vegetaaaaaaa

I wish more people understood this.

Do I use LLMs to write software for my personal use? Sure. I still try to build it in an “incremental” way the same way I would write software manually so I don’t get 10k SLOC written in a week, but at some point, even reviewing 100 LOC changes takes time, so I just take a cursory look at the diff, yolo-merge-and-run to test it. It’s not critical. This is fine. I’m just exploring the problem domain and solutions.

But would I go as far as sharing it, making a damn git repo and advertising it on Lemmy? Fuck no. This is unreliable, inscrutable slopware tailored for my own use. Anyone with a local LLM or a 20 euro claude/codex/z.ai subscription can do the same thing in a few minutes of work a day.

A single 10-line patch/contribution to a human-written project, with contributors who understand the code, or even a well-curated comment in a bug tracker that helps devs debug an issue or clearly, has more value for the community than 50 vibe-coded projects.

I didn’t find one that would work entirely authelssly and which would allow negative entries

Have you considered filing feature requests for these?

Yes, even writing a proper issue report probably entails more work and brainstorming than prompting your way to a shitty solution. No offense meant, I do it as well. I know using a LLM and pumping out a working solution to a complex problem in a week gives a feeling of euphoria and power; this wouldn’t have been possible at all a few years ago. But there is absolutely no value in proactively sharing and advertising it.

I appreciate OP being transparent about it though.

they don’t go out of their way to make the self hosted option easy

Just follow the docs docs.nextcloud.com/…/source_installation.html ? The manual install is a simple webserver + PHP-FPM + postgres setup

I have automated it with an ansible role [1], there’s nothing complicated about it, really.

• Small 4B models like gemma3 will run on anything (I have it running on a 2020 laptop with integrated graphics). Don’t expect superintelligence, but it works for basic classification tasks, writing/reviewing/fixing small scripts and basic chat
• I use github.com/ggml-org/llama.cpp in server mode pointing to a directory of GGUF model files downloaded from huggingface. The use it from the built-in web interface or API (wrote a small assistant script)
• To load larger models you need more RAM (preferably fast VRAM/GPU but DDR5 on the motherboard will work - it will be noticeably slower). My gaming rig with 16GB AMD 9070 runs 20-30B models at decent speeds. You can grab quantized (lower precision, lower output quality) versions of those larger models if the full-size/unquantized models don’t fit. Check out whatmodelscanirun.com
• For image generation I found github.com/vladmandic/sdnext which works extremely well and fast wth Z-Image Turbo, FLUX.1-schnell, Stable Diffusion XL and a few other models

As for the prices… well the rig I bought for ~1500€ in september is now up to ~2200€ (once-in-a-decade investment). It’s not a beast but it works, the primary use case was general computing and gaming, I’m glad it works for local AI, but costs for a dedicated, performant AI rig are ridiculously high right now. It’s not economically competitive yet against commercial LLM services for complex tasks, but that’s not the point. Check old.reddit.com/r/LocalLLaMA/ (yeah reddit I know). 10k€ of hardware to run ~200-300B models, not counting electricity bills