Zammad-AI

Appearance

Prometheus Metrics ​

Zammad-AI exposes Prometheus metrics for the backend API, triage flow, answer generation, and Kafka processing.

Metrics Endpoint ​

The application exports metrics on the standard Prometheus /metrics endpoint.

The metrics server is started during application startup when prometheus.enabled is true.

The default port is 9090.

Exported Metrics ​

HTTP Metrics ​

Collected by the FastAPI middleware in the backend service:

  • zammad_ai_http_requests_total
  • zammad_ai_http_request_duration_seconds

These metrics are labeled by HTTP method, resolved route path, and status code.

Triage Metrics ​

Collected while the triage service processes tickets:

  • zammad_ai_triage_run_duration_seconds
  • zammad_ai_triage_runs_in_progress

Answer Metrics ​

Collected while the answer service generates responses:

  • zammad_ai_answer_run_duration_seconds
  • zammad_ai_answer_runs_in_progress

Kafka Metrics ​

Kafka processing metrics use the zammad_ai_kafka_* prefix and are emitted by the Kafka middleware layer.

Business counters derived from Kafka processing are also exported:

  • zammad_ai_kafka_events_total
    • Counts only main-topic Kafka events that pass payload parsing and request/action filtering.
    • Retry-topic events and filtered events are not counted.
  • zammad_ai_kafka_ticket_outcomes_total
    • Counts Kafka-driven ticket outcomes with labels category, action_type, and outcome.
    • action_type values are normalized to ai_answer, static_answer, no_action, or unknown.
    • outcome values currently include answer, shared_draft, manual, and aborted_with_error.
    • When a ticket aborts before a triage result exists, category="unknown" and action_type="unknown" are used.

Configuration ​

Prometheus is configured in config.yaml under the prometheus section:

yaml
prometheus:
  enabled: true
  port: 9090

Environment variables use the ZAMMAD_AI_PROMETHEUS__ prefix.

Local Development ​

The local compose stack includes a Prometheus container and a Grafana container.

  • Prometheus scrapes host.docker.internal:9090 every 15 seconds.
  • Grafana is available on port 3000.
  • Prometheus persistence is stored in the prometheus_data volume.