Ask the Experts: How NVIDIA OpenShell Secures Autonomous Agents | Nemotron Labs

Autonomous agents that read files, call APIs, and run code across long-horizon tasks need more than behavioral guardrails. They need a runtime that enforces what they can actually do.

OpenShell is the safe, private runtime for autonomous AI agents — providing isolation, identity, policy, and audit below the harness layer. The harness guides what an agent tries. OpenShell determines what an agent can do.

This livestream goes inside the runtime with the product manager and engineers who built it. They'll walk through OpenShell's design principles, show how security profiles work across different workload types.

Why security controls belong at the runtime layer, not in harness logic

How OpenShell enforces policy: identity binding, credential scoping, and audit for every agent action

How security profiles work: isolated, connected, production, and adversarial workloads

How to get started on GitHub