Skip to main content

Customer Service

Typical pattern:

  1. Route user prompts through POST /v1/chat/completions.
  2. For RAG/knowledge-base workflows, pre-scan retrieved context via POST /v1/rag/scan.
  3. Log and review blocked events from management/reporting APIs.

Outcome goals:

  • Reduce successful prompt-injection attempts.
  • Enforce consistent policy behavior for support prompts.
  • Improve auditability of unsafe prompt handling.