Stop babysitting your AI coding agents.
When your agent reaches for something risky — rm -rf, force-push, deploy — your phone buzzes with the exact command (secrets redacted). Approve or Deny from anywhere; it keeps moving.
🔒 Private by design
Only the one command awaiting approval leaves your machine — API keys, tokens and DB URLs masked first. Your session never leaves. Open source, self-hostable.
🧩 Every agent, one layer
Claude Code via an automatic hook; Codex, Gemini CLI, Cursor and any MCP agent via a request_approval tool. One approval flow for all of them. Set up Codex / Gemini / Cursor →
🛑 A tripwire, not a cockpit
Auto-approve the safe stuff; get buzzed only for the dangerous moves; veto from your phone. A remote kill-switch — so you run agents more autonomously.
Set up in ~2 minutes
Install the plugin on your Mac
Run this once in your terminal:
curl -fsSL https://pwa-artvibes-projects.vercel.app/install.sh | bashThis installs the Claude Code hook at ~/.claude/plugins/push-bridge. Source on GitHub (MIT).
Enter your license key
Find your key in your LemonSqueezy receipt email. Free BYO mode? Skip this and use the plugin alone.
Enable notifications on this device
Pricing
Free
$0
Open-source plugin, bring your own webhook (ntfy.sh). Get pinged when Claude Code needs you — no hosted account.
Pro
$9/mo
Hosted relay + iPhone app. Review the command and Approve/Deny from your phone. Annual $79 (save 27%).
Get Pro →On iPhone you tap the notification, then Approve/Deny in the app (Apple doesn’t allow buttons directly on web-push notifications yet).