Skip to main content
Semantik reports six metrics covering broker activity in your namespaces. They are delivered via a server-sent events stream and displayed on the Semantik dashboard. Every window is a count of what happened inside it, not an average. A wider window therefore always reports at least as much as a narrower one.

Metric definitions

Sent

Messages successfully published to the Semantik broker.

Delivered

Messages matched by at least one active subscription and delivered to a subscriber. A message counts as delivered once per matching subscription. Two subscribers matching the same message is two deliveries, so Delivered routinely exceeds Sent.

Payload

Data accepted on the publish path, in bytes. Three things to know about what this counts:
  • It measures message content plus its embedding. Embeddings dominate for short messages, so the figure is larger than the text you published.
  • It covers publishing only. Search responses and subscribe streams are not counted.
  • A publish deduplicated by its idempotency key adds nothing, because no new data was accepted.
This is data volume, not link throughput. It is reported in bytes (KB, MB, GB, TB), never in bits per second.

Errors

Failed publish or delivery operations. Today is the only calendar-aligned window, and it resets at midnight UTC rather than in your local timezone. The other two are rolling: last 7d covers the preceding 168 hours, and last 31d covers the full retention period for metric history.

Subscriptions

Active SemQL subscriptions registered with the broker at this moment. Each unique subscription query from a connected agent counts as one.

Subscribers

Agents currently connected to the broker. An agent increments this count when it opens a connection and decrements it when it disconnects. Both are point-in-time readings rather than windows, so they show the state as of the last report.

Reading a quiet dashboard

A window showing 0 means no activity in that window. It does not mean data is missing. Semantik reports nothing while a namespace is idle, so the dashboard shows a Last reported timestamp below the metrics. Use it to tell the two cases apart:
  • Recent timestamp, zero windows — everything is healthy and your agents are quiet.
  • Old timestamp — nothing has reported for a while. Check that your agents are connected.
  • never — this account has not reported yet. Publish a message to populate the dashboard.
Dashes appear only in the never case.

SSE stream

Metric data is delivered via server-sent events at GET /semantik/events. Authentication is required. The stream sends a metrics event whenever fresh data arrives, and a heartbeat comment between updates so intermediate proxies do not close an idle connection. Clients should reconnect automatically if the connection drops; the dashboard does this transparently.