[Pr0m3th3us]

TLDR: Can i get a LLM installed on my system that scans my OS so it knows when i ask it something specifically pertaining to my linux setup it knows what dependencies are already installed, what setting i have, what files are already edited and how...etc?



So im not the best on linux and my memory is so bad i have a hell of a time remembering commands. Often i just use chatgpt to give me the commands for things. (i know its bad) Although as i go along and use it more often i eventually remember some stuff, but its pretty basic. I often find that LLM's just guess for the most part, or throw stuff out there and hope it sticks, anyone else get this?

Example: Yesterday i was going to use my Nordic NRF52840 Dongle to scan for Bluetooth devices using wire-shark. I have already set it up and used it about a couple months ago so i know its good but i don't remember the exact steps. So i ask chatgpt to help me set it up, mainly so i could just browse through the steps and remember seeing things that i know i had already done. (yea my memory is really bad)

This was my prompt

(i have a nordic nrf52840, how can i scan for devices with it on parrot os? i think i used nrf connect to upload a program to it already but its been a long time and don't remember)

some of the things i remember doing almost immidietly after i read them, kinda like "oh yea i remember now" moment. But then i got to this step

Code

git clone https://github.com/NordicSemiconductor/nRF-Sniffer-for-Bluetooth-LE.git

Im almost sure i had done this already but i figured it wouldn't hurt to run it again. So when i did it asked for my GIT Username and PW....odd but i typed them in and it didn't download, gave me some error which i input into chatgpt.....this started a whole 1.5 hours of BS. First chatgpt assumed that my config for GIT was broken, so it had me do a bunch of commands to get rid of the UN/PW, most of which it just fucking repeated almost ever time i would give it a new prompt or paste an error, which didn't work, after trying about 20 different commands and every time it was sure "this will fix it". It never did. So eventually i got the mindset that chatgpt doesn't even realize the link is broke, and sure as i get shit on my ass every-time i wipe THE FUCKING LINK WAS BROKE. When i told it this it apparently instantly knew? Like wtf how did you not know an hour ago?

Anyways i tried a couple other GIT clone links and they worked fine no UN/PW. Just download and install. Good to go. Fuck though almost every time i use a LLM to help me set something up on linux it take me around hoops and corners and honestly i cant rely on it being correct the first time it gives me steps 1-5 (or whatever).

[Wake]

TLDR: Can i get a LLM installed on my system that scans my OS so it knows when i ask it something specifically pertaining to my linux setup it knows what dependencies are already installed, what setting i have, what files are already edited and how...etc?



So im not the best on linux and my memory is so bad i have a hell of a time remembering commands. Often i just use chatgpt to give me the commands for things. (i know its bad) Although as i go along and use it more often i eventually remember some stuff, but its pretty basic. I often find that LLM's just guess for the most part, or throw stuff out there and hope it sticks, anyone else get this?

Example: Yesterday i was going to use my Nordic NRF52840 Dongle to scan for Bluetooth devices using wire-shark. I have already set it up and used it about a couple months ago so i know its good but i don't remember the exact steps. So i ask chatgpt to help me set it up, mainly so i could just browse through the steps and remember seeing things that i know i had already done. (yea my memory is really bad)

This was my prompt some of the things i remember doing almost immidietly after i read them, kinda like "oh yea i remember now" moment. But then i got to this step

Code

git clone https://github.com/NordicSemiconductor/nRF-Sniffer-for-Bluetooth-LE.git

Im almost sure i had done this already but i figured it wouldn't hurt to run it again. So when i did it asked for my GIT Username and PW....odd but i typed them in and it didn't download, gave me some error which i input into chatgpt.....this started a whole 1.5 hours of BS. First chatgpt assumed that my config for GIT was broken, so it had me do a bunch of commands to get rid of the UN/PW, most of which it just fucking repeated almost ever time i would give it a new prompt or paste an error, which didn't work, after trying about 20 different commands and every time it was sure "this will fix it". It never did. So eventually i got the mindset that chatgpt doesn't even realize the link is broke, and sure as i get shit on my ass every-time i wipe THE FUCKING LINK WAS BROKE. When i told it this it apparently instantly knew? Like wtf how did you not know an hour ago?

