Usage

lxask "what you want to do"

Say what you want, the way you would ask a colleague. Name the files, sizes, ages and ports you mean: lxask checks that they appear in the command.

#Reading an answer

$ lxask "delete all .tmp files in this folder"
$ find . -name '*.tmp' -exec rm {} \;
  find  Find files or directories under a directory tree, recursively
    -name  Base of file name (the path with the leading directories removed) matches shell pattern pattern
    -exec  Execute command; true if 0 status is returned
  rm    Remove files or directories
  warning: deletes every file the find expression matches

• The line starting with $ is the command. lxask never runs it.
• Below it, each program and option, as the documentation describes it. Where the documentation says nothing, nothing is printed: an explanation can be incomplete, but it is not invented.
• warning: marks commands that delete, overwrite or reformat.
• unverified: marks what lxask could not confirm. Treat that command with more care:

$ top -b -n 1 -l 1 -o %MEM | grep -v top
  unverified: top: -l is not a documented option
  closest documented example: ps --sort size

Common unverified lines:

lxask has already tried up to three answers before showing one with such a line.

#Requests with several parts

$ lxask "find the busiest process, see what files it has open, and stop it gracefully"
1. Find the process with the highest reported CPU usage
   $ ps -eo pid,comm,%cpu --sort=-%cpu | head -n 2
   ...
2. List the files open by the selected process
   $ lsof -p PID
   ...
3. Send SIGTERM to the selected process
   $ kill -TERM PID

Each part becomes its own step, answered on its own. Words in capitals like PID are for you to fill in from the step before. A note: at the end says if something you asked for is in none of the commands.

#Ruling a program out

Say "without X", "don't use X", "instead of X" or "X isn't installed":

$ lxask "which process listens on port 8080, without lsof"
$ ss -ltnp 'sport = :8080'

These requests, and requests about proxies, are answered with a version of the model trained for them (how it works).

#When lxask asks back

A request too vague for one command gets a question instead of a guess, and exit status 3:

$ lxask "add a proxy"
"add a proxy" can mean several things. Which one? Ask again naming it, e.g.:
  this shell session  lxask "set an http proxy for this shell session"
  the whole system    lxask "set an http proxy system-wide for all users"
  Git                 lxask "set a proxy for git"
  APT                 lxask "set a proxy for apt"
  ...

So far this covers proxies only.

#Straight into your prompt (Ctrl+G)

Add one line to your shell's startup file:

eval "$(lxask --shell-init bash)"     # in ~/.bashrc
eval "$(lxask --shell-init zsh)"      # or in ~/.zshrc

Then type what you want to do at the prompt, as plain words, and press Ctrl+G:

$ count the lines in data.txt          <- you type this, then Ctrl+G
$ wc -l data.txt                       <- the line is now the command

The command waits on your prompt, to edit or run with Enter; nothing runs by itself. What lxask says about it (unverified:, warning:) is printed above the prompt. A request with several steps becomes one line joined by &&. When lxask asks back or answers about itself, your line stays as you typed it.

With lxask --serve running, Ctrl+G answers in about 2 seconds. To use another key, bind the widget yourself: bind -x '"\C-x\C-l": __lxask_widget' (bash) or bindkey '^X^L' __lxask_widget (zsh).

#Keep the model loaded

Most of an answer's time on a CPU goes into loading the model. The installer asks whether to keep lxask loaded between questions (recommended, and the default):

• On (serve = auto): your first question starts lxask in the background and is answered by it; later questions take about 2 s instead of about 7 (Accuracy and limits). It holds about 1.3 GB of memory while loaded and stops itself after 30 minutes without a question, so the next question starts it again.
• Off (serve = off): every question loads the model and frees it after.

Change it in ~/.local/share/lxask/settings, or for one shell with LXASK_SERVE=auto or off. Only one copy is ever started: questions asked while it is starting wait for it.

By hand, whatever the setting:

lxask --serve &        # loads once, then waits; stops after 30 minutes without a question
lxask "show disk usage of each folder here"
lxask --stop

Every lxask call uses a running server when it can, and answers the same way. It listens on a socket only your user can open ($XDG_RUNTIME_DIR/lxask/, else ~/.local/share/lxask/run/). --idle MINUTES changes the timeout (0: never stop).

To start it with your session, a systemd user unit:

# ~/.config/systemd/user/lxask.service, then: systemctl --user enable --now lxask
[Service]
ExecStart=%h/.local/bin/lxask --serve --idle 0

#When an answer is wrong

lxask --bad                                       # the last answer was wrong
lxask --bad "should list directories only: ls -d */"   # with a note (quote * and ?)

The last answer is kept in ~/.local/share/lxask/last.json; --bad adds it, with your note, to feedback.jsonl beside it. Nothing leaves your machine by itself.

To send reports to lxask's developers:

lxask --share          # the last report
lxask --share 3        # the last three (up to 10)

This prints exactly what would be sent (your question, lxask's commands, your note, the lxask version and model file) and a link to a new GitHub issue filled in with it. Open the link, remove anything private (the issue is public), and submit. lxask itself sends nothing; a GitHub account is needed to submit.

Without a link: open an issue with the "Bad answer" template on GitHub.

#Options

#About lxask

$ lxask "who made you?"
lxask 0.11.0: an offline Linux command assistant.
Made by Aaron Thomas (arxhr007).

  project  https://github.com/arxhr007/lxask
  GitHub   https://github.com/arxhr007
  website  https://arxhr007.github.io
...

lxask --about prints the same. Questions about lxask itself ("who made you", "who created lxask", "what are you") get this answer instead of a command, without the model; a Linux question that only sounds similar ("who created this file") is answered as usual.

#Exit status

#In scripts

cmd=$(lxask --command-only "count the lines in data.txt")
echo "$cmd"      # read it before you run it