All you need to do is include a `[tool.pixi.workspace]` table in your `pyproject.toml` and you can use `pixi`.

The only time I fall back to using `uv` is for thirdparty libraries that don't have a `[tool.pixi.workspace]` defined.

To use `pixi` you do need to add (a tiny amount of) extra config.

I'm pretty sure that to solve an environment with wheels requires actually downloading the wheels as not enough metadata is available from PyPI. The packaging experts may have to weigh in on the correctness of that though.

You touched on the "contains all compiled dependecies" in the wheel package which increases the environment size whcih is an important consideration for storage costs, and container pull latency, but they also duplicate all compiled symbols *in memory* so they require more memory to run as well.

And another one on the amazingly powerful `pixi global` feature:
prefix.dev/blog/using-p...

There's an excellent series of blog posts on `conda` packages starting with:
conda.org/blog/conda-i...

I use `pixi` to manage PyPI projects. You don't need to duplicate your dependencies if you're ok with using wheels.

`pixi` will automatically map your existing `[project.dependencies]` array to `[tool.pixi.pypi-dependencies]` and will also map `[optional-dependencies]` to `pixi` features.

The new dashboard feed.