Repository

A repository is the storage location for all files of a software project. You can think of it as a project folder with an especially good memory. That’s because it doesn’t just store today’s version of the files, but every single change that was ever made. Each change also comes with a note: who made it, when, and why. That’s why anyone can look up at any time what a file looked like three months ago. In short, a repository is usually just called a “repo”.

Why teams can’t work without repos

Larger programs are rarely worked on by individuals alone. For an operating system or a banking app, it’s hundreds or thousands of people. Everyone is simultaneously changing files that depend on one another. Without a shared, organized storage location, this would immediately end in chaos. The repository is the authoritative truth about what the project currently looks like.

The second major advantage is the ability to go back in time. Almost every new version of a program brings bugs with it, some of them severe. Because every previous state is saved, a team can simply jump back. The question “Since when has this bug actually existed?” can also be answered by searching through the history. In companies, this log additionally serves as proof of who approved which change.

For AI models, repositories have become doubly important. On one hand, the code of the models themselves lives there. On the other hand, public repositories are one of the most important sources from which language models learned to program. Billions of lines of openly accessible code are contained in their training data.

Commits, branches, and the way back

The technology behind a repository is called version control. By far the most widespread program for this is Git. Whoever has changed something turns it into a commit: a completed package of changes with a comment. These commits line up into a chain that forms the entire project history. Practically nothing is ever deleted in the process; only new things are added.

So that people don’t get in each other’s way, there are branches. A branch is an offshoot on which someone can experiment undisturbed. If the result works, the branch is merged back with the main version. If two people edit the same line differently, Git reports a conflict and demands a decision.

Usually a copy of the repository sits on a server on the internet, and every participant has a complete copy on their own computer. You pull down other people’s changes and push up your own. A common misconception is that a repository is just storage space like a cloud drive. The difference is the history: a cloud mainly knows the newest file, a repo knows every intermediate step.

GitHub, open source, and models to download

The best-known provider for repositories is GitHub, which belongs to Microsoft and has over 100 million user accounts. Competitors are GitLab and Bitbucket. When tech news reports that a company has “released” a piece of software, it almost always means: the repository is now publicly viewable. This is how open source works – software whose source code anyone may read and reuse.

In the AI world, the platform Hugging Face plays this role as well. It hosts repositories with the finished model files, for instance from Meta's Llama or from Mistral. Anyone who wants to run such a model themselves downloads it from there. Programming aids like GitHub Copilot also tap directly into the repository: they read the project’s existing code in order to make fitting suggestions.

Repositories therefore also come up in disputes. Whether models were allowed to be trained on public code is occupying the courts. And students who program often put their projects on GitHub. This public repo serves as a work sample when applying for jobs in the software industry.