I agree, it’s even what I mostly use the ORM for, writing is so easy that way and results in far more succinct code.

Data comes out as a map or keyword list, which is then turned into the repository struct in question. If you want raw db data you can get that too. And you can have multiple structs that are backed by the same persistent dataset. It's quite elegant.

Queries themselves are constructed using a language that is near SQL, but far more composable:

Repo.one(from p in Post, join: c in assoc(p, :comments), where: p.id == ^post_id)

Queries themselves are composable

query = from u in User, where: u.age > 18

query = from u in query, select: u.name

And can be written in a keyword style, like the above examples, or a functional style, like the rest of elixir:

User
|> where([u], u.age > 18)
|> select([u], u.name)

None of these "queries" will execute until you tell the Repo to do something. For that, you have commands like Repo.all and Repo.one, the latter of which sticks a limit: 1 on the end of the provided query