I have over 3k Steam entries (~2.5k real games). First I opened the Storepage of every single new Game, read the Tags, added every Tag (most of the time I tried to choose the first 10 Tags) to the Game to Categorize it. ~10 times “Add to…” per game. Fun isn’t it?
Then I found Depressurizer which was the only tool that made this bearable - but it’s Windows-only!
Sadly I didn’t manage to run it on Linux. Tried it under different Wine and Bottles setups, nope not for me! (Maybe I’m just to stupid to get it up and running?)
Three months ago I finally quitted Windows and forced myself to use Linux as daily driver! Glad I did it.
And I told myself: Before I start Windows just to sort my game library, let’s start to make one for the Linux Community! Directly on Linux, for Linux!
So I built SteamLibraryManager with PyCharm from scratch on CachyOS.
My App is available as AppImage (good for SteamDeck), AUR, .deb, .rpm, tar.gz:
yay -S steam-library-manager
GitHub: SwitchLibraryManager
What it does (just the highlights - check the GitHub README for the full feature list):
Smart Collections with full Boolean logic (AND/OR/NOT + nested groups) - Steam’s dynamic collections have been AND-only since 2018. So I had the Idea with my own “Dynamic Collections” called “Smart Collections”.
If you like to see a short Video of SLM
Auto-categorize by 17 rule types: Tags, Genres, ProtonDB rating, Steam Deck status, HowLongToBeat, Achievements, PEGI (Age Ratings), and more
Import all your non-Steam games: Epic, GOG, Amazon, Lutris, Bottles, itch.io, Flatpak, even ROMs with 16 emulator definitions
Metadata that survives Steam updates - we overlay your edits on top of Steam’s data so they don’t get wiped
Built-in auto-updates for AppImage users - downloads in background, atomic replace with rollback if something goes wrong.
Steam Deck: Responsive UI that adapts to 1280x800. AppImage works in Desktop Mode, survives SteamOS updates. No pacman hacks needed.
Tested on both of my SteamDecks - LCD (512GB) and OLED (1TB). On the LCD one it was a bit tricky because I installed CachyOS Handheld Edition on it and installed the AUR, Oled is original SteamOS where I used the AppImage!
It’s my first App, please be patient with me 🙃 I just want to give something back instead of using it just for my own.
TBH: AI tools helped during development - mostly for boilerplate, tests, docs and docstrings because I really hate writing documentation 🙄).
Architecture decisions, feature design, and all the tricky stuff (VDF binary parser, Smart Collections engine, Steam OAuth2) were done by me. Every line was reviewed and tested manually.
I’m not gonna pretend AI doesn’t exist in 2026, but this isn’t a ChatGPT copy-paste job.
It’s a vision I brought to life to help myself, and that I want to share now with the best OS community out there. No matter what Distro!
Linux is awesome, sadly it took me 30 years to realize that, using Windows only!
Greetings from Germany
Open Source community wasn’t happy that their code got used to train LLMs either.
Well, if you’re open sourcing your code, you can’t really complain when someone uses it, no?
There many different kinds of open source licenses. Not every license that comes with open source software, gives you the permission to do with it whatever you want.
None of the open source licenses prohibit the use of code for learning purposes.
A whole bunch prohibit commercial use, don’t be obtuse
Well, now we venture into legalese.
LLMs are not incorporating the used code (in theory), so the copy-left clause does not apply.
It’s like if you read an GNU/GPLv3 source code from something, learned from it, and therefore any time you write any future code you MUST apply GNU/GPLv3 to it. It’d be insane.
If we assume that training an LLM is like training a person, then obviously the copy-left clause does not apply.
If we assume that training an LLM means actually incorporating the code into the product, then the entire thing either needs to be open sourced, or cannot be used commercially.
Why the fuck would we assume training an LLM is anything like training a person?
Because it kinda’ is…?
The type of knowledge retained is massively different (understanding of concepts vs memorisation of patterns), but the concept is the same.
And, again, if you want to consider the use of copy-left code in LLMs as infringing upon open source licenses, do you also want to consider using effectively the same pieces of copy-left code as said infringement?
If a human writes a simple “Hello World” in C++ after learning how to do it from a copy-left tutorial, does their “Hello World” now requires the use of an identical license?
Human output is innately transformative, and our “training data” is lived experience vs pattern recognition. You acknowledged that but imo don’t really appreciate what it means. We are doing more than regurgitating inputs even when our output happens to be repetitive. A LLM is only regurgitating inputs even when its output happens to be unique. It is no more than the sum of its parts, and stolen parts mean stolen output.
Ah yes you are right. I am so sorry, you have convinced me.
As far as I can understand it the code that was used for training was publicly available under some sort of permissive license. As in “you are allowed to read it and use it, but conditions apply”.
If the llm now creates code based on that it should be under an open license as well. In this case, it is. In most cases it’s not.
i think the big problem for the open source/FOSS communities isn’t that their code is being used via LLMs for other open source projects, which I’m pretty sure is fine, but for closed source commercial projects, which is NOT fine and a clear licensing violation.
it’s the free–>commercial “loophole” (really outright theft, but here we are) part that’s problematic
There are different Open Source licenses. And the result is only one license. In example there are stricter licenses and you are not allowed to re-license it under MIT License. Therefore your quoted argument is not valid.