VibeKit 🖖 The Safety Layer for Your Coding Agent

Hey YC! 👋
We’re building VibeKit, an open-source safety layer for Claude Code, Gemini CLI, and other coding agents.

❌ The Problem

We love coding with AI agents — until the day Anthropic emailed us to say they had found our API keys in a public GitHub repo and revoked them.

The leak came from our own coding agent. That’s when we realized most developers have no way to see — let alone control — what their AI agents are doing:

• Which files they read or write
• What shell commands they run
• What data leaves their machine
• Whether secrets get exposed in API calls

🖖 The VibeKit Solution

VibeKit is a universal CLI wrapper that runs any coding agent in a local, isolated Docker sandbox,

It comes with:

• Secret redaction before anything leaves your machine
• Full observability into file changes, commands, API calls, and outbound data
• Offline-first, open-source (MIT), and works with any agent — Claude, GPT-4o, Gemini, etc.

💻 Try it out

# Install globally
npm install -g vibekit

# Run Claude Code with VibeKit
vibekit claude

Then watch everything your AI is doing — safely.

🎥 Demo video: https://youtu.be/vDwJ4xcdRBM\

📂 GitHub: https://github.com/superagent-ai/vibekit\

📖 Docs: https://docs.vibekit.sh

🙏 Our Ask

If you use coding agents day-to-day:

• Run them through VibeKit — see what they’re really doing.
• Share your weirdest or most alarming finds with us — we’re building a public library of “agent fails.”

💌 Contact: founders@vibekit.sh

---