Instead of Conda: Pixi

Why Pixi is better

• Pixi writes pixi.lock for every platform you target, so an environment recreates exactly. Conda has no lockfile of its own.
• The manifest, lockfile and environment live in the project folder, so sharing a project shares the environment, without a global list of named environments to keep in sync.
• Through uv, one Pixi manifest resolves conda-forge and PyPI packages together, which removes the usual mix of conda install and pip install.
• The default channel is conda-forge, with over 30,000 packages, so existing conda dependencies carry over.
• pixi run replaces the Makefile or shell scripts that conda projects usually grow alongside environment.yml.

When to keep Conda

Keep conda for a long-lived shared base environment on a cluster or teaching machine, where people activate one environment by name. Also keep it where documentation and tooling assume conda activate. Pixi has no base environment by design.

Switching

Pixi can import an existing environment.yml into a workspace and export one back for colleagues still on conda. Commands map one to one: conda install becomes pixi add, conda activate becomes pixi shell, and conda run becomes pixi run. CI needs the Pixi installer instead of Miniconda.

Watch out

Pixi has no stable major release yet and ships new versions often, so expect occasional behaviour changes. It is built by prefix.dev, a small company. If you used Anaconda's defaults channel, check its terms, which Anaconda changed in 2024 to require fees in most commercial cases; Pixi uses conda-forge instead.

Also worth a look

• uv The better pick when a project only needs packages from PyPI and no conda-only binaries.

More newer, better package managers and environment tools

• npm →, use pnpm
• pip →, use uv
• Poetry →, use uv
• pyenv →, use uv