Just letting you know the link on your website to "Muni Town Blog" is linking to muni.town/blog when I think it should link to blog.muni.town.
Cool project!

This video explains them very clearly:
youtu.be/S3VBi6kHw5c?...

Flakes allow you to much more easily compose nix packages from different places. If you see a program on GitHub that has a flake, and you want to use it in my project, it's very easy to do. You can add it as an input to your flake and Nix will handle pinning it to a specific version and such.

I can offer a little insight:

IMO flakes are almost always a better option, just because sometimes you kind or need them, and it's simpler to just have a single approach.
There are much better long form explanations of them but I'll try to give a very simple one here: (see next)

If you were smart you could probably make an MCP server that provides the symbols and types and then allows the LLM to ask for the documentation separately, that way you can get visibility on way more of the API surface in context and only fetch the details of functions it's likely to use.

I'm very familiar with LSP so I'm sure it's possible.

The actual quantity of text should be tiny compared to loading full files with heaps of private implementation details not needed to understand how to use the module.

I wonder if I'm missing something, but I don't really get why LLM editors aren't using LSP output. I'd think providing basic stuff like the types and names for all exported symbols from all files in your codebase, would eliminate a lot of hallucinations/ improve big picture understanding.

Hey, I recently had to delve into the poorly documented world of ppx. I created this: github.com/faldor20/ppx...
It's quite small, has lots of comments and some demos, hopefully it's a helpful example on your journey.

The thing that was most confusing for me was having args for the ppx.