Stops cost abuse on LLM, SMS and payment APIs before the paid call happens. Every button below sends a real request to this API.
A DeBERTa prompt injection model (protectai v2, 8 bit build, about 256 MB) can run right here in your browser. It downloads only when you ask. The server still enforces with its regex guard; the browser verdict is shown next to it.
Loading this server instance's own counters.
Run a scenario to see the exact request.
Honest note: this demo runs on serverless functions with in memory state and a public demo only key. State is per instance, so budgets can reset on a cold start or when a request lands on another instance. The published benchmarks (6,291 decisions/s, p99 32 ms, 0 failed of 287,403 requests in a Kubernetes rolling restart) come from the Docker Compose and Kubernetes stack with Redis, not from this page. Admin endpoints stay behind login and the API keeps its rate limit on.