Screen every prompt and completion for prompt injection, jailbreaks, PII, and secret leakage. An OpenAI-compatible proxy that runs entirely on your own infrastructure. No prompt ever leaves your box.
We've sent your download link and quick-start. It's on its way.
↓ Download nowWant it wired into advisory, policy, assurance, risk & inventory with a full control plane? See the Cyber Suite →
Bidirectional, in-line, on-prem — the same detection engine that ships inside the AI Cyber Central suite, as a standalone you can run today.
Direct and indirect injection, role-play and refusal-suppression jailbreaks — normalised past base64 and homoglyph obfuscation before scanning.
Redact emails, IDs, keys and tokens before they reach the model — and catch secrets on the way back out.
Screen completions for harmful content and exfiltration channels before the answer reaches your app.
Start in monitor mode to tune with zero risk, then flip to enforce — allow, redact, rewrite, or block.
The firewall makes no outbound call except to your model provider. No analytics, no phone-home. Map every control to the frameworks your auditors already use.