[NVIDIA Releases Open-Source AI Agent Runtime OpenShell 0.1.0] According to Deep Tide TechFlow, on September 28, NVIDIA officially launched the open-source AI agent runtime platform OpenShell 0.1.0. This platform aims to establish secure boundaries and runtime controls outside the workload without modifying the AI agent's own code. OpenShell integrates technologies such as kernel-level sandbox execution, external regulatory services, external credential injection, and formal policy verification. It supports fine-grained interception of HTTP, GraphQL, and MCP network traffic (e.g., allowing data reading but blocking write operations) and prevents sensitive credential leaks. Additionally, the platform includes a built-in formal logic policy analyzer to prevent AI agents from bypassing permission checks through adversarial prompts. Currently, OpenShell supports mainstream frameworks such as Codex and Claude Code and is applied in fields like chip design, enterprise automation, and robotics.