Limitations

What the agentOS VM does not support, and how to work around it.

agentOS is a hybrid model. The lightweight VM handles most agent workloads (coding, scripting, file I/O, API calls) with near-zero overhead. When a workload hits a VM limitation, agents can escalate to a full sandbox on demand without changing code. The two share a filesystem, so files written in the VM are available in the sandbox and vice versa.

See agentOS vs Sandbox for a detailed comparison.

No native Linux binaries

agentOS runs JavaScript, TypeScript, Python, and shell commands. Native x86/ARM binaries compiled for Linux cannot run directly.

• Precompiled CLI tools (e.g. docker, kubectl, terraform)
• Native language runtimes (Go, Rust, C++ binaries)
• System packages installed via apt or yum

No real Linux kernel

The VM uses a virtual kernel. System calls that require a real Linux kernel are not available.

• Raw sockets and custom network protocols
• Kernel modules and eBPF
• Container runtimes (Docker-in-Docker)
• Low-level filesystem operations (inotify, fuse)

No hardware access

The VM has no access to GPUs, USB devices, or other hardware.

• GPU-accelerated ML inference
• Hardware-dependent testing