I am sure to upset a lot of purists with this, but as a hobby “sysadmin” of my own linux server at home, an AI built into the shell would do wonders to me.
I found bash-ai, but that seems to me like little more than a wrapper for ChatGPT queries.
In an ideal scenario, I’d like to employ a pre-trained, locally run AI that has access to all files, including config files and the list of installed packages, so it can, for example, tell me what specific lines in config files do or if a package I have already installed is good for a task I want to do or what the merge conflict between configs during an upgrade is about.
Is there such a built-in AI?
Might wanna hold off on that friend
I am aware that someone let their coding agent delete their database. I am not looking to vibe-code. I want a “Clippy”, if that makes sense. Not an agent that does stuff for me, but that answers my questions.
Far be it from me to yuck anyone’s yum, but did you have to pick almost the worst possible analogy to describe it? XD
“I don’t want it to be like the butt of today’s jokes, I want it to be like the butt of 30-year-old jokes”
The only differentiating factor here is time.
And an anthropomorphic representation, I guess.
Are you really ready to trust it? I use DDG’s agent frequently, when my query-fu fails, but I’ve gotten bogus, outright lies in response. They are good for helping me reframe my questions for the regular search engine. I don’t think I’d rely on LLMs yet for technical guidance.
In dealings with current LLMs I’m regularly reminded of the old truism: “it’s better to spend ten years looking for the right teacher, than to spend ten years learning from the wrong one.”
Have you tried solutions like “thefuck”? Maybe a non-LLM shell helper would be safer, and be less likely to teach you wrong or bad solutions.
Not sure if it will handle queries on bash commands. But, this video covers setting up and running a local large language model (LLM). IIRC, he specifically sets up DeepSeek, but it looks like you have a couple options. However, one caveat is that these LLMs run much faster with with a good GPU backing them up. If you don’t have a reasonable gaming GPU, you may be waiting a bit for answers.
I think there’s a bunch of AI enhanced shells and shell assistants… Shell-AI, Vity, Warp Terminal, Shell Genie, Butterfish, yai, aichat, aider, tgpt… You can get plugins for zsh or fish and other shells, we have agents with tool access, it’s in some IDEs…
You can get something functional using ollama for the model, aichat for the front end, and then making a script to generate a session file with all the relevant data you want included. I don’t know how effective it will be because there will be so much data.