Anyways i tried a couple other GIT clone links and they worked fine no UN/PW. Just download and install. Good to go. Fuck though almost every time i use a LLM to help me set something up on linux it take me around hoops and corners and honestly i cant rely on it being correct the first time it gives me steps 1-5 (or whatever).

Understand where you are coming from completely with AI making you jump through hoops, I have never used It but have you tried something like OpenClaw? You can ask it to install things run terminal commands edit config files etc. I would recommend installing it on a second machine though. Some members might have better experience with It that can help you more. Are you still trying to work out the Washing system we spoke about before? Give me a convo I can help you troubleshoot It.

[DramaKing]

TLDR: Can i get a LLM installed on my system that scans my OS so it knows when i ask it something specifically pertaining to my linux setup it knows what dependencies are already installed, what setting i have, what files are already edited and how...etc?



So im not the best on linux and my memory is so bad i have a hell of a time remembering commands. Often i just use chatgpt to give me the commands for things. (i know its bad) Although as i go along and use it more often i eventually remember some stuff, but its pretty basic. I often find that LLM's just guess for the most part, or throw stuff out there and hope it sticks, anyone else get this?

Example: Yesterday i was going to use my Nordic NRF52840 Dongle to scan for Bluetooth devices using wire-shark. I have already set it up and used it about a couple months ago so i know its good but i don't remember the exact steps. So i ask chatgpt to help me set it up, mainly so i could just browse through the steps and remember seeing things that i know i had already done. (yea my memory is really bad)

This was my prompt some of the things i remember doing almost immidietly after i read them, kinda like "oh yea i remember now" moment. But then i got to this step

Code

git clone https://github.com/NordicSemiconductor/nRF-Sniffer-for-Bluetooth-LE.git

Im almost sure i had done this already but i figured it wouldn't hurt to run it again. So when i did it asked for my GIT Username and PW....odd but i typed them in and it didn't download, gave me some error which i input into chatgpt.....this started a whole 1.5 hours of BS. First chatgpt assumed that my config for GIT was broken, so it had me do a bunch of commands to get rid of the UN/PW, most of which it just fucking repeated almost ever time i would give it a new prompt or paste an error, which didn't work, after trying about 20 different commands and every time it was sure "this will fix it". It never did. So eventually i got the mindset that chatgpt doesn't even realize the link is broke, and sure as i get shit on my ass every-time i wipe THE FUCKING LINK WAS BROKE. When i told it this it apparently instantly knew? Like wtf how did you not know an hour ago?

Anyways i tried a couple other GIT clone links and they worked fine no UN/PW. Just download and install. Good to go. Fuck though almost every time i use a LLM to help me set something up on linux it take me around hoops and corners and honestly i cant rely on it being correct the first time it gives me steps 1-5 (or whatever).

Git is programmed to assume that if a clone returns a 404, to assume that the repo is private, not that the link is wrong. Simple mistake. Chatbots can be so dumb. Reasoning and higher-tier models are better, but this was probably GPT-5.5 Smart. I can almost guarantee you that Gemini and Grok handle the technical stuff better.

Anyway, use Gemini CLI or Claude Desktop. OpenClaw is good, but it's more complex and beyond what you asked for.

[5chn4pp]

The reason it didn't work out for you is because you used Instant instead of Thinking.

Instant:

https://chatgpt.com/share/6a0654ce-fabc-...4078dda68d

Thinking:

https://chatgpt.com/share/6a06537a-d1a8-...2c38df61ba

See how thinking gets it correct while instant has the broken link?

Also for local knowledge Claude Code is probably your best bet.

[Pr0m3th3us]

Understand where you are coming from completely with AI making you jump through hoops, I have never used It but have you tried something like OpenClaw? You can ask it to install things run terminal commands edit config files etc. I would recommend installing it on a second machine though. Some members might have better experience with It that can help you more. Are you still trying to work out the Washing system we spoke about before? Give me a convo I can help you troubleshoot It.

I looked into OpenClaw, it seems pretty awesome but also needs alot of technical setup. Also it requires API's from whatever LLM i choose. Im not quite ready to pay for a yearly sub yet and if i did i might as well just use Claude Code.

