Installing packages from PyPI with conda#

Note

The conda-pypi channel is free to use for all users. This channel is not subject to the licensing requirements or payment obligations described in Section 1 of the Anaconda Terms of Service.

In conda version 26.9 and later, you can use the conda-pypi channel to install supported pure Python wheels from the Python Package Index (PyPI) alongside conda packages directly with conda install.

What does the conda-pypi channel do?

The conda-pypi channel, a channel maintained by Anaconda and hosted for free on anaconda.org, indexes all pure Python wheels from the public PyPI index. The channel repodata is generated from package metadata obtained from PyPI and converted into a format conda understands, so the reliability and security of packages available through conda-pypi matches that of PyPI directly.

When you add this channel to your conda configuration, conda can resolve and install those packages alongside packages from conda channels like conda-forge or defaults in a single operation. The channel repodata tells conda what packages are available and at what versions. When you install a package, conda fetches the wheel directly from PyPI at install time. This is by design and keeps the channel lightweight, but it means you need a PyPI connection at install time.

One important thing to note is that this is not a pip replacement. Conda still manages the environment, the solver still handles the full dependency graph, and the result is a reproducible, exportable conda environment. Using pip install directly inside conda environments can work initially, but conda loses track of pip-installed packages, making future environment updates and installs unreliable. For more information on using pip with conda, see our blog Conda and pip are two ecosystems, not just tools.

Prerequisites#

The conda-pypi workflow requires the following:

  • conda 26.9 or later:

    Check your version:

    conda --version

    If necessary, update conda:

    conda self update

  • Rattler solver:

    Set the rattler solver as your default solver:

    conda config --set solver rattler

Setting up the conda-pypi channel#

To make supported packages from PyPI available to conda, add the conda-pypi channel to your conda configuration (.condarc) file. Append it after your existing channels to give it a lower channel priority:

conda config --append channels conda-pypi

Your channel order in ~/.condarc should look similar to this:

channels:
  - conda-forge
  - conda-pypi
Channel priority and ABI compatibility

Normally, when mixing channels, strict channel priority is recommended to avoid ABI incompatibilities between compiled packages built in different environments. The conda-pypi channel contains only pure Python wheels, so there are no compiled binaries and, therefore, no ABI risk.

Flexible priority (the default) also works correctly when pairing conda-pypi with conda-forge or defaults. For more on channel priority, see Channel configuration best practices.

Install packages#

Before installing, you can check what is available in the conda-pypi channel using conda search. Once the channel is configured, conda install resolves the full dependency graph across all configured channels, so packages from PyPI and conda channels can be installed together in a single command.

Find available packages#

To check whether a package is available through conda-pypi:

conda search conda-pypi::<package-name>

Install packages from PyPI and conda channels together#

conda install <conda-forge-package> <conda-pypi-package>

When a package is available from both conda-forge and conda-pypi, conda selects a build according to your channel priority setting and the environment's dependency requirements.

You can also use conda create with packages from PyPI:

conda create --name myenv python=3.12 <conda-forge-package> <conda-pypi-package>

Install with extras support#

Extras are named groups of optional dependencies. When you request an extra, its dependencies are resolved together with the package's regular dependencies. For full details, see CEP 44.

Use the following syntax to request extras:

# Single extra
conda install 'package[extras="EXTRA"]'

# Multiple extras (comma-separated string)
conda install 'package[extras="EXTRA1,EXTRA2"]'

# Multiple extras (list syntax)
conda install 'package[extras=["EXTRA1","EXTRA2"]]'

For example, to install httpx with its http2 and cli extras:

conda install 'httpx[extras="http2,cli"]'

Note

Extra names are case-sensitive and must match the pattern [a-z0-9_.+-]{1,64}. Requesting an extra that is not defined in a package has no effect and doesn't produce an error.

Disable the conda-pypi workflow#

If you want to stop using the conda-pypi channel, remove it from your configuration:

conda config --remove channels conda-pypi

Suppress the conda-pypi suggestion tip#

To suppress the suggestion about the conda-pypi channel shown when conda newly installs pip into an environment, set the following configuration option:

conda config --set plugins.conda_pypi_pip_warning false

Limitations and security considerations#

The security posture of this workflow is equivalent to using pip or other PyPA tools to install from the public PyPI index directly. Packages are fetched from PyPI at install time. Conda does not re-sign or additionally vet the wheels. Keep the following limitations in mind:

  • Public PyPI only. The conda-pypi channel indexes packages from PyPI. It does not include packages from private or alternative package indexes (such as a corporate artifact repository).

  • No additional vetting beyond PyPI. The conda-pypi channel does not perform additional security scanning beyond what PyPI provides. Using the --exclude-newer flag is recommended as a lightweight mitigation for supply-chain risk.

  • conda client only. Supported in the conda CLI. Support in other clients (mamba, micromamba, and so on) is under active CEP discussion.

  • Pure Python wheels only. Packages with compiled extensions must come from conda channels.

  • rattler solver required. The classic conda solver is not supported for this workflow.

  • PyPI connection required at install time. The channel stores metadata only. Wheels are fetched from PyPI during installation.

Troubleshooting#

Packages not found in conda-pypi#

If conda install or conda create fails with a PackagesNotFoundInChannelsError for a package you expect to be available through conda-pypi, the most likely cause is an outdated version of conda or the rattler solver.

For example:

PackagesNotFoundInChannelsError: The following packages are not available
from current channels:
  - flask

Current channels:
  - defaults
  - https://conda.anaconda.org/conda-pypi

To diagnose, check your conda and rattler-solver versions:

conda --version

conda list | grep rattler

If either is below the minimum required version (26.9 for conda and 0.1.1 for conda-rattler-solver), update them:

conda self update

conda install --name base conda-rattler-solver

Additional resources#

On this page
Edit on GitHub
Show Source