{"slug": "layerfs-the-filesystem-for-ai-coding-agents", "title": "LayerFS: The filesystem for AI coding agents", "summary": "LayerFS has been introduced as a filesystem built for AI coding agents that commits every file tool call as an atomic Store operation and mounts an ephemeral workspace for each Bash tool call, letting agents on different computers share one serverless Store without a database server. LayerFS reports that 10 copies of content compress to a ratio of 0.31 after deduplication (10 copies to 1.05) and compression, that forks are metadata-only and cost nothing, and that mounting a workspace with dependencies takes 0.05 seconds versus roughly 5 minutes to reinstall them. The project targets recursive fork exploration for reinforcement learning training and recursive self-improvement agents, with same-file conflicts refused and per-line blame naming the step that wrote each line.", "body_md": "Compute exists only while a command runs.\n\n- **Compute:** Agents spend ~80% thinking, ~10% in Bash.\n- **FS:** Each file tool call is one atomic Store op.\n- **Sandbox:** Each Bash tool call mounts an ephemeral workspace, then commits.\n\nAgents on different computers share one serverless Store.\n\n- **Shared storage:** Every agent commits to one Store, with no DB server.\n- **Distributed computing:** Each agent runs on its own computer.\n\nOne project, many branches. One branch, many agents.\n\n- **Commits:** Every tool call is a commit on its branch.\n- **Conflict:** A same-file clash is refused with Conflict.\n- **Blame:** Each line names the step that wrote it.\n\nFork recursively any commit, at any depth.\n\n- **Zero cost:** A fork is metadata and copies no files.\n- **Recursive:** A filesystem for RSI agents and recursive exploration for RL training.\n\n10 copies compress into 0.31.\n\n- **Dedup:** Each content chunk is stored once.\n- **Compression:** The stored chunks shrink further.\n- **Ratio:** 10 copies → 1.05 dedup → 0.31 compression wtf???\n\nDependencies come with every mount. No wasting time for re-installation.\n\n- **Git / other AgentFS:** Drops ignored files; reinstalls.\n- **LayerFS:** Ignored files stay in every commit.\n- **Comparison:** ~5 min to install vs 0.05 s to mount.", "url": "https://wpnews.pro/news/layerfs-the-filesystem-for-ai-coding-agents", "canonical_source": "https://layerfs.ai/", "published_at": "2026-10-21 12:34:46+00:00", "updated_at": "2026-10-21 12:45:35.302386+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "ai-infrastructure"], "entities": ["LayerFS"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/layerfs-the-filesystem-for-ai-coding-agents", "markdown": "https://wpnews.pro/news/layerfs-the-filesystem-for-ai-coding-agents.md", "text": "https://wpnews.pro/news/layerfs-the-filesystem-for-ai-coding-agents.txt", "jsonld": "https://wpnews.pro/news/layerfs-the-filesystem-for-ai-coding-agents.jsonld"}}