LayerFS

LayerFS is an open-source file system layer from Ephemeral-AI-Lab that its README calls "the substrate under your agents." It gives coding agents writable workspaces on top of a shared history, and their changes can be branched, merged and rolled back the way code can. The history is stored in a local SQLite database on your machine, with no need to upload it to a LayerFS cloud. The currently recommended setup pairs it with DeepSeekHarness (DSH).

Who it's for:Developers running several AI coding agents at once who want to keep their changes separate, merge them and roll back at any time. Readers who don't write code won't have a use for it.

Before you start: The recommended route needs Node 22.19 or newer, pnpm and Docker, and you build its native background program yourself; installation downloads packages, Rust crates and container images. The docs ask you to create a new DSH_HOME profile rather than reuse your existing ~/.dsh. It runs with high privileges: the code mounts /var/run/docker.sock, all of Docker's volume directories and a writable home directory, which effectively gives it control of Docker on your computer, and it also mounts a file system locally. Once connected to DSH, the project content your agents read goes to the model provider you choose, which needs its own account or key. The README says the current version doesn't guarantee your data survives a crash or power loss, so don't use it as the only copy of anything. The code is MIT-licensed and the project itself is free; models, compute and Docker-related costs are extra.

Based on the project's public documentation. Not tested by AI Inspiration World.