Is it possible to use JGit's in-memory repository to add one small file to a big Git repository without checking it out locally?

I have a large (several GB) Git repository. I want some application to create small files within that repository and commit the changes. This should happen without checking out those gigabytes to the disk.

I found a JGit code sample in which a remote repository is cloned into an in-memory respository.

Can I use JGit (something like shown below) in order to add a file to a remote repository without checking it out locally (i. e. without transferring gigabytes of data to the machine where that code will run)?

DfsRepositoryDescription repoDesc = new DfsRepositoryDescription();
InMemoryRepository repo = new InMemoryRepository(repoDesc);
Git git = new Git(repo);
git.remoteAdd();
git.commit();

Update 1: The size of the entire directory (tracked files plus .git) is 1.2G. The size of .git alone is 573M.

% du -hs .
1.2G	.
% du -hs .git
573M	.git

Note that a checkout in RAM will download the exact same data from the remote. It will just use 1,2G in RAM, rather than on disk, after the clone is completed.

You may also want to see how you can create and add a "file" using JGit api, to see if it is a convenient way to add the data you want to your repo.

If you want to limit the amount of stuff that's downloaded from your git server : you can create a shallow clone.
see this SO answer for example :

git clone --depth 1 <repo_url> -b <branch_name>

This will download the inner files for the latest commit (a compressed version of the 500M in your case), and run an actual checkout (the 500M) of this commit.

You can run this either on disk, or in RAM, depending on your needs.

In your situation, this would still require something in the 600-700M usage, in RAM or on disk, including the checked out files,
but it would download only ~100M (I'm guessing : check the actual size by running the shallow clone command from your workstation) from the server for the compressed version of the head commit.