LayerFS

Overview

LayerFS is the substrate under AI agents. Each agent mounts a complete, writable Workspace over shared committed history, performs its work, commits what changed, and goes away. One SQLite file holds every Branch, Commit and content object without a database server.

Key Features

• Ephemeral computing: compute exists only while a command runs; each Bash call mounts a Workspace, executes, commits and unmounts.
• Decentralized agents on different computers share one serverless Store, with no database server.
• Multi-agent coordination through branches: each tool call is a commit, same-file clashes are refused as Conflict, and blame shows which step wrote each line.
• Recursive filesystem forks any Commit at any depth; a fork is metadata and copies no files.
• Storage optimization through content deduplication and compression, while dependencies and ignored files stay available on mount.

Use Cases

• Parallel multi-agent software development with isolated writable Workspaces over shared history.
• Agent workflows that need checkpoints, rewind, branching and merge.
• Recursive exploration and branching for RL training or agent search.
• Running agents across multiple computers against one shared Store.

Getting Started

• Try LayerFS through LayerFS × DSH, the dsh-layerfs plugin for DeepSeek Harness, using the DSH install guide.
• Consult the LayerFS documentation and core ideas before setup.

Deployment & Requirements

• Linux, macOS through DSH, and Windows through WSL2 are listed as supported platforms.
• The Store is a single SQLite file opened directly by each daemon and SDK process; no database server is required.
• Running LayerFS × DSH requires the dsh-layerfs plugin for DeepSeek Harness.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• The FUSE view intentionally omits certain POSIX node types and metadata.