# LayerFS

> LayerFS gives each AI agent an ephemeral writable filesystem Workspace over shared, deduplicated committed history.

LayerFS is an open-source filesystem substrate for multi-agent development from Ephemeral-AI-Lab. Each agent mounts a complete, writable Workspace over shared committed history, does its work, commits what changed, and goes away. Version 1.0.0 is the first official release, published as source under the MIT license.

## What It Is

LayerFS stores every Branch, Commit and content object in a single SQLite file, so there is no database server. File tool calls go to the Store in process through the SDK and each becomes a Commit. Each Bash call mounts a FUSE-based Workspace inside a Sandbox daemon, runs the command, commits, and unmounts. Unmounting deletes the overlay but never history.

## How Coordination and Storage Work

- Branches are writable lines of Commits, and publishing a Branch to main adds a Layer.
- Forks are metadata-only, so agents can fork any Commit recursively at any depth.
- Two Commits that change the same file are refused with a Conflict, unless a .lww rule says the newest edit wins.
- Content is chunked, deduplicated and compressed, and ignored files such as dependencies stay in every commit, so mounts come with them already installed.
- Every step is a checkpoint that can be forked from or read as it was, and blame names the step that wrote each line.

## Setup Path and Current Status

LayerFS currently runs through LayerFS × DSH, the dsh-layerfs plugin for DeepSeek Harness, which shows each agent's Branch live in the Work and History tabs. The docs describe a paste-a-prompt setup for a coding agent, which needs Node 22.19+, pnpm and Docker. The JavaScript SDK needs a Linux host, Docker Engine, Node 20+ and an ext4, xfs or btrfs filesystem. The README states the Store runs only in a Disposable profile, with no crash or power-loss durability, so it should not be the only copy of data you cannot rebuild. The project publishes benchmark suites, which the README describes as release measurements rather than a guarantee for any workload.

## Features
- Ephemeral writable Workspaces mounted over shared committed history
- Every tool call recorded as a Commit with auto checkpoints
- Metadata-only recursive forking of any Commit
- Branches, merging and publishing to main
- Conflict detection with .lww last-writer-wins rules
- Content deduplication and compression in a single serverless SQLite Store
- Instant bootstrap with ignored files and dependencies preserved in commits
- Append-only content-addressed objects
- Per-line blame naming the step that wrote each line
- JavaScript SDK with fs and project APIs

## Integrations
DeepSeek Harness (DSH), Docker, SQLite, FUSE

## Platforms
WINDOWS, MACOS, LINUX, API, DEVELOPER_SDK

## Pricing
Open Source

## Version
1.0.0

## Links
- Website: https://layerfs.ai/
- Documentation: https://layerfs.ai/docs/
- Repository: https://github.com/Ephemeral-AI-Lab/layerfs
- EveryDev.ai: https://www.everydev.ai/tools/layerfs