SE4Sci

Pip / uv and PyPI

• PyPI hosts packages for pip / uv
• pip: the canonical package manager for Python
• uv: a Rust rewrite of pip, 10-100x faster
  • About 34% of PyPI downloads come from uv (Feb 2026)!
  • Better defaults: uses .venv, never installs to system/user without a flag
  • Also has a high-level API

Conda/Mamba/MicroMamba/Pixi and conda-forge

• Conda: multipurpose packager with Python focus
  • Written in Python
  • Now contains the libmamba solver (in C++)
• mamba: a faster conda written in C++
• micromamba: Used to be different, now (2.0) a statically linked mamba
• pixi: a Rust rewrite of conda, with a new high-level API
  • No low-level API to match conda/mamba

Which to use?

• PyPI is the official ecosystem, most packages are there
• Package owners control PyPI, while conda-forge is community driven
• If you develop a package, PyPI is first
• Conda tools can use PyPI packages
• Conda is often preferred in some fields (like ML); use pixi there

So we'll focus on PyPI via uv.

The course website is a pixi project, by the way.

Getting started with uv

There are lots of ways to install uv. It's just a single Rust compiled binary.

https://docs.astral.sh/uv/getting-started/installation/

curl -LsSf https://astral.sh/uv/install.sh | sh
wget -qO- https://astral.sh/uv/install.sh | sh
pipx install uv
pip install uv
cargo install --git https://github.com/astral-sh/uv uv
brew install uv
# Windows:
winget install --id=astral-sh.uv -e
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Using uv

Low level API:

uv pip <...>
uv venv <...>

High level API:

uv run <command...>

Using packages

You should never just run "pip install <package>"; installing into your global site-packages is a bad idea, and installing into your user site-packages is worse!

Python has a solution for this: virtual environments.

But if you just want to install a command-line tool, you don't have to set them up yourself!

uv tool install <package>
<package> <args...>

Quick run

You can even do it in one line:

uvx <package> <args...>

This will create a temporary virtual environment, install the package, run it. The environment is cached for a time, so reruns are fast.

If the command and package names differ, use --from:

uvx --from <package> <command> <args...>

pipx has the same features: pipx install and pipx run (--spec instead of --from). pipx is preinstalled on GitHub Actions, and there is an action to set up uv.

(I believe this name was inspired by npx from JavaScript, which does the same thing.)

Quick scripts

You can get this functionality for scripts, as well:

# /// script
# dependencies = ["rich"]
# requires-python = ">=3.11"
# ///

import rich

rich.print("[blue]Hello, world!")

uv run myscripts.py <args...>

Virtual environments

uv venv

This creates a .venv folder with a virtual environment in it. You can then activate it:

source .venv/bin/activate  # Linux/macOS
.venv\Scripts\activate  # Windows

Note that uv commands will automatically use a .venv folder if it exists.

Using venvs

When a venv is "active", commands will run in that venv.

Most common command:

uv pip install -e. --group dev

You can always delete the venv and restart, rerunnning uv venv replaces the venv by default.

High level API

uv can help you setup and run a project. The high level API does the things you would normally do manually:

uv init <...> # start a new project
uv add <package> # add a package to the project
uv run <command...> # run a command in the project environment

For an existing project, you just need uv run!

What does uv run do?

• Creates a .venv if not present
• Installs the project's dependencies
• Installs the project in editable mode, if it has a build backend
• Installs the dev dependency group, if it exists
• Uses the lockfile if it exists, otherwise creates one

Works with a standard pyproject.toml; no uv-specific configuration needed.

Pixi

Pixi only has a high-level interface, and it needs a manifest (pixi.toml or [tool.pixi] in pyproject.toml).

• Installs conda and PyPI packages
• Lists platforms and channels in [workspace]
• Has a task system: pixi run <task>

pixi run lab  # starts JupyterLab for this course

Other uv features

See uv --help

• uv python can manage installed Pythons
• uv build can build packages
• uv publish can publish packages
• uv tree can show the dependency tree
• uv self can manage uv itself