A protective gateway in front of your real LLM API key. Every request flows through the AgentShield Security LLM - which acts as a security analyst, producing a full threat report (risk score, severity, attack chain, recommended action) - before being forwarded. Your real key never leaves the server.
Point your existing SDK at AgentShield instead of the provider. Every request flows through the security firewall automatically - no code changes beyond the base URL.
agk-…
virtual key in your app
Send a request through the gateway using one of your virtual keys.
Every request flowing through the gateway. L1 = regex shield, L2 = Security LLM judge. Refreshes every 3 seconds.
Ask Security LLM directly: "is this request safe?" Bypasses regex and shows the raw verdict from your trained model.
Point your app at the gateway and use a virtual key. Your real provider key stays on the server.
The Security LLM continuously attacks itself in a safe sandbox. Run the full attack suite (prompt injections, jailbreaks, encoding tricks, harmful content, role confusion, goal hijacking) and get a security score with TP/FP/FN/TN breakdown.
Ask about your API-key activity - spend, blocked attacks, virtual keys, money saved, anomalies. Every answer comes from live data, powered by the Security LLM.