Sorry, I just saw the recommendations. I’m using a Matrix server myself. And it’s connected to the internet, since I use it 24/7 and on my phone, etc.

I guess technically, most protocols can be used in an internal network. But maybe you’ll need to put in some extra effort, for example if a platform requires SSL certificates or something like that.

I mean you could try… If it asks for a hostname, just put a local hostname in. Or the IP address. Or set up a DNS entry on the router. And see if it works.

Or try something like RocketChat, or depending how your team’s workflow is, maybe you don’t want a messenger. But some (online) collaboration platform more focused on documents, like Nextcloud.

I think the added benefit of an OpenWRT router is, you get 3 more ports (for your TV, Playstation and PC), plus a Wifi network. And it’s really hard to break it. But a MiniPC with OPNsense, of course will be more powerful. And some more advanced things have been notoriously difficult to set up in OpenWRT, maybe OPNsense does it a bit better.

I think some people here recommended Snikket. It’s supposed to be easy to install and modern. I don’t know what components it’s made up of. It’s a dockerized XMPP server + Apps.

I dislike it. Usually I’d use packages from my Linux distribution. Or package it myself and maybe upstream the effort if my distro has a user repository. Now (this way) it’s down to everybody download random files from the internet and execute them. Specifically what every Linux tutorial instructs you not to do. Plus there’s no updates, no security, no version control or transparency. It’s not licensed in any free way, so I can’t fix it or adapt it to my liking, I can’t help you write better Python code…

But it’s your software project. You’re perfectly fine to do whatever you want with it. And it’s certainly commendable to write software, whether you do it for yourself, or put it out there in some way.

Yes. As far as I know, any gguf file should be completely safe. There had been some bugs/security vulnerabilities early on in llama.cpp, but they fixed that and I think overall, they have a good track record.

Issues might come after that, if you run some Agents on top of it, and give them access to your computer. But you don’t have to do that. If you just talk to it, I don’t see any reason to be alarmed. Other than the usual stuff. Keep using your own brain once in a while, and don’t blindly trust what AI Chatbots tell you, they give inaccurate information all the time 😅

Shouldn’t the upgrade also update the bootloader’s default entry to a new kernel? The way I’ve been doing it was apt update && apt dist-upgrade. And then reboot once every 1 to 2 years if I feel like it, am bored, or there’s all these news articles about a severe bug in the kernel.

Syncthing or Nextcloud. There’s a bunch of Linux sync software: https://awesome-selfhosted.net/tags/file-transfer--synchronization.html

Traditionally, you’d just put it on a NFS volume and be done with it. Or make it a boring plain old independent laptop with nightly backups configured, if your users always work from the same machine and don’t like… switch to a different computer in the middle of a task.

I’ve never heard that story. I think they might be hallucinating or trolling. Of course if you pull random Docker containers or execute some Github project to try new AI, you’re running other people’s code, and that could do arbitrary things…

But that’s not what we do. Usually, we download models in safetensors format, or gguf. And those are specifically designed to prevent this very thing, and not contain executable code.

Tools and MCP servers are a different story. Once you give your LLM access to the internet, it …well… has access to the internet. It mostly does what it’s supposed to do. But there’s occasional stories how someone’s AI Agent deleted all their email. Or reproduced some scifi story tropes and tried to use the internet to blackmail their user. AI can also make mistakes. Like you tell it to write a software project and it accidentally includes your password and API key. Or tell private information about you to other people if you grant it generous access to everything. The news about OpenClaw is full of hilarous anecdotes about things going wrong.

I didn’t have any luck with some uncensored Qwen 3.5 either. It always reasons about the guardrails. And it leans towards weaseling itself out of the situation. And the 3.5 version goes on for 1500 tokens anyway, just to think about how to respond to “Hello”.

I didn’t do a lot of LLM stuff lately. I’m also looking for a new local model which isn’t censored nor a sycophant, nor overly verbose and repetetive. But I guess I see that with a lot of models. And lots of the supposedly uncensored ones will give you the kids version of a murder mystery story, because they’re still averse to violence, conflict, taboo and all kinds of things.

And a lot of internet recommendations are older models from at least a year ago?! At least I didn’t find any perfect fit (yet).

I have a port forwarding without any tunnel to third parties and Wireguard.

The entire page is an advertisement for an AI tool that helped uncover it. Guess that’s the demonstration on how it augments a report.

