Kern

This product is a rootless container runtime and sandbox that runs without a background daemon. It is distributed as a single static binary and is designed to run agent tool calls and LLM generated code inside kernel enforced containers, giving code that has not been reviewed a container of its own each time it runs. It targets developers and teams building AI agents that execute model written code, as well as anyone wanting a daemonless way to run container images. It works with real OCI images, so users can pull, build from a Dockerfile, commit, push, save, and load. It reads existing Docker Compose files. For untrusted code, a single security profile flag applies user, PID, mount, network, UTS, and IPC namespaces, a deny by default seccomp allowlist, dropped capabilities, and cgroup v2 limits. CPU, memory, disk, and device limits can be declared once in a configuration file and attached by name. A separate Python package called Kern Sandbox creates a fresh container for every call a model makes and destroys it when the call returns. It runs on Linux for x86_64 and ARM directly. On Windows it runs through WSL2, and on macOS inside a Linux VM, since the vendor states there is no native macOS port. An install script verifies a SHA256 checksum before installing. No pricing is published.