NVIDIA OpenShell: Secure and private runtime environment for autonomous AI agents

OpenShell is a secure, private runtime for clustered autonomous AI agents. It uses kernel-level enforcement policies and formal verification to control agent file access, system calls, and network connections, ensuring agents work safely without accessing unauthorized data and credentials.

Hacker Newsniemandhier

OpenShellDockerPodmanOpenCodeOpenRouter

Added 2026-10-31 · originally published 2026-10-22

Steps

1Prepare the environment: Ensure the system is Linux, macOS (Apple Silicon), or Windows (with WSL 2), and install Docker,Podman, or host virtualization.
2Run the installation script to install the CLI and local gateway: Execute `curl -LsSf https://raw.githubusercontent.com/NVIDIA/OpenShell/main/install.sh | sh`
3Create a sandbox named demo: Run `openshell sandbox create --name demo`
4Install public OpenShell skills for the coding agent: Run `npx skills add NVIDIA/OpenShell`
5Refer to the official documentation to run your first agent (such as running OpenCode for the OpenRouter model), and review and approve new access permissions via policies as requested by the agent.

Copy-ready prompt

As an AI assistant, please help me write an OpenShell sandbox policy file based on the following requirements:

1. Target Agent Name: [Agent Name]
2. Allowed File Paths: [File Path List]
3. Allowed Network Hosts/API Endpoints: [Host or Endpoint List]
4. Allowed System Calls or Process Rules: [Rule Details]

Please help me generate a declaration that complies with the OpenShell kernel-level policy format, and provide risk warnings when the policy might introduce new sensitive access.

View original post Hacker News discussion

#AI Agent#NVIDIA#Security#Sandbox Isolation

Replicating Stripe's internal AI agent workflow

Handle repetitive tasks using the same AI agent setup used internally by Stripe, reducing task completion time from 20 days to 4 days compared to standard Claude Code. This system is built using over 100 saved prompts (one per step) with multiple agents collaborating to handle the build work.

Deploy visual AI workflows locally using AgentFlux and Docker

AgentFlux is an AI platform for visual workflows and autonomous agents, supporting Docker deployment on local machines or servers. It connects documents, data sources, and language models, enabling automated workflow generation from videos with results saved locally.

Generate an interactive codebase map with Claude or Codex

Use the Floorplan plugin with Claude Code or Codex to transform your application codebase into a plain-English interactive map. It displays files, modules, and cross-repo network connections, while also running app health checks and tracking recent AI modifications.

Found this useful? Get daily updates on Telegram or X.

Join on Telegram Follow @CaiXiang23766

Summarized by AI from the original source - refer to the original post for details. Copyright belongs to the original author.

Steps

Related use cases