Interactive Playground

Type or paste text below to see how opensentry's Tier 0 pipeline detects prompt injection attacks. The analysis runs entirely in your browser — nothing is sent to any server.

How it works

opensentry runs a Tier 0 pipeline entirely in your browser:

1. L0 Front Gate — checks input size and detects flooding (repeated-character runs)
2. L1 Normalization — strips invisible/control characters, folds confusable Unicode, collapses whitespace
3. L2 Statistics — script-mixing detection, entropy analysis, special-token scanning, decode-and-rescan of encoded payloads
4. L3 Structural Regex — 12 regex patterns covering instruction override, role spoofing, template forgery, exfiltration lures, jailbreaks, and more
5. Scoring — evidence is aggregated via noisy-OR and compared against flag/block thresholds

Tier 1 (local ML) and Tier 2 (remote guard) are available when you install the full package — they add semantic/paraphrase attack detection on top of these deterministic heuristics.