run with this

pip install rwt==4.4.3


tests Ruff Code style: Black

pip-run provides on-demand temporary package installation for a single execution run.

It replaces this series of commands (or their Windows equivalent):

$ virtualenv --python pythonX.X --system-site-packages $temp/env
$ $temp/env/bin/pip install pkg1 pkg2 -r reqs.txt
$ $temp/env/bin/python ...
$ rm -rf $temp/env

With this single-line command:

$ py -X.X -m pip-run pkg1 pkg2 -r reqs.txt -- ...

Features include

  • Downloads missing dependencies and makes their packages available for import.
  • Installs packages to a special staging location such that they're not installed after the process exits.
  • Relies on pip to cache downloads of such packages for reuse.
  • Leaves no trace of its invocation (except files in pip's cache).
  • Supersedes installed packages when required.
  • Relies on packages already satisfied [1].
  • Re-uses the pip tool chain for package installation.

pip-run is not intended to solve production dependency management, but does aim to address the other, one-off scenarios around dependency management:

  • trials and experiments
  • build setup
  • test runners
  • just in time script running
  • interactive development
  • bug triage

pip-run is a compliment to Pip and Virtualenv, intended to more readily address the on-demand needs.

[1] Except when a requirements file is used.


pip-run is meant to be installed in the system site packages alongside pip, though it can also be installed in a virtualenv.


  • as script launcher
  • as runtime dependency context manager
  • as interactive interpreter in dependency context
  • as module launcher (akin to python -m)
  • as a shell shebang (#!/usr/bin/env pip-run), to create single-file Python tools

Invoke pip-run from the command-line using the console entry script (simply pip-run) or using the module executable ( python -m pip-run). This latter usage is particularly convenient for testing a command across various Python versions.

Parameters following pip-run are passed directly to pip install, so pip-run numpy will install numpy (reporting any work done during the install) and pip-run -v -r requirements.txt will verbosely install all the requirements listed in a file called requirements.txt (quiet is the default). Any environment variables honored by pip are also honored.

Following the parameters to pip install, one may optionally include a -- after which any parameters will be executed by a Python interpreter in the context or directly if prefixed by !.

See pip-run --help for more details.


The examples folder in this project includes some examples demonstrating the power and usefulness of the project. Read the docs on those examples for instructions.

Module Script Runner

Perhaps the most powerful usage of pip-run is its ability to invoke executable modules and packages via runpy (aka python -m):

$ pip-run cowsay -- -m cowsay "moove over, pip-run"

< moove over, pip-run >
   \   ^__^
    \  (oo)\_______
       (__)\       )\/\
           ||----w |
           ||     ||

cowsay example animation

Module Executable Runner

Some package tools, like ranger, are invoked with a unique executable instead of a module. pip-run can run an executable from a package if it is prependend by a !:

$ pip-run ranger-fm -- '!ranger'

Command Runner

Note that everything after the -- is passed to the python invocation, so it's possible to have a one-liner that runs under a dependency context:

$ python -m pip-run requests -- -c "import requests; print(requests.get('').status_code)"

As long as pip-run is installed in each of Python environments on the system, this command can be readily repeated on the other python environments by specifying the relevant interpreter:

$ python3.7 -m pip-run ...

or on Windows or Unix with [pylauncher](

$ py -3.7 -m pip-run ...

Script Runner

pip-run can run a Python file with indicated dependencies. Because arguments after -- are passed directly to the Python interpreter and because the Python interpreter will run any script, invoking a script with dependencies is easy. Consider this script "":

#!/usr/bin/env python

import requests

req = requests.get('')

To invoke it while making sure requests is present:

$ pip-run requests --

pip-run will make sure that requests is installed then invoke the script in a Python interpreter configured with requests and its dependencies.

For added convenience when running scripts, pip-run will infer the beginning of Python parameters if it encounters a filename of a Python script that exists, allowing for omission of the -- for script invocation:

$ pip-run requests

Script-declared Dependencies

Building on Script Runner above, pip-run also allows dependencies to be declared in the script itself so that the user need not specify them at each invocation.

To declare dependencies in a script, add a __requires__ variable or # Requirements: section to the script:

#!/usr/bin/env python

__requires__ = ['requests']

# or

# Requirements:
# requests

import requests

req = requests.get('')

With that declaration in place, one can now invoke pip-run without declaring any parameters to pip:

$ pip-run

The format for requirements must follow PEP 508.

Single-script Tools and Shebang Support

Combined with in-script dependencies, pip-run can be used as a shebang to create fully self-contained scripts that install and run their own dependencies, as long as pip-run is installed on the system PATH. Consider, for example, the pydragon script:

#!/usr/bin/env pip-run
__requires__ = ['requests', 'beautifulsoup4', 'cowsay']
import requests
from bs4 import BeautifulSoup as BS
import cowsay
res = requests.get('')
b = BS(res.text, 'html.parser')
cowsay.dragon(b.find("div", class_="introduction").get_text())

This executable script is available in the repo as examples/pydragon (for Unix) and examples/ (for Windows [2]). Executing this script is equivalent to executing pip-run pydragon.

