Swarmfile is now a git-LFS server

Most teams that outgrow plain Git don't actually want to leave it. They want somewhere else for the files that don't belong in a diff. A 4K plate, a trained model checkpoint, a texture set. Git-LFS was built for exactly that split: keep source in Git, point large objects somewhere else. The "somewhere else" is usually the part that falls short: bloating your git host's storage bill, capping object size, or requiring a second piece of infrastructure to run and pay for.

Swarmfile can now be that somewhere else.

What you get#

Point git-LFS at a Swarmfile project and your large binary assets get real storage, deduplication, and quota, while your source code stays exactly where it already lives, in the git host your team uses today.

One thing to be precise about is encryption, because it's the one place LFS differs from the mounted drive. Objects that go through Swarmfile's transfer agent in blocks are encrypted at rest like anything else on a private project. Objects uploaded directly by the stock git-LFS client, and uploads of 64 MiB or more through the desktop app's built-in agent, are stored as single objects without application-layer encryption, and git-LFS isn't available on end-to-end-encrypted projects at all. If a binary has to be encrypted at rest, keep it on the drive. The git-LFS guide has the path-by-path table.

A few things you get:

• Pushes up to 256 GiB. Stock git-LFS sends each object in one request, which our network edge caps at ~100 MB. A small transfer agent that speaks the git-LFS custom-transfer protocol pushes larger objects in chunks, up to 256 GiB each, resuming where it left off.
• One lock system for git and the drive. For a file on the mounted drive, git lfs lock and a drive lock are the same lock, so it can't be checked out twice by two different routes. Files that live only in git can be locked too.
• No lock-in. Pull everything back out with stock git-LFS and no Swarmfile software installed, then repoint at any other host. The exit is built in from day one, not bolted on later.
• A first-class view of your objects. Anything you push through LFS can also be projected onto the mounted drive as real files, browsable and previewable the same way as everything else in a project.

Getting started#

If your team already uses git-LFS, the migration is a config change, not a new workflow: point .lfsconfig at your Swarmfile project, authenticate, and push. See Git-LFS for setup, credentials, and how deduplication interacts with a mounted drive if you use both.