LayerFS v1.0: Filesystem for Ephemeral AI Coding Agents

Details

LayerFS v1.0 is a filesystem substrate for AI coding agents that uses ephemeral workspaces and a durable, shared commit history to support massive parallelism.
The system achieves a 0.31x storage ratio through deduplication and compression, reducing 10 logical copies to 0.31 stored copies.
Instant bootstrap is enabled by committing dependencies, allowing workspaces to mount in 0.05 s compared to the ~5 min typical install time for standard Git-based agent filesystems.
Agents operate on different computers sharing a single serverless Store, with each file tool call acting as an atomic commit and each Bash call mounting an ephemeral workspace.
Conflict resolution uses .lww rules where the newest edit wins for specified files, while same-file clashes on other files are refused with a Conflict error.
Recursive forking allows agents to explore branches at any depth with zero file-copy cost, supporting recursive self-improvement and reinforcement learning workflows.

Key point