Keep your Git. Version the rest.
Your source is fine in Git. It’s the builds, binaries, model weights, datasets and shipped assets that break it. Offload them to Swarmfile over Git-LFS today, then branch, review, roll back, and stream them like code, on a drive that just mounts.
Git ends where your big files begin.
Git-LFS is a bolt-on, metered.
Large files were never Git’s job, so LFS staples them on, and your code host meters its storage and bandwidth, while every checkout still downloads the files it needs. The moment a repo passes a few gigabytes, the model creaks.
Datasets and builds don't belong in the repo.
Model weights, training sets, container images, release artifacts, captured media: drop them in Git and clone times explode; keep them out and they lose all history and review. Neither is version control.
Code and its shipped assets live apart.
The 3D, audio, and video assets that ship inside your app sit on a NAS or a Drive folder, disconnected from the code’s branches and history. Two sources of truth, one of them un-versioned.
Swarmfile picks up exactly there: the gigabytes, versioned.
Built for how software teams actually ship.
Bring the Git you already use.
Point a repo’s .lfsconfig at Swarmfile and offload the big files today. We’re a Git-LFS server, so your source stays exactly where it is and only the gigabytes route to us. No server to stand up, no workflow to relearn. Stock git-LFS pushes objects up to ~100 MB; add our small transfer agent for chunked, resumable pushes up to 256 GiB. Walk away whenever you want: git lfs fetch --all pulls everything back with stock git-LFS, no Swarmfile software required. See the on-ramp.
Every file is git-lfs-pullable, with zero duplicate storage.
A file written to the mounted drive is addressable by its LFS object id, so git lfs pull fetches the exact same bytes the drive already holds. And the reverse works too: an object you git lfs push can be projected onto the drive as a first-class file. One copy, both directions.
git clone it, too.
Turn on git view for a project and anyone with access can git clone swarmfile://<org>/<project>: the remote helper ships in the desktop installers, so there’s nothing else to install. Clone and fetch work today, shallow clones included; push isn’t built yet. Git & Swarmfile · How hosted LFS compares
Version datasets and builds like code.
Branch a training set, review a change to a multi-gigabyte artifact, and roll back a bad build. It’s the version-control workflow you have for source, extended to everything Git was never meant to hold.
Stream it. Don't clone it.
Open a 400 GB dataset without downloading it, and hand a build agent, a CI job, or an AI sandbox a scoped slice of a huge project instantly, with no full clone and no wait. Running a coding-agent fleet? One mounted workspace per agent, one shared cache.
Encrypted, and exact about where.
On paid plans, files you save through the drive are encrypted with your project’s key before they leave your machine; turn on the opt-in end-to-end tier and not even we can read them. Two exceptions, stated plainly: Free and public projects are stored unencrypted, and git-LFS objects uploaded directly by the stock client (or 64 MiB+ through our LFS agent) are stored without application-layer encryption, so git-LFS isn’t available on end-to-end projects (see the details).
Public projects and releases, today.
Publish a project and anyone can browse and stream it without an account; tag releases with provenance, let others fork a release into their own org, and run a headless runner on every commit or tag. Not built yet: git push, issues, and stars. So today Swarmfile is the home for the heavy half your code host can’t hold.
Prefer to start from the terminal? The developer quickstart has the real commands for Git-LFS, mounted workspaces, and CI.
Keep your workflow. Lose the file limits.
Trial Swarmfile on one repo’s big files with a 7-day Pro trial. Card on file, nothing charged until day 7. A trial covers up to 2 seats; add more once it converts. Bring your own storage or use ours.
Cancel anytime before day 7.