NVIDIA OpenShell
VerifiedOfficialby NVIDIA
Official listing: built and published by NVIDIA. More from NVIDIA
Runs autonomous agents in isolated sandboxes with kernel-level policy enforcement over files, syscalls, and network, injecting credentials only for approved endpoints.
- Access
- CLIAPI
- Auth
- None
- Skills
- 4
Skills
Agent Sandboxes
Creates and manages isolated agent sandboxes from the openshell CLI, with configurable images, runtimes, GPU access, and lifecycle.
Sandbox Policies
Enforces declared filesystem, network, and process rules at the kernel level so every file access and connection is checked at runtime.
Policy Change Review
Uses the policy advisor and formal-verification prover to flag risky new access in a policy change and hold it for human review.
Credential Providers
Keeps real credentials out of the sandbox and attaches them only to requests bound for approved endpoints, including inference routes.
Related Agents
NVIDIA NemoClaw
by NVIDIA
Open-source security runtime for multi-agent AI. Enforces policy-based access control, sandboxes kernel interactions…
OpenSandbox
by OpenSandbox Group
Runs AI-agent workloads in isolated Docker or Kubernetes sandboxes, exposing sandbox lifecycle, command, filesystem…
HOL Guard
by Hashgraph Online
Blocks risky AI coding agent actions at runtime on your machine — shell commands, secret access, MCP servers, plugins…
Praesidia
by Squad Technology SRL
Govern AI agent and MCP server access with runtime policy, approvals, guardrails and a signed audit trail, through a…