I have several Docker hosts in my home lab as well as a k3s cluster and I'd really like to use k3s as much as possible. But when I want to figure out how to deploy basically any new package they say here are the Docker instructions, but if you want to use Kubernetes we have a Helm chart. So I invariably end up starting with the Docker instructions and writing my own Deployment/StatefulSet, Service, and Ingress yaml files by hand.
That doesn't seem to have anything to do with what apps you have installed, just whether you have Play Protect enabled. I have Play Protect enabled, and I can still install apps without having to scan them first.
I wanted to dip my toe in the AI waters, so I bought a cheap Dell Precision 3620 Tower i7-7700, upgraded the RAM (sold what it came with on eBay) and ended up upgrading the power supply (this part wasn't planned) so I could install a RTX 3060 GPU. I set it up with Ubuntu server and set it up as a node on my home kubernetes(k3s) cluster. That node is tainted so only approved workloads get deployed to it. I'm running Ollama on that node and OpenWebUI in the cluster. The most useful thing I use it for is AI tagging and summaries for Karakeep, but I've also used it for a bunch of other applications including code I've written in Python to analyze driveway camera footage for delivery vehicles.
It came slightly after the IBM Portable PC (5155) which was released in 1984. That was a real luggable very similar to the Compaq. So I'd say the 5140 (which I've seen but never owned, I did think I was getting one once from a contest) was thought of as a luggable, but an improvement over what came before it.
I just saw this and created a quick deployment to my k3s installation. I created deployment.yaml, service.yaml, and ingress.yaml. I've got a few things already set up like wildcard DNS, cert-manager, and homepage so I've got a few extras in these files and kubeforge is already showing up in my homepage and deployed with https.
I ran into errors when I tried to download the schema, but then it suddenly started working, not sure why.
My first impression is that even with a high resolution laptop screen you're going to end up doing a lot of zooming in and out even for trivial deployments.
I imported the directory where I created those 3 yaml files and now have 3 connected boxes for the ingress, two for the deployment, and three for the service, but no interconnections between those three groups. It would be nice if the labels were cross connected between those groups, even better if when you were creating those groups from scratch you could specify the labels on one side then draw the edge connecting them and the label would get filled in on the other side and even get updated when one side changes. For example if I created a deployment with app.kubernetes.io/name of kubeforge and then was able to create the service object and link the edge and have that label connect.
2. To minimalize scrolling around for deployments I want to incorporate a "minimize" feature on nodes so that users can free up space on the graph. (https://github.com/kubenote/KubeForge/issues/6)
3. Interconnectivity is built in but currently only with inline data on nodes. Adding the capability to object refs does make sense. In the future I also want to create generic "string nodes" as a first pass at templating via the node graph. (https://github.com/kubenote/KubeForge/issues/7)
Thank you for kicking the tires and giving me feedback on how to improve!
Did you find the concept refreshing? And what would you like to see in it for it to be something that you would actively use?
After playing with it a bit more it looks nice but has fallen into an uncanny valley of usefulness. It provides a nice looking way to view the structure and a very syntax assisted way to change that structure, but no guidance whatsoever as to what you probably need to do. For example for me to deploy kubeforge I picked a set of 3 yaml files from something else and copied them to a new directory and then it was basically a search and replace to change the names and image file. So I had to really understand the files once and now I'm doing fairly mindless work to reuse them. But with kubeforge I have to deeply understand the details every time and select them from pulldowns - that's actually harder. I know I could have imported those yaml files first, made changes and exported them, but then I'd be going to the fields and changing them one by one, there is no equivalent of global search and replace. Now if I need to do something new/different then kubeforge can make it easier to figure out what options are at any given spot but still provides no guidance - I recently used initContainers for the first time in a project and could have used a "oh you want to set up an initContainer, here are the required elements for that".
I'd be happy to discuss, we'd need to figure out a better communication path but that's doable.
Some quick wins:
- make the workspace persistent so you can resume what you were doing the last time you visited - the Swagger UI does this, not sure how but I suspect browser side storage. This might be easy for you.
- more user feedback when you add a new block to the workspace. For example I kept clicking items from the toolbar wondering why nothing was happening until I finally realized a buttload of new objects had appeared in the workspace but weren't highlighted enough to notice them.
Some kind of auto-arrange would help a lot, starting with a pile of rectangles that you have to manually drag around to have any chance of making sense of it isn't ideal.
Graph diagrams look great but they fill a screen VERY quickly. There has to be a lot of thought into how to make the boxes as small as possible so someone can get the overall layout of the land, but can quickly access any details.
And the harder stuff would be some kind of assist dialog when you add a new toolbar item (I'm NOT going to suggest AI here, maybe just a dialog box which lists mandatory items and them other items in the order from most to least common). I understand this is tricky because you intend to support all API versions.
Your comment got me a bit curious and so I spent time playing around creating a simple Android app using webView for my personal website and got it working, the only permission I added was INTERNET. So what's the next level of awfulness - do I add additional permissions and then additional information can be presented to my website server, or would I actually have to implement an additional path to collect the kind of info these apps are trying for?
I depended heavily on Pocket for over a decade as a free user. It started to get bogged down with about 20k bookmarks. I used to spend hours manually tagging saves and the search function never seemed to actually return results. This time around I wanted a self hosted solution.
I looked at Walabag and Shiori before I decided on Karakeep. I just didn't like the UI of the first two. I already have an Ollama server and the AI tagging feature of Karakeep is far better than Walabag's, in fact the tag management feature in general is. And Meilisearch adds a really fast search engine to Karakeep that has allowed me to discover new value to the 16k bookmarks from Pocket after cleaning down from the 20k I exported, it's super impressive.
Now the less great news, Karakeep is much newer and less mature than the other options and currently only supports a SQLite backend and I really hope that changes. The only API for Karakeep goes through its web interface and so I don't think I even could export all my bookmarks. If the data was stored in a standalone real database like MySQL or PostgreSQL other options would be possible.
The AI tagging is AMAZING but it generates a LOT of tags and that makes the tag management screens in Karakeep difficult or impossible to use because they are overwhelmed. I am looking forward to the next and future releases which aim to help with this.
I use the Android app which works really well.
Karakeep does make your server into a web crawler and because of the little war on AI LLMs we're experiencing these days an unfortunate number of websites have started to fight all crawling. Karakeep uses a SingleFile browser extension which allows you to prove you are a human or log in to a website and then capture a page and submit it to Karakeep. This is a little awkward because you may end up bookmarking something once using the regular Karakeep extension and then see that you didn't get what you want and have to do it again via SingleFile. I'm hoping that at least a config list will be added so that the regular Karakaap browser extension will automatically invoke SingleFile for websites known to block bots.
