Self-hosted runtime control plane for AI agents. Blocks the rogue tool call before it executes: dangerous commands, secret leaks, prompt injection. Works with any ai agent like Claude Code, Codex, and framework SDKs. - View it on GitHub