# LayerFS

> LayerFS is an open-source (MIT) filesystem for AI coding agents: each tool call gets a complete, writable Workspace over shared history.

LayerFS is a developer preview (v0.1.7), MIT-licensed, source at https://github.com/Ephemeral-AI-Lab/layerfs.
Everything that opens a project's Store must share one Linux kernel, with the Store on a local
ext4, xfs or btrfs filesystem; setup needs a Linux host, Docker Engine and Node 20 or newer.
The core ideas: ephemeral computing (compute exists only while a command runs), agents in many
Sandboxes sharing one serverless Store (one SQLite file, no database server), multiagent
coordination on Branches with same-file conflicts refused, a recursive filesystem (forks copy
no files), deduplicated and compressed storage, and instant bootstrap (dependencies come with
every mount).

## Docs

- [Docs](https://layerfs.ai/docs/): LayerFS is an open-source filesystem for AI coding agents. Set it up with one prompt for your coding agent, then browse the docs.
- [Core ideas](https://layerfs.ai/docs/core-ideas/): The ideas behind LayerFS, one at a time, each with its animated card.
- [Concepts](https://layerfs.ai/docs/concepts/): Project, Store, Branch, Commit, Layer, Workspace, Sandbox, merge and history: the words LayerFS uses, on one page.
- [SDK reference](https://layerfs.ai/docs/sdk/): The LayerFS JavaScript SDK: the client, every call in project, branch, fs and sandbox, and its errors.
- [Platform and limits](https://layerfs.ai/docs/platform/): Where LayerFS runs, how durable its Store is, how parallel work behaves, its limits, and what history does not hold.
- [Help](https://layerfs.ai/docs/help/): Fixes for LayerFS refusals, answers to common questions, and a glossary.
- [How LayerFS compares](https://layerfs.ai/docs/compare/): LayerFS against git worktrees, copying the project, overlayfs and Btrfs snapshots for AI coding agents, and when to use something else.
- [Integrations](https://layerfs.ai/docs/integrations/): LayerFS inside the tools agents already run in: LayerFS × DSH, and more to come.
- [LayerFS × DSH](https://layerfs.ai/docs/integrations/dsh/): The dsh-layerfs plugin bundle: agents in a DeepSeek Harness session work on a LayerFS project from bash, and a LayerFS tab draws what each did.

## Architecture and benchmarks

- [Architecture](https://layerfs.ai/architecture/): How the Store, Workspaces, FUSE and the daemon fit together.
- [Benchmarks](https://layerfs.ai/benchmarks/): Speed and storage, measured, with the method and receipts.
- [dsh-dedup · Benchmarks](https://layerfs.ai/benchmarks/dsh-dedup/): What a LayerFS Store keeps when a Workspace commits 157 snapshots of a real project through the JS SDK, beside a Git pack of the same trees.
- [fio · Benchmarks](https://layerfs.ai/benchmarks/fio/): fio data I/O jobs on a LayerFS Workspace through the JS SDK, against the same Sandbox's own disk.
- [fs-bench-pro · Benchmarks](https://layerfs.ai/benchmarks/fs-bench-pro/): Agent filesystem workflows on a real Store, release by release, from each release's CSVs.
- [fs-bench · Benchmarks](https://layerfs.ai/benchmarks/fs-bench/): LayerFS against Cloudflare Computer on Computer's own twelve-scenario filesystem benchmark, run unchanged on the same image.
- [fsstress · Benchmarks](https://layerfs.ai/benchmarks/fsstress/): xfstests' fsstress through the JS SDK: random concurrent operations on a LayerFS Workspace, then the Commit remounted and compared entry by entry.
- [fsx · Benchmarks](https://layerfs.ai/benchmarks/fsx/): xfstests' fsx through the JS SDK: every byte of millions of random reads, writes, mapped I/O and truncates checked on a LayerFS Workspace.
- [mdtest · Benchmarks](https://layerfs.ai/benchmarks/mdtest/): IOR's mdtest metadata phases on a LayerFS Workspace through the JS SDK, against the same Sandbox's own disk.
- [pjdfstest · Benchmarks](https://layerfs.ai/benchmarks/pjdfstest/): pjdfstest POSIX conformance on a LayerFS Workspace through the JS SDK, against the same Sandbox's own disk.
- [smallfile · Benchmarks](https://layerfs.ai/benchmarks/smallfile/): smallfile small-file operation rates on a LayerFS Workspace through the JS SDK, against the same Sandbox's own disk.

## Optional

- [Full text of the docs, architecture and benchmarks](https://layerfs.ai/llms-full.txt)