Storage Gateway
How agents see a normal POSIX filesystem while the orchestrator transparently enforces security and routes I/O to the right backend.
Storage Gateway
Inside an agent container, files just work. The agent reads, writes, lists, and seeks against ordinary paths — the same POSIX operations any program expects. There is no special SDK, no signed-URL dance, no object-store API to learn. Code that runs locally on a developer's laptop runs unchanged inside an AEGIS execution.
Underneath that ordinary filesystem, the Storage Gateway is doing real work. Every read and write passes through a transport-agnostic security layer that validates the file handle, checks that the execution actually owns the volume it is touching, canonicalizes the path to block traversal attacks, squashes UID/GID so kernel permissions cannot leak across tenants, and emits an audit event. The transport mechanism varies by runtime — NFS for Docker, a host-side FUSE daemon for rootless Podman, virtio-fs for Firecracker microVMs — but the authorization logic is written once and reused. The agent sees one thing; the platform enforces another.
Key ideas
- POSIX everywhere — agents use normal file operations; no custom storage SDK in the agent runtime.
- Transport-agnostic FSAL — a single security layer (the File System Abstraction Layer) backs every transport.
- Per-volume mounts — each execution mounts only the volumes it owns, scoped to its lifetime.
- Pluggable backends — the same agent code works against SeaweedFS, S3, or local disk without change.
- Full audit trail — every file operation is recorded for forensics and Cortex learning.
Learn more
- Storage Gateway — the security and authorization model in depth.
- Storage Backends — the swappable backends behind the gateway.
- Configuring Storage — how to wire volumes and mounts in your deployment.
- SeaweedFS — running the default object backend.