Playground

Try to break it.

Semantic analysis, not keywords. Four detection ports, trajectory tracking, covert channel identification. Paraphrase the attack however you like.

Intent analysis 4-port detection Trajectory monitoring Measured false-positive rate
Lightweight preview. A deterministic subset of the real detection layer — enough to show semantic intent, trajectory escalation, and which boundary catches what. The full engine (including the cost-gated LLM-judge this preview skips) runs locally once installed.
Reworded injection
Paraphrase the attack however you like — dropping or swapping a word won't help. The engine scores malicious intent, not keywords.
attacker > payload
Ready
Pick a scenario or type your own payload.
Port 1
Input
Port 2
Tool call
Port 3
Tool result
Port 4
Output
Scan detail will appear here.
Measured false-positive rate, published · Strong on known attacks · Lower on novel attacks

This preview runs a deterministic subset of the engine in your browser. Installed, it adds stateful tracking across a whole session, the full bypass-tested corpus, a cost-gated judge for the ambiguous cases this preview does not run, and a hash-chained log on your own disk. Some attacks get through here; that's honest security, not theater. Exact measured rates and methodology at the security page and security.clawmont.com.

What you're seeing vs. the full product

This is a lightweight preview of the real layer, a deterministic subset that demonstrates semantic intent analysis, multi-turn trajectory tracking, and 4-port coverage. The full 4-port engine runs locally once installed in your OpenClaw gateway: deeper analysis, stateful session tracking across every tool call, the full bypass-tested corpus, the cost-gated LLM-judge that escalates the ambiguous cases this preview doesn't run, and a hash-chained audit log.

Some attacks will get through, and some legitimate prompts get wrongly blocked. That's honest security: strong detection on known attack patterns, far lower on novel attacks the engine hasn't seen, and a false-positive rate we measure and publish rather than claim away. The exact measured rates, each with the instrument that produced it, plus full methodology and raw measurements, are on the security page and at security.clawmont.com.

Like what you see?

Every detection layer runs locally inside your OpenClaw gateway. API keys never leave your machine. Full 4-port protection from $9.99/mo.

See pricing Security details Setup guide