Git is programmed to assume that if a clone returns a 404, to assume that the repo is private, not that the link is wrong. Simple mistake. Chatbots can be so dumb. Reasoning and higher-tier models are better, but this was probably GPT-5.5 Smart. I can almost guarantee you that Gemini and Grok handle the technical stuff better.

Anyway, use Gemini CLI or Claude Desktop. OpenClaw is good, but it's more complex and beyond what you asked for.

I will try those out, i also used to use perplex and it basically was the exact same problems as i described in the OP.

The reason it didn't work out for you is because you used Instant instead of Thinking.

Instant:

https://chatgpt.com/share/6a0654ce-fabc-...4078dda68d

Thinking:

https://chatgpt.com/share/6a06537a-d1a8-...2c38df61ba

See how thinking gets it correct while instant has the broken link?

Also for local knowledge Claude Code is probably your best bet.

Link arent working for me. I googled it and it shows i can change it but i don't see the settings on the web app. I asked GPT and it said i literally have to tell it to think...rofl wtf.

[NoSuchAgency]

Codex CLI is phenomenal. Much better than Claude code in my experience. Plus if you already pay for chatgpt you will have access to it from OpenAI.

Run

`npm i -g @openai/codex`

[5chn4pp]

Link arent working for me. I googled it and it shows i can change it but i don't see the settings on the web app. I asked GPT and it said i literally have to tell it to think...rofl wtf.

Weird, here they are pasted on pastebin:

https://pastebin.com/kqxKVtdY

https://pastebin.com/FGQd7MiQ

[BritneyLeary]

I'll look into this. I know someone very skilled in LLMs.

[Pr0m3th3us]

Codex CLI is phenomenal. Much better than Claude code in my experience. Plus if you already pay for chatgpt you will have access to it from OpenAI.

Run

`npm i -g @openai/codex`

Yea see i don't have any subs and really don't want to pay lol But codex CLI looks good for sure. So basically its between that and Claude Code.

Weird, here they are pasted on pastebin:

https://pastebin.com/kqxKVtdY

https://pastebin.com/FGQd7MiQ

Working now idk what happened.

[DramaKing]

I looked into OpenClaw, it seems pretty awesome but also needs alot of technical setup. Also it requires API's from whatever LLM i choose. Im not quite ready to pay for a yearly sub yet and if i did i might as well just use Claude Code.

There are free LLMs.

Link arent working for me. I googled it and it shows i can change it but i don't see the settings on the web app. I asked GPT and it said i literally have to tell it to think...rofl wtf.

It's literally just clicking the '+' button.

[Pr0m3th3us]

There are free LLMs.


It's literally just clicking the '+' button.

Damn im blind af....

[NoSuchAgency]

Yea see i don't have any subs and really don't want to pay lol But codex CLI looks good for sure. So basically its between that and Claude Code.


Working now idk what happened.

Brother, I'm pretty sure codex cli has a free version that is relatively generous.

AND, if you have a PC that can run like a 30b model you could get a lot done forcing it to use to like the new gemini local model.

But yeah, try using the free chatgpt!

[Pr0m3th3us]

Brother, I'm pretty sure codex cli has a free version that is relatively generous.

AND, if you have a PC that can run like a 30b model you could get a lot done forcing it to use to like the new gemini local model.

But yeah, try using the free chatgpt!

I did some research and ChatGPT recommended for my system “Qwen2.5 14B”. There is a caveat though….

My current system for this Linux (parrot security) is a dell optiplex 7020 (I5-4590, 16gb ddr3) I am going to put my 3060 12gb in it and a 750w PSU. However the current case is SFF, so I have to swap cases with my optiplex 3010 mid ATX (my plex machine) and maybe do some mods to get shit to fit lol.

[Pr0m3th3us]

So after using Claude's sonnet for a couple of days helping me clone a drive using my BORG backup, im fucking sold. No bullshit replies that chatgpt does like "this will def work this time" and "why this works" (spoiler it never worked). Claude just went straight to the point and did it fast and right the first time almost every time, and its not even the code version. So when i can i will buy a year of Code for sure. Just need to find a seller here on HF lol.

[Pittbull88]

This is easy to do multiple ways the way I have done it I have notion with an mcp connected and my notion is built with instructions and best practices this is a part of my local ai memory