Library / SDK
pdm-project/pdm avatar
pdm-project/pdm

PDM runs on twenty-three dependencies and its own build backend

A modern Python package and dependency manager supporting the latest PEP standards

8,665 stars526 forksPythonMIT

At a glance

What is it?
PDM reads PEP 621 metadata from pyproject.toml, manages virtual environments and a pdm.lock file, and builds itself with pdm-backend. Its two experimental switches, PEP 582 and uv, cannot be turned on at the same time.
Who is it for?
PDM suits teams that need a lockfile, several virtual environments, and freedom over the build backend, and it is a workable default for library work where Pipenv stops at packaging. It is the wrong choice if your policy rules out piping a downloaded script into a shell, or if you expect uv and PEP 582 to work together, which the README rules out.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 3 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • pdm-project/pdm8,665 ★ · -4
Star history
Repository2026-09-222026-11-15
pdm-project/pdm8.7k8.7k

-4 Stars in the last 54 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

PDM's own pyproject.toml is built by pdm-backend

PDM is built by the backend it tells you to consider. Its own pyproject.toml puts `pdm-backend` and `pdm-build-locked` in the build requirements and names `pdm.backend` as the build backend, so the project dogfoods the path it recommends. That file marks the version as dynamic, which means no version string is written into pyproject.toml at all: 2.29.3 shipped on 2026-10-19, 2.29.4 on 2026-10-27 and 2.29.5 on 2026-10-30, and those numbers live in release tags instead. Support is declared as `requires-python = ">=3.10"`, and the classifiers stop at Python 3.14, so 3.15 carries no classifier and anything older than 3.10 is out of range by rule rather than by accident. What this file cannot tell you is which backend a project you have not created yet will end up using. The optional dependency groups include `pytest = ["pytest", "pytest-mock"]`, `copier = ["copier>=8.0.0"]` and `cookiecutter`.

Twenty-four runtime dependencies, one of them conditional on Python 3.11

The runtime list has twenty-three entries. Environments come from virtualenv>=20, interpreter discovery from findpython>=0.7.0,<1.0.0a0, and interpreter installation from pbs-installer[download]>=2026.10.25, which is the piece behind PDM's ability to install Pythons built by astral-sh's python-build-standalone. Resolution sits on resolvelib>=1.1, package discovery on unearth>=0.19, dependency logic on dep-logic>=0.5, and the installation step on installer>=1 with build isolation from pyproject-hooks. Around those sit rich>=12.3.0 for terminal output, argcomplete>=3.6.3 for shell completion, platformdirs for paths, tomlkit>=0.11.1,<1 and python-dotenv>=0.15 for configuration, shellingham>=1.3.2 to read the calling shell, filelock>=3.13 for lock file writes, hishel[httpx2]>=1.4.0 as the cache layer, truststore>=0.10.4 with certifi>=2026.10.19 and httpx2[socks]>=2.9.1,<3 for transport, packaging>22.0 for metadata, blinker for signals, and id>=1.5.0. A single entry is conditional: tomli>=1.1.0 applies only when python_version is below 3.11. There is no trimmed mode, which is the practical limit. Floors such as certifi>=2026.10.19 and the double pin httpx2[socks]>=2.9.1,<3 mean a base image carrying older copies has no room for PDM, even though the everyday command surface is a handful of verbs. The findpython ceiling <1.0.0a0 is another hard edge, and it will not move without a change to this file. The opt-in centralized installation cache, described by analogy to pnpm, changes where packages land on disk rather than what the resolver decides.

The recommended install pipes a downloaded script into your shell

PDM needs Python 3.10 or higher. The recommended route on Linux and Mac is one line that fetches a script and runs it in the same breath:

bash
curl -sSL https://pdm-project.org/install.sh | bash

On Windows the equivalent is a PowerShell one-liner:

powershell
powershell -ExecutionPolicy ByPass -c "irm https://pdm-project.org/install.ps1 | iex"

If you would rather not execute a script you just downloaded, the README points at the standalone binary in the release assets, and at an installation section in the documentation for the alternatives it describes as a Python script and package managers without enumerating them. The repository root shows the pieces plainly: install-pdm.sh, install-pdm.ps1, install-pdm.py and install-pdm.py.sha256. A checksum file therefore exists for the Python installer, and neither recommended command shows a verification step. That gap matters on CI images and for anyone whose policy refuses to pipe a download into a shell, because the fallback is then the release asset and the README hands you no checksum to compare it against.

pdm new, pdm add, and a lock file you read by diffs

The quickstart is two commands. Creating a project:

bash
pdm new my-project

You answer the prompts, and a project with a pyproject.toml file is ready to use. Adding dependencies:

bash
pdm add requests flask

Several packages go into one command, and resolution then writes what it settled on into pdm.lock, the file you open to see what is locked for each package. What this flow cannot do is explain a lock file for you. The README documents no command that prints a single package's entry, no update or rollback path for one dependency, and no way to check a lock against an index. The consequence lands on review: a dependency bump shows up as a diff in pdm.lock, and nothing in the README tells a reviewer which package forced a transitive pin to move, so somebody reads the file line by line. The scaffold step has the same blind spot. `pdm new` asks questions and writes pyproject.toml, but the README does not say which build backend the generated file selects, which leaves you to discover that decision at build time rather than at creation time.

