MCP server for AI coding agents that keeps raw tool output out of the context window and persists session state across compaction.
Ephemeral-AI-Lab/layerfs
A SQLite-backed, content-addressed filesystem store that gives AI agents disposable workspaces and durable, deduplicated checkpoints of shared history.
- Stars
- 620
- Forks
- 34
- Open issues
- 37
- Last commit
- 13 hours ago
What is it?
LayerFS is a Rust project for managing filesystem state for AI agents. Each filesystem-affecting tool call runs in its own ephemeral Workspace, and retained results become immutable Commits in a LayerStack timeline. Content-addressed storage, content-defined chunking and copy-on-write keep one shared base plus only the unique changes, so new states cost roughly what changed rather than the size of the whole environment.
From any Layer or Commit, a caller can fork zero-copy Branches, run parallel workspaces, discard failed attempts, roll back by forking an earlier state, and promote a Branch with Add. The project ships a Rust SDK, a CLI, host materialization and Linux FUSE container workspaces. The current v0.1.6 release is a developer preview for local evaluation and agent-runtime integration, and it does not yet provide crash or power-loss durability guarantees.
Key features
- 01Content-addressed storage that reuses identical data across files and states.
- 02Content-defined chunking keeps localized edits from rewriting whole large files.
- 03Zero-copy Branches and copy-on-write Commits from any Layer or Commit.
- 04Workspace execution and commits through a Rust SDK and CLI.
- 05Linux FUSE container workspaces, with host materialization as an alternative.
Use cases
- Running many parallel agent rollouts over the same codebase without cloning environments.
- Rewinding an agent's filesystem to an earlier checkpoint after a failed attempt.
- Testing environment or dependency changes in isolated branches before promoting one.
Topics
FAQ
- What is layerfs?
- A SQLite-backed, content-addressed filesystem store that gives AI agents disposable workspaces and durable, deduplicated checkpoints of shared history. LayerFS is a Rust project for managing filesystem state for AI agents. Each filesystem-affecting tool call runs in its own ephemeral Workspace, and retained results become immutable Commits in a LayerStack timeline. Content-addressed storage, content-defined chunking and copy-on-write keep one shared base plus only the unique changes, so new states cost roughly what changed rather than the size of the whole environment.
- Is layerfs free to use? What license does it have?
- layerfs is open source under the MIT license. Check the license terms before using it in commercial products.
- Is layerfs still maintained?
- Actively maintained. The last commit to layerfs was on Oct 18, 2026.
- What can I use layerfs for?
- Running many parallel agent rollouts over the same codebase without cloning environments. Rewinding an agent's filesystem to an earlier checkpoint after a failed attempt. Testing environment or dependency changes in isolated branches before promoting one.
- Where can I get layerfs?
- The source code and installation instructions are on GitHub: https://github.com/Ephemeral-AI-Lab/layerfs
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A self-hostable workspace where humans and AI agents share channels, built on a Nostr relay you run yourself.
A distributed, Dolt-backed graph issue tracker that gives coding agents persistent, dependency-aware task memory via the bd CLI.
A Rust CLI proxy that compresses shell command output before it reaches an AI coding agent, cutting LLM token use on common dev commands.
A minimal, cross-shell prompt written in Rust that shows relevant project and environment info and works with Bash, Zsh, Fish, PowerShell and more.
A command-line tool that exposes all of Google Workspace (Drive, Gmail, Calendar, Sheets, Docs, Chat, Admin) with structured JSON output for humans and AI agents.