fonix232

“The best” depends on what you really need.

Authentik and Authelia are two of the top most quoted solutions, but they aim to be an “everything in one” package, with tons of features a simple homelab will rarely need. They’re aimed more at enterprise-like setups.

On the other end of the spectrum you have e.g. Pocket-ID, which only does OIDC by itself (you can hook up an LDAP server as user database though); but that tradeoff means it uses little to no CPU time, and, in my experience, around 130MB RAM. Given RAM is getting more expensive than gold, one needs to consider memory footprint, especially for older homelab setups. We can’t all be running 256GB nodes after all.

So instead of hunting the mythical “best” solution, decide what features you really need, grab a list of providers that cover those features, and compare them to see what fits your needs best.

Funny thing is that we have multiple examples of these AI fine-tunes backfiring spectacularly.

For example all the Grok fails on Xitter, and Elon promising to fix those results constantly, without any of those sticking.

I think the bigger problem is that there’s a segment of the population that’s gotten so dependent on “AI” (LLM chatbots), they literally can’t learn without it. Even the most mundane things they could’ve easily looked up via a search engine, end up as a question to ChatGPT. And no matter how many times you show them that the chatbot gave insanely wrong results, they’re still going to use it.

It’s incredibly vague still, that’s the problem.

Like, I see, if I wanted to roll my own shitcoin, I can’t publish the source code of it on Codeberg.

But, say, a Blockchain lookup app - technically it’s a cryptocurrency related project, but at the end of the day it’s just an app that connects to a public API and lists data from it. Is it also banned?

Or, I make an open source smart clock, from an LED matrix and an ESP32. This smart clock can link to YouTube/Twitch/etc. for viewer count, to GitHub for project stars/forks, to currency exchanges, and one includes a crypto exchange rate tracker for displaying current mainstream coin valuations (say, the top 5 most used coins). Is this project, since it’s crypto related, banned?

Or let’s say I come up with something blockchain adjacent, that could be utilised for cryptocurrency use… and what if I just publish a new cryptographic approach that then gets used by a new cryptocoin that suddenly becomes famous?

Or if my project uses a cryptographic system that comes from a dependency that was broken out of a cryptocurrency project?

Where do we draw the line of “related”? What level of relationship is allowed with the world of cryptocurrency? What degree of separation is deemed appropriate?

The recommendations so far are quite solid.

I’d also add that crowdsec with the right collections as a primary filter within your reverse proxy setup can be extremely useful, potentially even better than fail2ban, as CS delivers a literal “scan every packet and put them against these rules” approach - where the rules are crowd sourced, thus are more up to date than any manual blocklists, filter list etc. can be, and can spot even 0day intrusion approaches.

It depends on the app. You can use Authentik as an OIDC provider and allow apps that offer OIDC to log in through that - and that usually works for TV sticks and such.

Uhm, no.

Only if the licence allows. I personally go for A/GPLv3 wherever I can because I don’t want corporations to just take and not contribute back. But that means I can’t just copy MIT/Apache software, need to do my own implementation from scratch. Which I prefer anyway.

I’d argue that AI does change things:

• on one hand it increases the percentage of invalid/unusable images. A mis-labeled image is still potentially useful, an image of a toucan in the Arctic isn’t. This means that even large datasets become largely unusable because of the volume of fake images.
• on the other hand any AI picture that slips through, taints the dataset and results in expensive manual cleanup to ensure data is reliable.

why would I try a book REQUESTING system when I’m working on my own AIO solution?

I’m actually working on an all in one solution that sort of fills the roles of Readarr AND Calibre - without all the baggage of the latter.

Oh, and of course a KOReader library plugin that foregoes OPDS and works like the Kindle interface (“available books” synced from this backend, one tap downloads them, plus potentially adding series handling too in the future).

But right now it’s mostly manual downloads + CWA ingest + OPDS via KOReader.

Ah fuck, I knew I shoulda shorted it…