MXC: Cross-Platform Sandboxed Code Execution System for Untrusted Code

Introduction to MXC

MXC is a sandboxed code execution system developed for running untrusted code, including model output, plugins, and tools. It supports Windows, Linux, and macOS, providing a consistent approach to isolating potentially malicious or unstable code within applications.

Unified Containment Model and Backends

The system features a unified containment model that abstracts various underlying containment technologies. These range from OS-native process sandboxes like ProcessContainer, Windows Sandbox, LXC, Bubblewrap, and Seatbelt, to full virtual machines such as MicroVM (Nanvix), Hyperlight, IsolationSession, and WSLC. This allows applications to select the appropriate level of isolation based on their needs and the operating system.

Policy-Driven Sandboxing and SDKs

MXC implements policy-driven sandboxing, enabling granular control over filesystem access (read-only, read-write, denied paths), network activity (proxy support, outbound controls), and UI interactions (clipboard, display, GUI access). Developers can integrate MXC into their applications using provided SDKs for Rust, .NET, and Node, which offer versioned APIs for both one-shot and state-aware execution of containers. The system also includes diagnostic tools to troubleshoot access-denied failures within a container.

Integration and Workload Execution

Applications integrate MXC as an SDK dependency. The application specifies the container type, containment rules, and the workload command. MXC then validates the request, selects the appropriate backend, and launches the workload within the isolated container. For scenarios where SDK embedding is not feasible, platform-specific executor binaries are available, accepting JSON container-creation requests based on a stable schema.