Observability

Signal Coverage

• Tracing: OpenTelemetry + Tempo/Jaeger.

• Metrics: Prometheus + Grafana.

• Logs: Loki.

• LLM Observability: Langfuse or Helicone.

Operational Guarantees

• End-to-end traceability from user prompt to tool calls.

• Metric baselines for latency, cost, and quality.

• Anomaly detection for retrieval drift and model regressions.