I think there’s pros and cons to everything. That way would have been less of a dickhead move towards the Forgejo developers. But a big letdown to admins as they don’t know what’s up with the software they’re running on their servers. The way the author chose gives some new intelligence to admins, and they can now act on it, since it’s public knowledge. But it’s annoying to the devs.

I guess I as a Forgejo user am kinda greatful they did it this way. Now I got to learn the story and can allocate 2h on the weekend to see if my personal Forgejo container is isolated enough and whether the backups still work.

(But that’s just my opinion after reading one side of the story. Maybe there’s more to the story and they’re being a dick nonetheless…)

Edit: And regarding just dropping the security team an informal mail… I don’t know if that’s clever. You’d normally either follow some security policy, or don’t engage. Sending them other kinds of mails which violate their policy (an internal carrot) might not be the best choice.

Thanks. Maybe Agent Zero is a bit too close to the “usual” security model for my liking. Seems they also tell me to run it isolated and not connect it to private data and production systems… But that’s kind of what I want. I’d like it to screen my email inbox or move the remaining spam mails to the spam folder. But I thought there maybe was some sane approach where a human programmed the email adapter of it. And I can just configure the agent to stick to read permissions only, so it’d be fine.

Thanks for the other link. From reading the list, I think crewAI and smolagents are closest to what I want. I mean I don’t have an exact use case. I just figured since everyone and their grandma supposedly has AI agents these days. And AI is supposed to my life better, I’d try it. Idk. Let it sift trough my email inbox. Some online RSS feeds and the changelogs of some open-source projects I follow and alert me if there’s something interesting. Or if there’s something going on in a pull request I was part of… Maybe it can help with some other things. Or be a FAQ bot for all the knowledge I stored on my computer… Or generate a cat picture and send it to me via chat at lunchtime to brighten up my day. Connect to my Home Assistant and ping me before I leave the house if the train is delayed, it’s icy conditions on the road. That’s roughly what could be my needs.

But I want something more grounded than OpenClaw. It’s probably easy to build in some permission system and come up with separate agents for tasks, so the email agent can’t blackmail me with information from my knowledgebase, or delete the inbox. And sure I could use LangChain. That’d do it. But I’ve tried, and that’s just a lot of work. I’ll end up coding all the workflows myself. Figure out the prompts or steal them from another project. Reinvent how planning and subdividing tasks works… Copy lots of boilerplate code to start a vector database and then do RAG. Memory, skills. I’ll have to write all the email, chat, RSS, webcrawler integrations myself. A scheduler, background tasks. Then code an entire UI because what they have is more for testing and very straightforward chatbots. And it just escalates to a 100h+ Python project. For what I think must have been written several times already?!

Yeah, OpenClaw is the lunatic approach I mentioned. It’s hilarious. And a nice idea. But it might also delete all your data, write a nasty email to your boss/wife, leave you with a $200 a month API bill… It can also do a lot of things and it is fun. Especially reading what crazy things are going on in that ecosystem. But it’s not what I’m looking for, right now 😅

Thx very much. That’s valuable info. I edited my comment and crossed it off my list of software to evaluate for future projects. I already got the vibe-coding and a bit of sketchiness by scrolling through the latest commits and issue tracker.

Thanks for pointing it out. Yeah it does. I just copy-pasted what I found and didn’t check.

Laptops are designed to be fairly power-efficient. I don’t know what yours does. But mine goes down to only a very few watts if idle and the display is switched off. There’s the Linux tool “powertop” which shows power consumption and it can also tune most components to go to low power. Sometimes there’s also a power profile setting. That shouldn’t be on “performance” or anything like that. I don’t think Linux has more to offer, except sleep with wake-on-lan.

Erotic roleplay and messing around 😊

Uh, it is a bit more involved. The Arch Wiki has a lengthy article on it: https://wiki.archlinux.org/title/Wake-on-LAN

Basically, you’d enable it in your BIOS/firmware. Make sure it’s enabled in the network driver. And then you need to figure out a way to send such a magic packet. You can make your router or another device in the network send it. Or do a port-forward or send it through a VPN.

I use my homeserver for it. It’s located in the broom closet and on 24/7. But there’s ways to do it with a laptop. You can inhibit standby and let it run contunuously. Or configure Wake on Lan and wake it up before you use it… I mean a switched-off computer obviously can’t do any computation.