Point Tzu at any address — a rival, a tool you're weighing up, or your own app. It reads what that app already tells the world — the stack, the hosting, the keys shipped in the JavaScript, the controls nobody switched on — and hands you back a teardown.
Passive observation only — the same requests your browser already makes. No signup to look, and nobody gets told you looked.
Point Tzu at any app. It fingerprints the stack, maps what's exposed, and tells you how solid that app actually is. Most apps built in a weekend look like it from the outside — and now you can tell which ones.
Free. No account. Any target.
Then prove you own your own app, and Tzu stops observing and starts testing. AI writes code that trusts the browser: row-level security never switched on, auth missing from an endpoint, keys left in the bundle, storage rules wide open. Tzu finds those, proves they're real, and gives you a fix to paste straight into your coding agent.
Verified assets only.
Actively scanning infrastructure you don't control is unauthorized access, so Tzu is built so it can't. The active engine takes no target parameter at all — it reads the target out of a verified-ownership record, which means there's no request you could craft to aim it somewhere else.
For a target nobody has verified, the output is characterizing, not exploitable: enough to make a business decision, never enough to launch an attack.
Everything Tzu reads passively is what the app publishes to anyone who loads it. That's a real signal, and it's a small slice of what's actually there. The keys in a bundle say a door might be unlocked; only testing your own app says whether it opens.
The engine behind this has been running against real client work for months. Tzu is the self-serve half, and it isn't live yet. Put an address in and you'll be among the first to point it at anything.