OpenShell is NVIDIA’s open-source runtime for running autonomous AI agents safely and privately. Agents are most useful when they can read files, install packages, call APIs and use credentials, but handing them unrestricted access to your data, secrets and network is risky. OpenShell solves that tension: you declare in a policy exactly what each agent is allowed to touch, and OpenShell enforces it at runtime. The project has gathered more than 16,000 stars on GitHub since its release and is under active development in 2026, with a stable 0.1.x release cadence.
OpenShell governs agents in two complementary ways. First, it instruments the kernel: every agent runs in an isolated sandbox, and policy is enforced on every file access, system call and network connection as it happens. Second, it applies formal verification to policy itself, checking what a proposed policy change would allow before the change is applied, so misconfigurations are caught before they become exposures. An extension surface and new APIs in the 0.1.x series make it easier to integrate OpenShell into existing agent frameworks and fleet deployments, letting many agents run with least-privilege access to tools and credentials.
OpenShell is free and open source under the Apache-2.0 license. There are no paid tiers; you run it on your own infrastructure.
OpenShell is for platform and security teams deploying autonomous agents in production who need provable, kernel-enforced limits on what each agent can read, execute and connect to.