Nextcloud Hub 10 – your unified, modular digital workspace - Nextcloud
Welcome Nextcloud Hub 10 - reinforced performance, deeper integration, and dozens of new features that will make your day easier!
You're viewing a single thread.
It seems that while this update once again contains lots of (unncessary) AI bullshit, it also has some other positive changes, such as :
Overall, I am still disappointed by the direction towards AI that Nextcloud has taken, but I suspect that it makes them more money. Also, apparently large organizations / enterprises who pay Nextcloud GmbH for support, are also allowed to influence the roadmap of the software, which probably has something to do with jumping on the AI bandwagon.
However, despite this new direction, Nextcloud still does the job for me: Easy file syncing, with document collaboration, calendar, contacts and basically act as a whole replacement for Microsoft 365 / Google Workspace / whatever with little to no problems.
The AI support doesn't hurt you if you don't use it - and they've done the right thing by making sure you can do things locally instead of cloud.
Here's what AI does for me (self-hosted, my own scripts) on NC 9:
When our phones sync photos to Nextcloud a local LLM creates image descriptions on all the photos, as well as creating five tags for each.
It is absolutely awesome.
Which LLM do you use? Ollama?
I didn't even think of trying to use a LLM for this. I have been kind of underwhelmed with dedicated plugins for this. Maybe Ollama will finally satisfy all my needs.
But it cannot recognise people, right? But tags and descriptions would be better than nothing.
Ollama as a general LLM server and then LLaVa as model
I agree. (Digital artist here) I hate the Ai hype machine madness but also understand trying to find ways to make it ethical and useful to eliminate tedium. As it should be for in the first place.
In the tech sphere, they likely have to play along to stay relevant since everybody's all hyped about it.
How's this local LLM of yours? Does it tend to be more accurate than Recognize? It's integrated into Nextcloud?
Recognizing objects and people in pictures locally is one of the best uses I think. ...and maybe if it can stop auto-tagging random EDM songs as "country" or "rap" when they sound nothing like, I'd be excited about that!😂
I've set up OpenWebUI with the docker containers, which includes Ollama in API mode, and optionally Playwright if you want to add webscraping to your RAG queries. This gives you a ChatJippity format webpage that you can manage your models for Ollama, and add OpenAI usage as well if you want. You can manage all the users as well.
On top, then you have API support to your own Ollama instance, and you can also configure GPU usage for your local AI if available.
Honestly, it's the easiest way to get local AI.
How did you get your nextcloud AI workers to actually function in docker? Im not using the AIO image and I attempted to edit the systemd script provided in the docs but it fails :/ manually running the do command that I modified works until it runs out of time
@troed: That sounds awesome, and something on my todo list. Do you mind sharing how your upload and tagging AI-scripts work?
Honestly I probably should do more research but how is NC for photos compared to something like Immich or PhotopPrism?
I've been using NC since single digit version numbers
The performance increase on v30 is VERY noticeable, especially loading photos.
it's come a long way in the last few majors. I've been using it since v7 and it's been a struggle at times with users complaining about how glacial it could get. These days it's pretty blazingly fast comparatively, I get no complaints.
damn, RTL text is no trivial matter, kudos to them for getting that done
Thanks for the summary!
