NVIDIA has open-sourced a pretty interesting project!


OpenShell, giving the AI programming assistant a "safety cage." Usually, when you ask Claude Code or Codex to help you write code, are you worried they might secretly access files they shouldn't or send data out?
This is the solution to that problem. AI runs in a sandbox, with the file system, network, and processes all controlled, so any malicious activity can be directly blocked.
API keys are not exposed to the sandbox, relying solely on policy-based routing and forwarding.
The coolest part is that policies are written in YAML, and after editing, hot reloads without needing a restart.
It also comes with a terminal monitoring panel similar to k9s, allowing real-time observation of what the AI is doing.
Supports mainstream agents like Claude Code, Codex, Ollama, and even GPU passthrough.
If you want AI to help you get work done without giving it too much permission, this is quite practical.
Project link is in the comments 👇
#NVIDIA #AI Security #开源 #ClaudeCode #DeveloperTools
View Original
post-image
post-image
post-image
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin