> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/openshell/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/openshell/_mcp/server. # Support Matrix This page lists the host platform, compute driver, software, runtime, and kernel requirements for running OpenShell. ## Supported Platforms OpenShell publishes multi-architecture gateway images for `linux/amd64` and `linux/arm64`. The CLI, policy prover, package-managed gateway, and standalone gateway binary are supported on the following host platforms: | Platform | Architecture | Status | | -------------------------------- | --------------------- | ------------ | | Linux (Debian/Ubuntu) | x86\_64 (amd64) | Supported | | Linux (Debian/Ubuntu) | aarch64 (arm64) | Supported | | macOS (Docker Desktop) | Apple Silicon (arm64) | Supported | | Windows (WSL 2 + Docker Desktop) | x86\_64 | Experimental | On Linux, the `openshell` CLI is a static musl binary and does not require glibc at runtime. ## Standalone Gateway Binary OpenShell publishes standalone `openshell-gateway` release assets for manual download on these platforms: | Platform | Artifact pattern | | --------------------- | --------------------------------------------- | | Linux x86\_64 (amd64) | `openshell-gateway-x86_64-unknown-linux-gnu` | | Linux aarch64 (arm64) | `openshell-gateway-aarch64-unknown-linux-gnu` | | macOS Apple Silicon | `openshell-gateway-aarch64-apple-darwin` | These artifacts are attached to GitHub releases. Kubernetes deployments should use the Helm chart and the published gateway image. On Linux, `openshell-gateway` requires glibc 2.28 or newer. Compatible systems include, for example, Ubuntu 20.04+, RHEL 8+, Rocky Linux 8+, Amazon Linux 2023+, and Fedora 32+. ## Standalone Policy Prover OpenShell publishes standalone `openshell-prover` release assets for manual download on these platforms: | Platform | Artifact pattern | | --------------------- | ---------------------------------------------------- | | Linux x86\_64 (amd64) | `openshell-prover-x86_64-unknown-linux-musl.tar.gz` | | Linux aarch64 (arm64) | `openshell-prover-aarch64-unknown-linux-musl.tar.gz` | | macOS Apple Silicon | `openshell-prover-aarch64-apple-darwin.tar.gz` | These artifacts are attached to GitHub releases. The Linux binaries are static and do not require glibc. All prover archives include the required solver linkage. ## Runtimes The gateway can manage sandboxes through several runtimes. | Runtime | Status | Notes | | ---------- | ---------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------ | | Docker | Supported for local development and single-machine gateways. | Requires Docker Desktop or Docker Engine on the gateway host. | | Podman | Supported for rootless local and workstation workflows. | Requires a Podman-compatible socket and rootless networking setup. | | Kubernetes | Supported through the [OpenShell Helm chart](https://github.com/NVIDIA/OpenShell/blob/main/deploy/helm/openshell/README.md). | Requires a Kubernetes cluster supplied by the operator. | | MicroVM | Supported for VM-backed sandboxes. | Uses the VM compute driver and libkrun-based runtime. | ## Software Prerequisites Install the software for the compute driver you use: | Component | Minimum Version | Notes | | ------------------------------- | ---------------------------- | -------------------------------------------------------------------------------------------------- | | Docker Desktop or Docker Engine | 28.0 | Required for Docker-backed gateways, local image builds, and Docker development workflows. | | Podman | 5.x | Required for Podman-backed gateways. | | Kubernetes | 1.29 | Required for Helm deployments and Kubernetes sandbox scheduling. | | Helm | 3.x | Required to install `deploy/helm/openshell`. | | kubectl | Compatible with your cluster | Required for Kubernetes operational inspection and secret creation. | | Host virtualization | Host dependent | Required for MicroVM-backed gateways. MicroVM uses Hypervisor.framework on macOS and KVM on Linux. | ## Workload Images OpenShell accepts standard Linux OCI images. The built-in default workload is `nvcr.io/nvidia/base/ubuntu:24.04` for `linux/amd64` and `linux/arm64`. It is a minimal Ubuntu Noble userspace and does not bundle agent CLIs. Operators can change the default, and users can select an explicit image with `--from`. ## Container Images | Image | Reference | Pulled When | | ---------------- | ----------------------------------------- | ----------------------------------------------------------------- | | Gateway | `ghcr.io/nvidia/openshell/gateway:latest` | Helm chart install or upgrade, or standalone container deployment | | Default workload | `nvcr.io/nvidia/base/ubuntu:24.04` | First sandbox creation unless preloaded or overridden | The Helm chart in `deploy/helm/openshell` deploys