LoonFS is a fast, scalable filesystem built on object storage. Durable and embeddable by design.
Use LoonFS as a personal workspace, share it with your team, or embed it within your platform.
Features you already know and that "just work" with your evolving workflows and tools.
Versions, rollbacks, forks, and preconditions — work faster, not looser.
LoonFS is built around a correctness-first protocol.
Read the docs →Atomic. A multi-file write commits as one unit — it either lands completely or fails safely and leaves the namespace in a valid state.
Isolated. Every query is resolved against an authoritative snapshot; another agent's in-flight writes stay invisible until they commit.
Consistent. Every read resolves against one committed revision, so a path never returns state between successful transactions.
Durable. Content, metadata history, manifests, checkpoints, and control state are stored as LoonFS objects in durable object storage.
Throughput measured against each provider's API on the same benchmark suite. Higher is better.
Content and metadata are durably committed directly to object storage, and reads are accelerated using a cache. LoonFS can be run on a server or embedded in-process with a single dependency: an S3-compatible store.
Read the docs →LoonFS is designed to be used and navigated like a native filesystem. All filesystem commands available by default in agent-friendly interfaces.
Typed clients for TypeScript and Python. The same namespaces, history, and sharing you get from the CLI: a few lines of code, ready to embed in your app or your agent.
Use LoonFS in your environment.
Works with most S3-compatible stores including S3, GCS, and R2.