Playground

Paste a prompt, tool output, or agent message. The same scanner that runs on every event ingested into agentwach flags prompt-injection, jailbreaks, leaked secrets, and destructive commands.

Try an example:
88 / 20,000 chars

Scan results

Run a scan to see flagged patterns.

How this works in production

When your agents send events to agentwach, this same rule engine runs on every payload — prompts, tool outputs, and message bodies. Hits are recorded as warnings and can trigger guardrail stops, Slack alerts, or webhook calls.

Start scanning your agents →