I built AgentVault in 3 hours last weekend after watching the OpenClaw security situation explode. THE PROBLEM: OpenClaw (Anthropic's Claude computer use framework) had a nightmare week - 5 CVEs discovered, 341 malicious skills in their marketplace, and 42,000 instances exposed to the internet. People are running AI agents with full terminal access, network access, and file system access with ZERO visibility into what's happening. I realized: we're giving AI the keys to our systems and just... hoping for the best? WHAT I BUILT: AgentVault is a security proxy that sits between Claude and your system. Every command, every network request, every file access goes through AgentVault first. You get real-time visibility into everything your agent does, automatic blocking of dangerous patterns (rm -rf, sketchy network requests), a permission system where YOU decide what happens, and complete audit trails. WHY SO FAST? Because the problem was urgent and nothing else existed. All the "solutions" are enterprise-only or vaporware. I needed this for my own Claude setup and figured others did too. Built with Node.js/TypeScript proxy, SQLite logging, Next.js dashboard. Self-hostable, lightweight, works today. WHAT'S NEXT: Right now it's built for OpenClaw but I'm expanding to general AI agent security - LangChain, AutoGPT, custom frameworks. Looking for early users to help shape what this becomes. Is AI agent security something you're thinking about? Would love to hear what features matter most to you.

AgentVault

Realtime security monitoring for AI agent for Openclaw

Realtime security monitoring for AI agent for Openclaw