so is built as a Bionic shared library, not a normal sailfish glibc library, It links against android camera libraries like camera2ndk and mediandk. Sailfish side loads it through libhybris, this makes Camera2 usable from Sailfish by placing the Android dependent code in a Bionic bridge library and keeping the Sailfish app side simple. And since Jolla opened source the camera app, I used it as the basis. I further built the bridge and made many changes for the app to “barely” function.
Interesting perspective, perhaps you are right.
Just want to jump in to this thread a bit and say this: despite what every other new user of 2026 is thinking, having AI-coded apps is not likely to grow the interest on Sailfish OS. I just stumbled across this blog post:
The author was expressly mentioning the fact that the vibe-coded landscape of app development/distribution is something making him reconsider using Sailfish OS:
This app (nr. it’s about the WhatsApp client from OpenRepos) works. It can send and receive messages and also lets you send and receive photos. That’s sort of what I need, though I would also enjoy being able to call.
The problem is with it’s reliability. This app too has a Schrödingers message-state, where I can see a message as sent from my phone only to have it not show up on any other devices. It has also at other times sent messages multiple times as duplicates, something that can apparently get you banned from whatsapp.
The app is also heavily vibe-coded and far from even a remotely defenseable way. First of all, every single commit is treated as a release meaning you’re always on a git branch of the program. The developer can also push 7 commits and 17 “versions” on the same day. This application is in other words severely untested.
I recently came across a good microblog by Veronica Explains about how she avoids vibe-coded programs because you cannot be sure that the developer understands the codebase and will be able or even want to support it if/when something breaks. This is a good point, and was something that worried me when I first installed the program. It was slightly dulled down by the idea that the port technically only does the frontend whereas the backend is powered by whatsmeow, meaning that the developer must at least know something, right? Sadly, I have only come over things that to me confirm the opposite. Some selected comments from the forums include:
I gave the link to the repository to Claude and it claims it sets the thing that's used for the counter right now. user: Great work. You are so fast, that is amazing.... There are so many ideas in my head .... dev: Keep them coming. I plan to end my subscription by 18th of August I won't go away, but then (after Aug 18) it will take longer, because i have less tokens to use up thenIt’s one thing to have a slopped application on your phone. It’s another thing to be depandant on phone. It’s again yet another thing to be dependant on having to trust one with a big part of your private messages.
Sure, there is an app existing right now and even thought it’s super jank I guess it works. But even then, there’s all of this shit to have to deal with, and also the date of which the app will “die”, or at least the day that what seems to be the main dev (an LLM) will “leave” the project is only five days away. What happens if/when the app breaks then and the user it out of tokens? Or what if some regulation changes how the LLMS can work, what if the company gets fucked, the bubble pops, or if the LLM just implodes onto itself? It’s way too hard for me to be able to actually trust this stuff, and writing about it makes me realize just how risky it actually feels to have this as one of my main dependencies for a working phone.
So, I don’t know, if you are okay with having a vibe-coded ecosystem on a platform that’s supposed to be an alternative to the big-two - who are also pushing AI left&right around their ecosystem - then fine. But I do not think that the sizable chunk of people willing to look for an alternative to these will be happy to find just another platform where AI development is rampant.
I personally try to avoid LLM generated apps as much as possible. It’s not easy, sometimes I do need something like a screen recorder or a youtube client and I have to give in. I also put up with the fact that some of these people do try to be helpful, willing to understand what they’re doing and receptive to feedback. But it does not mean that I actually enjoy the idea of having LLM-based apps installed and using them.
It’s like me being your upstairs neighbor and ask you “hey, do you like to be waked up by my driller every Saturday on 8 AM in the morning? Of course you like it”. No, you do not like it. You just have no other choice. You wake up because you cannot not hear it (probably you’re too broke to get some proper earplugs or something or they don’t help you at all) and you cannot get back to sleep afterward. You do not enjoy being waked up by my drill every 8 AM on a Saturday morning.
Just my 2c as a new user and a non-developer.
PS: I am not using the Matrix client and neither the WhatsApp one 
I do think you’re right that some people, like me, might leave a project alone if it’s getting a lot of LLM attention because we can’t get enough juicy bits of work done without colliding with the bots. That could be. Though I don’t mind battling bots
This is great! Decoupling is everything!
Thank you for continuing to make sensible points in every discussion!
LLMs seem perfect for simple noarch apps, QML is super human readable/reviewable, py/js also mostly. It’s really hard to hide something malicious there. The black box problem starts with the monstrosities that ship their own everything so you get one big black box rpm with mystery systemd services included that do everything from managing your cpu governor to slicing bread. Problem is even bigger as LLMs (just like most new devs) include main.cpp in their project even in apps where there is no need for it (blame jolla docs for that), so you get a precompiled binary even when all logic is happening in qml/js/py and app should’ve been easily reviewable on device. As long as the source is provided the anti-LLM hysteria is getting really overboard though, as per OP, even devs who clearly label their creations as LLM assisted are getting hate (haven’t seen any of it, but maybe the comments were deleted?). Just don’t install LLM-assisted apps and leave the dev alone, how hard is it? (if they lie and/or avoid answering a simple question, might be deserving some, but yeah whatevs, the grifter case is not OT here)
Edit: and to quote the classic
Really hope this gets its own thread, amazing finding hidden 100 posts deep in a disaster of a thread
And another thing with LLMs, people will create useless elaborate hoops without understanding the underlying tech, just because LLM said so, you’re leaking all your metadata to third parties for literally no reason??? LLM said so!
