Skip to content

Manage Project#

Choose a Python interpreter#

If you have used pdm init, you must have already seen how PDM detects and selects the Python interpreter. After initialized, you can also change the settings by pdm use <python_version_or_path>. The argument can be either a version specifier of any length, or a relative or absolute path to the python interpreter, but remember the Python interpreter must conform with the python_requires constraint in the project file.

How requires-python controls the project#

PDM respects the value of requires-python in the way that it tries to pick package candidates that can work on all python versions that requires-python contains. For example, if requires-python is >=2.7, PDM will try to find the latest version of foo, whose requires-python version range is a superset of >=2.7.

So, make sure you write requires-python properly if you don't want any outdated packages to be locked.

Working with Python < 3.7#

Although PDM run on Python 3.7 and above, you can still have lower Python versions for your working project. But remember, if your project is a library, which needs to be built, published or installed, you make sure the PEP 517 build backend being used supports the lowest Python version you need. For instance, the default backend pdm-pep517 only works on Python 3.7+, so if you run pdm build on a project with Python 3.6, you will get an error. Most modern build backends have dropped the support for Python 3.6 and lower, so it is highly recommended to upgrade the Python version to 3.7+. Here are the supported Python range for some commonly used build backends, we only list those that support PEP 621 since otherwise PDM can't work with them.

Backend Supported Python Support PEP 621
pdm-pep517 >=3.7 Yes
setuptools>=60 >=3.7 Experimental
hatchling >=3.7 Yes
flit-core>=3.4 >=3.6 Yes
flit-core>=3.2,<3.4 >=3.4 Yes

Note that if your project is an application(without name metadata), the above limitation of backends don't apply, since you don't need a build backend afterall, and you can use a Python version up to 2.7.

Build distribution artifacts#

$ pdm build
- Building sdist...
- Built pdm-test-0.0.0.tar.gz
- Building wheel...
- Built pdm_test-0.0.0-py3-none-any.whl

The artifacts will be available at dist/ and able to upload to PyPI.

Configure the project#

PDM's config command works just like git config, except that --list isn't needed to show configurations.

Show the current configurations:

pdm config

Get one single configuration:

pdm config pypi.url

Change a configuration value and store in home configuration:

pdm config pypi.url ""

By default, the configuration are changed globally, if you want to make the config seen by this project only, add a --local flag:

pdm config --local pypi.url ""

Any local configurations will be stored in .pdm.toml under the project root directory.

The configuration files are searched in the following order:

  1. <PROJECT_ROOT>/.pdm.toml - The project configuration
  2. <CONFIG_ROOT>/config.toml - The home configuration

where <CONFIG_ROOT> is:

If -g/--global option is used, the first item will be replaced by <CONFIG_ROOT>/global-project/.pdm.toml.

You can find all available configuration items in Configuration Page.

Publish the project to PyPI#

With PDM, you can build and then upload your project to PyPI in one step.

pdm publish

You can specify which repository you would like to publish:

pdm publish -r pypi

PDM will look for the repository named pypi from the configuration and use the URL for upload. You can also give the URL directly with -r/--repository option:

pdm publish -r

See all supported options by typing pdm publish --help.

Configure the repository secrets for upload#

When using the pdm publish command, it reads the repository secrets from the global config file(<CONFIG_ROOT>/config.toml). The content of the config is as follows:

username = "frostming"
password = "<secret>"

url = ""
username = "frostming"
password = "<secret>"


You don't need to configure the url for pypi and testpypi repositories, they are filled by default values.

To change the repository config from the command line, use the pdm config command:

pdm config repository.pypi.username "__token__"
pdm config repository.pypi.password "my-pypi-token"

pdm config ""

Cache the installation of wheels#

If a package is required by many projects on the system, each project has to keep its own copy. This may become a waste of disk space especially for data science and machine learning libraries.

PDM supports caching the installations of the same wheel by installing it into a centralized package repository and linking to that installation in different projects. To enabled it, run:

pdm config install.cache on

It can be enabled on a project basis, by adding --local option to the command.

The caches are located under $(pdm config cache_dir)/packages. One can view the cache usage by pdm cache info. But be noted the cached installations are managed automatically -- They get deleted when not linked from any projects. Manually deleting the caches from the disk may break some projects on the system.


Only the installation of named requirements resolved from PyPI can be cached.

