The aoc package is opinionated, it creates modules with the parse/2, part_one/1 and part_two/2 functions, and it creates tests if you fancy some TDD. If you do not like that, you can still use it just to fetch the inputs locally.

If you want to do your setup yourself I guess starting with a bash script that generates a module can be enough.

I’d suggest to try aoc right now, see the Are you trying this library before December 1st ? section so you know if it is useful to you before day 1.

This is my bash command to work with my package that I run every morning:

play:
  mix aoc.create | rg '\.ex' | rg -v Compiling | xargs code
  mix format
  nohup firefox $(mix aoc.url | rg "https") >/dev/null 2>&1 &
  echo 'Good Luck :)'

Here’s a similar thread, which also contains my opinion

I have one repo for alllllll of my puzzle solutions that I keep building in every year - GitHub - sevenseacat/advent_of_code: My Elixir solutions to Advent of Code (spoilers: includes solutions for 457 stars and counting)

My daily process is pretty simple -

• I run mix day <year> <day> eg. mix day 2025 1 to generate a skeleton module/test module for the day’s solution (which will be in lib/y2025/day01.ex and test/y2025/day01_test.exs)
• I manually download the puzzle input for that day and put it in lib/y2025/input/day01.txt
• And away I go! Now I can run code like Y2025.Day01.part1() in iex to test out my part 1 solution, etc.

I keep it super-minimal; each day is a subdirectory with part1.exs and part2.exs files (and inputs) and I run code with elixir part1.exs etc

Usually that file defines a module, then has top-level code afterwards that uses the module.

Some of the problems can be solved with literally just a chain of pipes, no module required.