By default, the script will assemble the dependencies on each invocation, which may be inconvenient for a script. See Environment Persistence for a technique to persist the assembled dependencies across invocations. One may inject PIP_RUN_MODE=persist in the shebang, but be aware that doing so breaks Windows portability.

[2] .PY must exist in the PATHEXT for Python scripts to be executable. See this documentation for more background.

Other Script Directives

pip-run also recognizes a global __index_url__ attribute. If present, this value will supply --index-url to pip with the attribute value, allowing a script to specify a custom package index:

#!/usr/bin/env python

__requires__ = ['my_private_package']
__index_url__ = 'https://my.private.index/'

import my_private_package

Extracting Requirements

After having used pip-run to run scripts, it may be desirable to extract the requirements from the __requires__ variable or # Requirements: section of a script to install those more permanently. pip-run provides a routine to facilitate this case:

$ py -m examples/pydragon
requests beautifulsoup4 cowsay

On Unix, it is possible to pipe this result directly to pip:

$ pip install $(py -m examples/pydragon)

To generate a requirements.txt file, specify a newline separator:

$ py -m --separator newline examples/pydragon > requirements.txt

And since pipenv uses the same syntax, the same technique works for pipenv:

$ pipenv install $(python -m

Interactive Interpreter

pip-run also offers a painless way to run a Python interactive interpreter in the context of certain dependencies:

$ /clean-install/python -m pip-run boto
>>> import boto

Experiments and Testing

Because pip-run provides a single-command invocation, it is great for experiments and rapid testing of various package specifications.

Consider a scenario in which one wishes to create an environment where two different versions of the same package are installed, such as to replicate a broken real-world environment. Stack two invocations of pip-run to get two different versions installed:

$ pip-run keyring==21.8.0 -- -m pip-run keyring==22.0.0 -- -c "import importlib.metadata, pprint; pprint.pprint([dist._path for dist in importlib.metadata.distributions() if dist.metadata['name'] == 'keyring'])"

IPython Inference

If IPython is specified as one of the dependencies, the Python interpreter will be launched via IPython (using -m IPython) for interactive mode. This behaviour may be toggled off by setting the environment variable PIP_RUN_IPYTHON_MODE=ignore.

How Does It Work

pip-run effectively does the following:

  • pip install -t $TMPDIR
  • cleanup

For specifics, see

Environment Persistence

pip-run honors the PIP_RUN_RETENTION_STRATEGY variable. If unset or set to destroy, dependencies are installed to a temporary directory on each invocation (and deleted after). Setting this variable to persist will instead create or re-use a directory in the user's cache, only installing the dependencies if the directory doesn't already exist. A separate cache is maintained for each combination of requirements specified.

persist strategy can greatly improve startup performance at the expense of staleness and accumulated cruft.

Without PIP_RUN_RETENTION_STRATEGY=persist (or with =destroy), pip-run will re-install dependencies every time a script runs, silently adding to the startup time while dependencies are installed into an ephemeral environment, depending on how many dependencies there are and whether the dependencies have been previously downloaded to the local pip cache. Use pip-run -v ... to see the installation activity.

The location of the cache can be revealed with this command:

py -c 'import importlib; print(importlib.import_module("pip_run.retention.persist").paths.user_cache_path)'


  • Due to limitations with pip, pip-run cannot run with "editable" (-e) requirements.
  • pip-run uses a sitecustomize module to ensure that .pth files in the requirements are installed. As a result, any environment that has a sitecustomize module will find that module masked when running under pip-run.

Comparison with pipx

The pipx project is another mature project with similar goals. Both projects expose a project and its dependencies in ephemeral environments. The main difference is pipx primarily exposes Python binaries (console scripts) from those environments whereas pip-run exposes a Python context (including runpy scripts).

Feature pip-run pipx
user-mode operation
invoke console scripts
invoke runpy modules  
run standalone scripts  
interactive interpreter with deps  
re-use existing environment  
ephemeral environments
persistent environments
PEP 582 support  
Specify optional dependencies  
Python 2 support  

Comparison with virtualenvwrapper mktmpenv

The virtualenvwrapper project attempts to address some of the use-cases that pip-run solves, especially with the mktmpenv command, which destroys the virtualenv after deactivation. The main difference is that pip-run is transient only for the invocation of a single command, while mktmpenv lasts for a session.

Feature pip-run mktmpenv
create temporary package environment
re-usable across python invocations
one-line invocation  
multiple interpreters in session  
run standalone scripts
interactive interpreter with deps
re-use existing environment  
ephemeral environments
persistent environments


The author created this package with the intention of demonstrating the capability before integrating it directly with pip in a command such as pip run. After proposing the change, the idea was largely rejected in pip 3971.

If you would like to see this functionality made available in pip, please upvote or comment in that ticket.


pip-run uses semver, so you can use this library with confidence about the stability of the interface, even during periods of great flux.


Invoke tests with tox.

For Enterprise

Available as part of the Tidelift Subscription.

This project and the maintainers of thousands of other packages are working with Tidelift to deliver one enterprise subscription that covers all of the open source you use.

Learn more.

Security Contact

To report a security vulnerability, please use the Tidelift security contact. Tidelift will coordinate the fix and disclosure.