LLM Guardrails: Five Gates Every Production Request Should Pass

AI teams deploying LLMs in production face a problem most tutorials skip: the model is only half the system. The other half is safety infrastructure — guardrails that inspect every request and response.

Five guardrails fire on every nRouter request from day one:

1. Prompt-injection detection — six attack classes blocked at HTTP 400, costing $0
2. PII redaction — 21 personal-data families masked, conversation-aware
3. Secret scanning — API keys, tokens, high-entropy strings caught
4. Abuse blocking — leet-speak + Unicode normalization, 403 guardrail_blocked
5. Response scanning — buffered, scanned, delivered; leaks withheld

Pre-call blocks cost $0 (inspection runs before credit reservation); post-call blocks incur provider cost. One policy hierarchy (Org > Team > Key), under 50ms overhead, on every plan.

Full technical catalog: https://nrouter.ai/product/guardrails