Deferred + grounded replies We acknowledge interactions within 3 seconds, run the RAG pipeline, and return follow-up messages that cite your knowledge base before GPT fallback.
Telemetry & admin alerts Monitor command mix, latency percentiles, KB vs GPT sourcing, and trigger admin notifications for unanswered or failed commands.