Security Model | agentOS Documentation
Summary
agentOS uses a three-component trust model to run untrusted code in isolated V8 or WASM executors, mediated by a trusted sidecar kernel that enforces a deny-by-default permission policy.
The kernel enforces the boundary.
The sidecar/kernel acts as the trusted computing base, owning the VFS, sockets, and permission policy to broker all requests from the untrusted executor.
It is responsible for enforcing the boundary against the executor.
Configuration is not an attack surface.
Because the client is trusted, providing malicious mount descriptors or network allowlists is considered a configuration choice rather than a sandbox vulnerability.
Guest syscalls are mediated.
Every guest syscall, including network fetch and child process spawning, is serviced by the kernel rather than the host to prevent direct access.
every guest syscall is serviced by the kernel rather than the host.
Mounts restrict guest visibility.
Host data enters the VM only through explicit configurations like files or mounts, which project specific subtrees to prevent access to the wider filesystem.
the guest sees only the subtree you mount
Side-channels are blunted.
To mitigate Spectre-style attacks, agentOS freezes high-resolution clocks and removes SharedArrayBuffer in its default mode.
Highlights
There are no host escapes: guest code cannot spawn a real host process
Configuration is not an attack surface.
The security boundary is sidecar ↔ executor.
No syscalls are bound to the system by default. Everything is denied until explicitly opted in.
There are no host escapes: guest code cannot spawn a real host process, touch the real host filesystem, or open a real host network socket.
The trusted computing base. It brokers client requests and owns the kernel, VFS, mount/plugin registry, socket table, and permission policy.
The boundary protects the host from the guest. It does not harden your host process against everything else.
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.