Can others
reproduce your results? Recipes for using uv and pixi
to manage Python dependencies

James Thomas, j.thomas@bristol.ac.uk

2025-09-09

You should try uv and pixi in your projects

• What can uv and pixi do?
• Some common pitfalls when using pip and conda
• How these are addressed by uv and pixi
• Practical steps and recipes for using them in projects

• Take home message is that you should use uv and pixi in your projects
• I’ll argue this by talking about…
• These are my experiences of using the tools, and I’m hoping you can share your own advice!

What can uv and pixi do?

Like venv/pip or conda

• create virtual environments
• install packages
• uv: PyPI
• pixi: conda-forge (or PyPI)

Like poetry, pipenv or pip-tools

• reproducible dependency installations using a lockfile

Like pyenv

• install multiple python versions

Like pipx

• install system-wide tools

And… a shared cache…

• uv and pixi (2023-24) incorporate functionality found in several separate tools
• Some of them, like venv and pip are builtin to python
• Others are more establish tools that you may be using already
• Like the combination of venv and pip, or conda, they can …
• Like more advanced dependency management tools, can … (see an example of a lockfile later)
• Like pyenv, … (so compact, don’t need python, good for bootstrapping)
• Like pipx, …
• Shared cache means new venvs are cheap to create (temporarily add packages, etc.)

Motivations for
using better tools

Why not just use pip?

• pip will allow you to install incompatible dependencies
  if you don’t install them all at the same time

  pip install -r requirements.txt
  pip install matplotlib  # don't do this!

• difficult to keep track of direct dependencies of a project
  (the ones you intended you use, not just all the dependencies)

pip will allow you to install incompatible dependencies
if you don’t install them all at the same time

python -V
ls
cat my_script.py
cat requirements.txt             # project is old
. .venv/bin/activate
pip install -r requirements.txt  # numpy 1.21.6, scipy 1.7.3
python my_script.py
pip install matplotlib           # note warning and numpy upgrade
python my_script.py              # broken

Why not just use conda?

• conda has a faster solver to determine version compatibility (libmamba)

but

• developing across different platforms is difficult
• still difficult to keep track of direct dependencies
• not everyone wants the conda ecosystem

developing across different platforms is difficult

ls
cat my_script.py
cat environment.yml             # project is old
conda env create
conda activate my_env
python -V
python my_script.py
conda export > environment.yml  # at some point will need to do this
cat environment.yml             # note the build specifications
git pull
conda env update --prune        # broken
cat environment.yml

How do uv and pixi
make your life easier?

Package installation, environment management

	uv	pixi
Configuration and lockfile	pyproject.toml uv.lock	pixi.toml or pyproject.toml pixi.lock

• Splits intent from actual
• May have seen in other tools (or done manually)

Package installation, environment management

	uv	pixi
Configuration and lockfile	pyproject.toml uv.lock	pixi.toml or pyproject.toml pixi.lock
Init a project	uv init --python 3.x	pixi init
Add a package	uv add pandas	pixi add python=3.x pandas
Run python a script	uv run python uv run my_script.py	pixi run python pixi run python my_script.py

• Init, add, run pattern
• No activating needed
• In conda, python is a package
• With uv, can skip python

Package installation, environment management

	uv	pixi
Configuration and lockfile	pyproject.toml uv.lock	pixi.toml or pyproject.toml pixi.lock
Init a project	uv init --python 3.x	pixi init
Add a package	uv add pandas	pixi add python=3.x pandas
Run python a script	uv run python uv run my_script.py	pixi run python pixi run python my_script.py
List packages	uv pip list uv tree	pixi list pixi tree
Update .lock/env and packages	uv sync uv sync --upgrade	pixi install pixi update

Using uv

Install	Init	Add	Run
docs.astral.sh/uv	uv init --python 3.x	uv add PACKAGE	uv run python uv run my_script.py

python                             # not installed
ls
curl -LsSf https://astral.sh/uv/install.sh | sh
restart
uv init --python 3.10
ls                                 # can remove remove main.py
cat pyproject.toml
cat .python-version
uv run python -V
uv add numpy==1.21.6 scipy==1.7.3
uv run my_script.py                # no activate
ls
cat pyproject.toml
cat uv.lock                        # version control (x3), don't edit uv.lock by hand
uv add matplotlib                  # older (compatible) version of matplotlib
uv run my_script.py

Using pixi

Install	Init	Add	Run
pixi.sh	pixi init	pixi add python pixi add PACKAGE	pixi run python pixi run python my_script.py

python
conda
ls
curl -fsSL https://pixi.sh/install.sh | sh
reset
pixi init
ls
cat pixi.toml
pixi add python=3.10
pixi run python -V
pixi add numpy==1.21.6 scipy==1.7.3
pixi run python my_script.py         # no activate
ls
cat pixi.toml
cat pixi.lock                        # version control (x2), don't edit pixi.lock by hand

Multiple platforms with pixi

pixi workspace platform add ...

pixi workspace
pixi workspace platform add osx-arm64 win-64
cat pixi.toml
cat pixi.lock

Recipes
and quality fo life improvements

Projects in uv

Init project	Project script	Inject packages
uv init --package	uv run test-coin	uv run --with jupyter jupyter-lab

# uv init --package
tree
cat src/my_project/probability.py
cat pyproject.toml
uv run
uv run test-coin
uv run --with jupyter jupyter-lab

In the notebook:

• Importing package
• Autoreloading

Projects in pixi

Init project	Project tasks
pixi init --format pyproject	pixi run test-coin (uses typer) pixi run lab

# pixi init --format project
tree
cat pyproject.toml                 # Talk through
pixi run
pixi run test-coin
cat src/my_project/probability.py
pixi run test-coin --help
pixi run test-coin --heads-observed 4
pixi run lab

System-wide tools

Without installing	Install system-wide
uvx COMMAND	uv tool install PACKAGE
pixi exec COMMAND	pixi global install PACKAGE

uvx cowsay -t Moo
uv tool install cowsay
cowsay -t Moo
# for pixi it's cowpy

Listing packages

List	Tree (why do I have this package?)
uv pip list	uv tree [--package PACKAGE] [--invert]
pixi list	pixi tree [PACKAGE] [--invert]

uv pip list
uv tree --package scipy
uv tree --package numpy --invert

Scripts with inline metadata PEP 723

uv init --script my_script.py
uv add --script my_script.py PACKAGE

cd ../uv_script_demo
ls
# uv init --script my_script.py
# uv add --script my_script.py numpy
cat my_script.py
uv run my_script.py       # 1.22.0 was released on 30 Dec 2021
uv run my_script.py

Wrapping up

• Find out more: 🔗 docs.astral.sh/uv, 🔗 pixi.sh

• Use for new projects

  • uv if PyPI, pixi if Conda
  • package in src/
  • commands uv run analysis, pixi run analysis

• When working with someone
  new to Python

  • uv run ... could be in a script
  • pixi run dashboard
    only command they need?

• Easier than conda/pip by themselves

James Thomas
j.thomas@bristol.ac.uk

jatonline.github.io/managing-
dependencies-using-uv-and-pixi