fulldelusion.png
[Hide] (115.7KB, 1889x684) >>19625
Hm. Nah, not yet.
>>19636
Atm it isn't completely offline yet. I've a shitton of toggles for whether to allow SearX to go find stuff online, have ollama look for model updates, etc..
>Is this hard to install?
No, because you can bootstrap it. Once I got Qwen going by following the instructions on ollama's website, I just kept asking the Qwen model stuff like:
>How can I get you to be able to edit the system config files yourself?
And then once that was going, it was basically off to the races, as I could just ask the Qwen model to dl other models and configure things for me like "go get the obliterated versions going for me" or "figure out why you aren't working with my GPU card correctly", and then code up the pages and local webserver, install/run comfyUI, install/run SearX (I've gotten to the point now where I can redirect SearX to only search this "offline" web), make SearX work together with Qwen queries, and then once I had that, I could just ask it to figure out how to find the open models for image creator, radios, etc., etc..
The project has grown quite a bit, but I've done VERY little work the entire time and have basically acted as a top-level project executive as I let it churn away for a few days to set things up while I was away at work. I am basically very lazily AI slopping the whole thing together. After I did this, I've been having it use these modules to try to make games for me, and I'm playing with the idea of making an offline version of itch.io right now.
>What are the system requirements?
I'm not sure what the actual reqs are, but atm I'm not running anything special--most of these parts came from PCs my workplace was throwing out. It's a Quadro M4000 and 48GB of RAM sticks.
>You have to run your own server to it, correct, but does it have to be a real imageboard's software or just look like it?
This is not a genius operation. I told it, "Put all the messages in a giant json file, and make something that displays it like a messageboard. Organize the json so which board/thread/timestamp/who posted can be identified, and then run a model to make replies to other posters/new threads, but mostly to me." I mean, a few more project description words that that, but that was the general idea, and then it just coded it up for me.
Listen, this is AI slop all the way.
>Why would you want literal fake news?