Never really thought about it before, but i guess the mainframes can run python. A CMR came thru this evening that i had to approve as the ( temporary ) manager of the vendor/mainframe group, "update bla bla python" ..  wasn't expecting that.

I suppose it makes sense just never really thought about it. Tho i started "official" professional life ( not freelance, which i did before ) in the mainframe world as a COBOL guy.. it needs to go away. its long past its prime. Long past.  ( and where i learned that being a full time developer was NOT my path in the IT world.. sure, as needed, but not my primary task  )

May not be useful to everyone, and i know we have a 'mail server' here, but i needed to test some mail sending, random from/to addresses and not so 'formal', and really just needed a minimal server that would accept anything sent to it, and then let me see it for review before i delete and run another test.

Was about to write one, but figured i should look first.. save me trouble. I ran across 'Inbucket' which is perfect for this... really cool tool, for this sort of edge-case.  its old, but not super old ( hey, if it works, why update often? ) and while the deb release was broke, their docker container worked fine ( too lazy to compile it )

shoot .... and saying 'here' i didn't mean i was thinking of spamming uncensored to run my tests...  Just that citadel has one i could have used locally, but its production level, so more overhead than i needed.

Lesson learned. Read the docs.

Not that anyone cares, but i use PVE and have for a long time now. I have some desktops accessible via Guacamole to friends and some family ( and myself back when i was in office ). Nothing fancy there i agree, just background. Ran into the issue of machines being shut off, and cant turn them back on unless you go in the admin interface, so users are outta luck and they have to ping me, i have to get on the server bla bla bla.

Guac supports WOL.  PVE does not, yet. ( donno why, everyone wants it ) Long ago i tried to write something that would monitor for packets, then either run a CLI on the PVE server or hit the API. Failed several times. Just could not figure it out and gave up.

Fast forward to AI times.. "hey bot make me a server.... "    In its instructions on setup, it said to insert the 'wol server' IP address for the broadcast destination, and NOT the subnet..  Doh.  Missed that totally before. But hey, now its not hard coded in the program and i have a pretty web interface to update things with. Lol. 

So this weeks side project is to create a web app dev tool, totally database driven, and xml import centric.

Basically you import xml files of what you what, and the db serves everything, pages, forms, data, script schedules, email templates, whatever, its all in the DB. No random pages or code floating around. Not really a new concept, but not real common as most still write 'files' around. ( and its python, my preference for code ).   there was a opensource .net tool perhaps 15 years go that i used that did this, it was a great thing. Then scumbag Oracle sued them into oblivion due to "you cant have a database fed app, that is our idea!!!". I dont know if they actually somehow won that insane argument in court, or just won due to attrition as they had more money. 

Instead of a 'graphical designer' like they all have, ( might do that later, donno ) its going to be 100% bot driven at least at first. Its got a chat create/edit page, "i want bla bla bla" and it creates the XML for you.  You can review then commit to the db. Of course you can export existing xml projects ( "modules" ) back into it, add/remove features, etc.  has some DB admin stuff, as well as other admin functions, so its not like you are locked out, just that the AI model is going to do all the heavy lifting for you and all you do is sit and tell it what you want and test... 

its about 2/3 done. And ya, its 100% bot code, with me *heavily* steering.  Sort of an experiment to see if i can pull it off.  If it works, it could be neat. Ill shove it out on GitHub with my other stuff. 

But the question is... is this going to be considered no code, or low code?  LoL

Doubt anyone cares, but its close enough to be put on GitHub..

https://github.com/Nurb4000/PythonAppFoundry

So, is it wrong to have a 'LLM coder helper' develop its replacement?  LoL. that is this weekend's project.  Mostly out of paranoia, if the project that im using goes commercial like nearly all the others, its gone.   This way i have 'mine' to use as it will never be sold out.  i have found that using a 'guidance tool' is soooooo much more effective than doing it "manually' via a regular interface, so its well worth the effort.

And ya, its up on GitHub now too. Still a few things i want to do, but its more than usable.

And i guess i need to sign up for openrouter, as they have a free tier. I want to test it on 'external' APIs and not just connecting to my cluster. And its the only free one i know of.   Don't care if it doesn't really do well with whatever they offer for free, just to test that it works.

ya, this is what happens when i'm bored and its too hot to go out.   lol.

I actually fed it its own code as a baseline for feature set and operations. Its open. Their back end is not, but the front end tool-set is.        It was more of a joke, like when you get a new hire 'go train him' and you are in reality training your replacement.  

Wed Jul 15 2026 09:13:55 EDT from IGnatius T Foobar

So, is it wrong to have a 'LLM coder helper' develop its replacement? 

"Wrong" ?

At this point that's how the commercial vendors are doing it too.

Remember, a code assistant is only as good as what's been fed into it. If you set it to work on a large, established code base, it's going to ape the style and patterns already found there. I use it every day now and the code it writes "looks like mine."

On the other hand, if you tell it "write me a replacement for XYZ" it's going to do the same thing a junior dev does: scrape crap off Stack Overflow and write a pastiche of what's already out there. Or if the source for XYZ is out there it'll probably just crib that.

...and added a KB system so it can track previous projects, and use them as 'guidance'. Should save on prompt context use ( a big deal with LLMs ), and just overall 'efficiency'.    And it now can create its own 'tools' when it determines its needed for a project.  ( tools are things it can 'do' on its own, like read files, edit, use an API, etc )

now i think its down to just fixing bugs and better model 'steering'. the core is there, and it works.

Turned it loose on refracting my 'app foundry' this morning, its still going..   Who wants to bet it wont work afterward ..  lol 

aaaannnddd it fried things.  Sort of what i expected, in the way i expected.  But, i think after a few iterations of 'hard steering' later tonight it can repair the damage without me doing a lot of copy/paste   (short of a rollback and redo)

Sun Jul 26 2026 11:24:42 EDT from Nurb432

Turned it loose on refracting my 'app foundry' this morning, its still going..   Who wants to bet it wont work afterward ..  lol 

And on a lighter note from my other posts..

After about a week of letting code bot run in the evenings..  i have a 3/4 baked GUI workflow system to test.  Not yet usable for daily life, but its close.  No real need, but since its free, might as well give it stuff to do that is interesting. But after this, not sure what to do start on next.  Finished by BBS thingie, music is ongoing and never done, but it needs ME to be heavily involved , i cant let it 'cook' .. I have my resurrection of the Python AppFoundry, and its done.  A replacement for opencode is functional in case it ever goes away, since you cant hardly trust anyone these days not to go commercial after getting free testing and bug fixes.. and yank their source off GitHub. They are already are selling LLM time.. so...  

Running out of ideas.