OpenShell – A safe, private runtime for fleets of autonomous AI agents

442 points · 201 comments on HN · read original →

Points and comments are a snapshot, not live.

NVIDIA released OpenShell, a safe runtime for autonomous AI agents.

OpenShell governs agent actions via kernel-level enforcement and formally verified policy changes.

It isolates agents in sandboxes, controls file access, system calls, and network connections, and replaces credentials only for approved endpoints. Policy changes undergo formal verification to flag risky new access before human approval.

Supports Linux, macOS on Apple Silicon, or Windows with WSL 2. Requires Docker, Podman, or host virtualization.

What commenters are saying

Commenters noted NVIDIA showcased OpenShell heavily at GTC Europe, calling it "absolutely awesome" with a great feature bundle.

Telemetry sparked debate: some dislike it on principle despite opt-out and compilation options, citing IP exposure or distrust. A few argued contributing usage data helps improve the free tool.

One commenter claimed the agentic industry moves away from k8s toward spikier schedulers, though others disagreed, saying most orgs still use OpenShift or bare k8s.