operationsJune 2, 2026· 3 min read

Three windows into one cache: dashboard, Prometheus, logs

The built-in dashboard for humans, /metrics for your Grafana, one JSON line per event for your pipeline — same truth, three consumers, zero adapters.

Observability fails in two directions: tools that demand their own stack (another agent, another bill) and tools that emit nothing useful. Crowkis threads it with three native windows aimed at three consumers, all reading the same engine state, none requiring an adapter.

In plain words: Watch it in the built-in dashboard, graph it in your existing Grafana, or read clean JSON logs — whichever your team already does, Crowkis plugs in without new tooling.

Humans get the dashboard: the live verdict feed with per-decision confidence and deciding stage, hit-type breakdowns, per-tenant economics, top queries and top misses, memory and storage pressure. It answers the operator's first question — what is this thing actually doing? — visually and in real time.

adoption is one port change

Four doors in, one cache, and the model only sees genuinely new questions.

Machines get /metrics in Prometheus exposition format and OpenTelemetry signals, so the cache lights up inside whatever Grafana or Datadog estate you already run; alerts ride your existing alerting. Pipelines get structured logs — one JSON line per significant event, no spam, rotation configured — so your log stack ingests without grooming.

The bottom line

The discipline underneath: a significant event logs once, an insignificant one logs never, and nothing requires a log-level safari to make usable. Observability is respect for the operator's attention, encoded.