Agents need more than model-level guardrails. They also need runtime boundaries that control what they can access and do.
For the NVIDIA Open Agent Safety Platform launch, we’re sharing how DeepInfra approaches agent isolation, our work with OpenShell, and a tested guide for
