Filesystem

Each VM presents a normal POSIX filesystem to guest code, backed by a virtual filesystem inside the kernel. At a glance:

• Per-VM virtual filesystem: Every VM gets its own isolated virtual filesystem. One VM cannot see or reach another VM’s files.
• Never touches the host disk: Guest filesystem calls are served entirely inside the kernel and never read or write the real host disk.
• Normal POSIX and Node APIs: The standard node:fs and node:fs/promises APIs work as usual against the virtual filesystem, so ordinary programs run unchanged.
• Mountable backends: You can project host-backed sources into the guest filesystem, Docker-style, including host directories and S3. Mounts are confined to their root, and the guest sees only the mounted subtree.
• nodeModules is a mount: NodeRuntime.create({ nodeModules }) is a convenience for a read-only host-directory mount. It projects a host node_modules tree at guest /tmp/node_modules using the same mount machinery as mounts, placed on the package-resolution path. See Module loading.

Full reference

The canonical filesystem API, including seeding files, host-boundary file exchange, and the full mount and backend configuration, is owned by agentOS.

agentOS: Filesystem The complete filesystem API plus mount and backend configuration details.