PyPI

Python is a widely used programming language, used among other things to build websites and AI systems. Those who program with it rarely write everything themselves. For tasks like drawing charts, analyzing tables, or recognizing images, there are ready-made building blocks that other people have already written. Such building blocks are called packages, and PyPI is the large public repository where they are stored. The abbreviation stands for Python Package Index, roughly meaning “directory of Python packages.” You can think of PyPI as an app store — only not for phone apps, but for code building blocks, and free of charge.

Why almost every Python project depends on it

Without PyPI, every team would have to program basic functions over and over again. That would be slow and error-prone. Instead, millions of projects rely on the same tested building blocks. PyPI now hosts over 600,000 different packages. Together, they are downloaded many billions of times per month.

This is especially significant for the AI industry. Tools like PyTorch, TensorFlow, or Transformers, which are used to build and run AI models, distribute their software via PyPI. When a company releases a new model, a matching package usually appears there shortly afterward. The path from announcement to the first test on one’s own computer therefore often takes only minutes.

This convenience has a downside: dependency. If PyPI goes down or an important package is removed, build processes come to a halt worldwide. This has happened several times. That’s why large companies operate their own copies of the repository, known as mirrors.

From upload to install command

Anyone who wants to publish a package packs their code together with a description file into an archive. This contains the name, the version number, and the other packages required. This archive is then uploaded to PyPI. From that moment on, it is visible and retrievable by everyone.

On the user side, a helper program called pip does the work. The command “pip install requests” fetches the package named requests from PyPI and places it on the user’s own computer. If this package needs further packages, pip automatically downloads them too. This chain is called dependencies, and it can easily comprise dozens of packages.

An important distinction: PyPI does not check the content of uploaded code. It is a repository, not a quality control system. Anyone with a free account may publish. Attackers exploit this and upload malware under names that look confusingly similar to well-known packages. A typo during installation can thus cause real damage.

PyPI in the news and in practice

In tech news, PyPI usually comes up in two contexts. The first is security incidents: researchers regularly discover malicious packages that steal passwords or crypto wallets. The operators then delete such packages, but often only after hundreds of downloads. The second context is outages and new rules, such as the requirement for two-factor authentication for authors of important packages.

In everyday life, one rarely encounters PyPI directly, but constantly benefits from it. Almost every data analysis, every AI prototype, and many web services largely consist of packages from there. Anyone starting out with Python types the pip command within the first few hours.

Comparable repositories exist for other languages: npm for JavaScript, Maven Central for Java, crates.io for Rust. They work on the same principle and share the same strengths and weaknesses. Together, they form what experts call the software supply chain — a web of dependencies that hardly anyone fully oversees.