Open-source, sandboxed AI agents for every employee, fronted by a gateway that injects scoped credentials so the agent never holds a password.
Microsoft Execution Containers (MXC)
Summary
Microsoft's open-source, policy-driven sandbox SDK for running agents, model-generated code, tools and plugins with only the files and network access you grant.
Description
Microsoft Execution Containers (MXC) is a policy-driven execution layer for untrusted or dynamically generated workloads. Developers declare which files, network destinations and UI an agent, tool, plugin or model-generated script may use, and MXC picks an appropriate container to enforce that policy at runtime, independently of the agent itself.
One JSON configuration schema works across Windows, Linux and macOS, with backends including process containers, Windows Sandbox, WSL containers, Bubblewrap/LXC on Linux and Seatbelt on macOS. Modes for enforcing, learning and permissive runs produce an activity report you can turn into a policy. SDKs ship for Rust, .NET and Node (npm @microsoft/mxc-sdk). MIT licensed; version 1.0 marked general availability in October 2026.
Reviews
No reviews yet — sign in to write the first one
Similar App Suggestions
An open-source layer that lets your customers describe a feature and get a working micro-app inside your product — sandboxed to their own permissions.
Open-source computer-use infrastructure: a desktop driver, cloud desktop fleets, local macOS VMs and a benchmark for agents that click.
Security scanner and runtime monitor for AI agents that audits skills, MCP servers and CLI tools for malicious code and prompt injection.
Write an agent as a TypeScript function and deploy it in seconds onto KVM-isolated Linux microVMs, with the loop, sessions, streaming and versioning managed for you.
Covered in the Weekly
- Onei AI Weekly #11 — October 12, 2026