Show the current Python environment#

$ pdm info
Python Interpreter: D:/Programs/Python/Python38/python.exe (3.8.0)
Project Root:       D:/Workspace/pdm
$ pdm info --env
  "implementation_name": "cpython",
  "implementation_version": "3.8.0",
  "os_name": "nt",
  "platform_machine": "AMD64",
  "platform_release": "10",
  "platform_system": "Windows",
  "platform_version": "10.0.18362",
  "python_full_version": "3.8.0",
  "platform_python_implementaiton": "CPython",
  "python_version": "3.8",
  "sys_platform": "win32"

Manage global project#

Sometimes users may want to keep track of the dependencies of global Python interpreter as well. It is easy to do so with PDM, via -g/--global option which is supported by most subcommands.

If the option is passed, <CONFIG_ROOT>/global-project will be used as the project directory, which is almost the same as normal project except that pyproject.toml will be created automatically for you and it doesn't support build features. The idea is taken from Haskell's stack.

However, unlike stack, by default, PDM won't use global project automatically if a local project is not found. Users should pass -g/--global explicitly to activate it, since it is not very pleasing if packages go to a wrong place. But PDM also leave the decision to users, just set the config global_project.fallback to true.

If you want global project to track another project file other than <CONFIG_ROOT>/global-project, you can provide the project path via -p/--project <path> option.


Be careful with remove and sync --clean commands when global project is used, because it may remove packages installed in your system Python.

Working with a virtualenv#

Although PDM enforces PEP 582 by default, it also allows users to install packages into the virtualenv. It is controlled by the configuration item python.use_venv. When it is set to True, PDM will use the virtualenv if:

  • a virtualenv is already activated.
  • any of venv, .venv, env is a valid virtualenv folder.

Besides, when python.use_venv is on and the interpreter path given is a venv-like path, PDM will reuse that venv directory as well.

For enhanced virtualenv support such as virtualenv management and auto-creation, please go for pdm-venv, which can be installed as a plugin.

Import project metadata from existing project files#

If you are already other package manager tools like Pipenv or Poetry, it is easy to migrate to PDM. PDM provides import command so that you don't have to initialize the project manually, it now supports:

  1. Pipenv's Pipfile
  2. Poetry's section in pyproject.toml
  3. Flit's section in pyproject.toml
  4. requirements.txt format used by pip
  5. setuptools

Also, when you are executing pdm init or pdm install, PDM can auto-detect possible files to import if your PDM project has not been initialized yet.


Converting a will execute the file with the project interpreter. Make sure setuptools is installed with the interpreter and the is trusted.

Export locked packages to alternative formats#

You can also export pdm lock to other formats, to ease the CI flow or image building process. Currently, only requirements.txt and format is supported:

pdm export -o requirements.txt
pdm export -f setuppy -o

Hide the credentials from pyproject.toml#

There are many times when we need to use sensitive information, such as login credentials for the PyPI server and username passwords for VCS repositories. We do not want to expose this information in pyproject.toml and upload it to git.

PDM provides several methods to achieve this:

  1. User can give the auth information with environment variables which are encoded in the URL directly:
url = "http://${INDEX_USER}:${INDEX_PASSWD}"
name = "test"
verify_ssl = false

dependencies = [
  "mypackage @ git+http://${VCS_USER}:${VCS_PASSWD}[email protected]"

Environment variables must be encoded in the form ${ENV_NAME}, other forms are not supported. Besides, only auth part will be expanded.

  1. If the credentials are not provided in the URL and a 401 response is received from the server, PDM will prompt for username and password when -v/--verbose is passed as command line argument, otherwise PDM will fail with an error telling users what happens. Users can then choose to store the credentials in the keyring after a confirmation question.

  2. A VCS repository applies the first method only, and an index server applies both methods.

Manage caches#

PDM provides a convenient command group to manage the cache, there are four kinds of caches:

  1. wheels/ stores the built results of non-wheel distributions and files.
  2. http/ stores the HTTP response content.
  3. metadata/ stores package metadata retrieved by the resolver.
  4. hashes/ stores the file hashes fetched from the package index or calculated locally.
  5. packages/ The centralized repository for installed wheels.

See the current cache usage by typing pdm cache info. Besides, you can use add, remove and list subcommands to manage the cache content. Find the usage by the --help option of each command.

How we make PEP 582 packages available to the Python interpreter#

Thanks to the site packages loading on Python startup. It is possible to patch the sys.path by executing the shipped with PDM. The interpreter can search the directories for the nearest __pypackage__ folder and append it to the sys.path variable.