PEP 582 and uv are two switches that exclude each other

Two settings under an Experimental heading change how PDM installs anything. The first turns virtual environments off and enables PEP 582:

bash
pdm config python.use_venv False

With that set, packages land in a local project folder instead of a venv, the way npm installs into node_modules. The second hands installation to uv:

bash
pdm config use_uv true

uv is described as a very fast Python package installer written in Rust. The README then draws the line that decides which of the two you can have: uv does not work with PEP 582. Their combination is therefore not a faster install path but an unusable one, and nothing in the README says which path stops applying when both keys are set in the same project. A team that wants uv keeps venvs; a team that wants PEP 582 keeps PDM's own installer and gives up uv. Both settings are also labelled experimental, so a config committed today is not a contract about the next release, and the repository root carries a CHANGELOG.md and a news/ directory where that kind of change would be recorded.

Any build backend, and nobody picks one for you

The README compares PDM with three tools, and the differences are about packaging rather than taste. Pipenv combines pip and venv and installs from a non-standard Pipfile.lock or Pipfile, but it does not handle packages related to packaging your code, so it is useful only for developing non-installable applications, and a library developer ends up reaching for setuptools anyway. Poetry manages environments and dependencies in a similar way, can build .whl files, uploads wheels and source distributions to PyPI, uses the pyproject.toml standard, and has an interface users can customize through a plugin. Hatch manages environments, allows several per project, keeps them in a central location by default but can be configured to place a project's environments in the project root, and manages packages without lockfile support. PDM's own claim is narrower: venvs in both project and centralized locations, metadata read from a standardized pyproject.toml, lockfile support, plugins shareable as distributions, and no limit to a specific build backend. That last point carries a cost. Coming from Poetry you must re-pick a backend and re-check your plugin set; coming from Hatch you gain the lockfile you were missing.

A plugin system with no published compatibility table

Extending PDM happens through plugins, and that surface is the thinnest documented part of the project. Plugins can be shared by uploading them as distributions, and Awesome PDM is a curated list of plugins and resources, yet the README publishes no compatibility table, plugin API version or minimum PDM release for any plugin. Adding a plugin therefore adds an unpinned piece of your build toolchain. The release rhythm makes that concrete rather than theoretical, with 2.29.3 on 2026-10-19, 2.29.4 on 2026-10-27 and 2.29.5 on 2026-10-30, three releases inside two weeks, and a last push to the default branch on 2026-11-02. A plugin chosen against 2.29.3 can meet 2.29.5 with nothing in its own notes saying so. The opt-in centralized cache is the same kind of dependency: a config key changes what lands on disk rather than a versioned contract promising it will keep working. The repository is set up for contributors, with AGENTS.md, CONTRIBUTING.md, SECURITY.md, .pre-commit-config.yaml, .pre-commit-hooks.yaml, tox.ini, typings/ and docs/ at the root, but none of that tells a plugin author which release to target.

Editorial conclusion

PDM suits teams that need a lockfile, several virtual environments, and freedom over the build backend, and it is a workable default for library work where Pipenv stops at packaging. It is the wrong choice if your policy rules out piping a downloaded script into a shell, or if you expect uv and PEP 582 to work together, which the README rules out. Before committing, check the 3.10 floor, try each plugin against the exact PDM version you pin, and read the pdm.lock diff yourself rather than trusting the resolution.

Frequently asked questions

What does PDM stand for?

The repository never spells out an expansion of the acronym. It describes PDM as a modern Python package and dependency manager supporting the latest PEP standards, written in Python, published under the MIT license, and developed at pdm-project.org.

What does PDM stand for in project management?

Nothing in this repository covers the project management sense of PDM. README.md and pyproject.toml both concern Python packaging, and every documentation link declared in pyproject.toml points at pdm-project.org.

What are PDM and PLM?

PLM is not defined in README.md or in pyproject.toml, and none of the twenty-three runtime dependencies refers to it. This repository describes only a Python package and dependency manager, so a reader looking for product data management terminology finds nothing here.

What is the PDM method in project management?

No management method is covered. What this repository documents instead is a two-command workflow: pdm new my-project scaffolds a project with a pyproject.toml file, and pdm add requests flask records dependencies that resolution writes into pdm.lock.

Official sources

  1. License: MIT
  2. pdm-project/pdm on GitHub
  3. Project website
  4. README
  5. Releases
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/pdm-project-pdm.svg)](https://hysenlabs.com/projects/pdm-project-pdm)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/pdm-project-pdm.svg)](https://hysenlabs.com/projects/pdm-project-pdm)

Was this useful?