Skip to main content
Sign In
Internals

Security Model

Trust boundaries, isolation guarantees, and the agentOS threat model.

agentOS is in beta and still undergoing security review. The security model described here is subject to change.

Trust boundaries

agentOS has two trust boundaries:

  1. Runtime boundary. The VM isolate that runs agent code. All code inside the VM is untrusted. The isolate prevents access to the host process, host filesystem, and host network.
  2. Host boundary. Your application code that configures and manages the VM. You are responsible for hardening the host process, validating inputs, and managing secrets.

VM isolation

Each agentOS actor runs in its own isolated VM:

  • Sandboxed execution. All agent code runs inside a V8 isolate with WebAssembly. No code escapes the isolate boundary.
  • Virtual filesystem. The VM has its own filesystem. Agents cannot access host files unless explicitly mounted.
  • Virtual network. The VM has no direct access to the host network. Outbound requests are proxied through the host with configurable controls.
  • Process isolation. No host process is visible or accessible from inside the VM.

What agentOS guarantees

  • Agent code cannot read or write host files outside configured mounts
  • Agent code cannot make network requests except through the host proxy
  • Agent code cannot access host environment variables or secrets
  • Each actor’s filesystem, sessions, and state are isolated from other actors
  • Resource limits (CPU, memory) are enforced at the VM level

What you are responsible for

  • Hardening the host process and deployment environment
  • Validating authentication tokens in onBeforeConnect
  • Scoping permissions appropriately for your use case
  • Managing API keys and secrets on the host side (use the LLM gateway to avoid passing keys into the VM)
  • Configuring resource limits and network controls to match your threat model

Further reading

Edit this page