Control

Observability

Every request carries a correlation_id, and the Executive Summary shows usage, models, latency and provider health on one screen.

Executive Summary

CardWhat it shows
Usage trendDaily input / output tokens; up to the last 30 days
Model breakdownRequests, token share, spend, latency (avg / p95)
LatencyAverage, p50, p95 (latency_ms)
Top usersTop 5 people by tokens, based on X-FFM-User
Provider healthFailed request rate by model and provider
Personal data protectionEvent and detection counts, by outcome and type

The cards group the usage record the gateway writes for every request by day, model and person. Until at least 20 requests have accumulated, p95 shows "—". The data comes from the GET /admin/dashboard/overview endpoint and is available only to the system_admin role.

Tracing with correlation_id

A request's audit, usage and log records are all linked by the same correlation_id. When you investigate an error or a complaint, this ID takes you to every record for that request.

AUDIT LOGOBSERVABILITY
QuestionWho did what?What happened where?
ScopeAuditing, compliance, permanent recordUsage, latency, where errors occur
SamplingNeverAllowed
Shared keycorrelation_idcorrelation_id

OpenTelemetry and license

BaseLLM can send its traces and metrics to your own monitoring system over OpenTelemetry (OTLP); Prometheus can also read the metrics at /metrics. That address runs on a port separate from the API.

  • Enterprise and trial on: Traces and metrics are sent.
  • Individual off: Nothing is sent; /metrics answers 403.
  • When the license expires stops: Sending stops within a minute.