the gateway workload, service account, service, optional persistent storage, and network policy for Kubernetes. It defaults to a StatefulSet for SQLite-backed installs and can render a Deployment for external database-backed installs. To override the default image references, use Helm values: | Helm value | Purpose | | ---------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | | `global.image.registry` / `global.image.tag` / `global.image.pullPolicy` | Override shared image settings for the gateway, supervisor, and trusted sandbox runtime. | | `gateway.image.registry` / `gateway.image.repository` / `gateway.image.tag` / `gateway.image.digest` | Override the gateway image reference. | | `supervisor.image.registry` / `supervisor.image.repository` / `supervisor.image.tag` / `supervisor.image.digest` | Override the supervisor image reference. | | `sandbox.image.repository` / `sandbox.image.tag` / `sandbox.image.digest` | Override the default workload sandbox image. | ## Kernel Requirements The sandbox boundary requires the following Linux kernel facilities, including when it runs inside a container or microVM: | Module | Requirement | Details | | -------------------------------------------------------------- | ------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | [Landlock LSM](https://docs.kernel.org/security/landlock.html) | Required | ABI 3 or newer, introduced in Linux 6.2, with Landlock enabled. The mandatory baseline protects private channel and bootstrap files, including against truncation. A filesystem policy's `best_effort` setting never disables this baseline. | | seccomp | Required | Nested user-notification filters and atomic `SECCOMP_IOCTL_NOTIF_ADDFD` with `SECCOMP_ADDFD_FLAG_SEND`, usable under the runtime's existing seccomp profile without added capabilities. The sandbox actively probes these operations before admitting the workload. | | Task-memory access | Required | The non-dumpable broker must be able to read and write a same-UID, dumpable workload child's memory through `process_vm_readv` / `process_vm_writev` or `/proc//mem`. The sandbox actively probes the production parent-to-child topology before admitting the workload. | | seccomp `WAIT_KILLABLE_RECV` | Recommended (Linux 5.19+) | Keeps a notified workload thread in a kill-only wait so the broker can safely write mediated results into workload memory. Without it (kernels \< 5.19, for example RHEL 9.x / RHCOS 5.14) the sandbox still starts, in a reduced **legacy read-only** mode described below. | A kernel version alone does not establish support. A disabled Landlock LSM or a runtime profile that blocks the required seccomp operations causes launch to fail closed. An upstream Linux 6.2 or newer kernel provides the required Landlock ABI; distribution backports must pass the same active qualification. The broker remains non-dumpable during qualification. A runtime may satisfy task-memory access through `/proc//mem` even when its kernel omits the `process_vm_readv` and `process_vm_writev` system calls; OpenShell qualifies the same parent-to-executed-child access shape used by mediated workloads. ### Legacy read-only mode (kernels before Linux 5.19) `SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV` was added in Linux 5.19. On older kernels — notably RHEL 9.x and RHCOS, which ship a 5.14 kernel — the sandbox cannot install a kill-only listener, so it falls back to a plain listener and runs in a **legacy read-only** cancellation mode. The sandbox starts and enforces the full isolation boundary (Landlock, the outer NetworkPolicy fence, DNS and TCP authorization); the only difference is that the broker refuses the mediated operations that write results back into workload memory, failing them closed with `EOPNOTSUPP`: * `getpeername`; * `accept` / `accept4` **when a non-null peer-address argument is supplied** (a null address argument still works); * `sendmmsg` paths that write per-message lengths back to the caller. Socket creation, `connect`, `bind`, `listen`, `sendto`, and `sendmsg` are unaffected. Outbound-oriented workloads generally run unchanged; server workloads whose accept wrappers request the peer address will see `EOPNOTSUPP` until the node runs a kernel that provides `WAIT_KILLABLE_RECV` (Linux 5.19+, or a distribution backport). The selected mode is reported in the sandbox qualification output as `seccomp_listener_mode` (`killable` or `legacy_read_only`). On macOS, these kernel modules run inside the Docker Desktop Linux VM, not on the host kernel. ## Agent Workloads OpenShell runs agents and tools that you install in a user-owned OCI image. The image must satisfy the selected compute driver's platform requirements. Refer to [Run Your First Agent](/about/run-an-agent) for the image, provider, and policy workflow.