Skip to content

Consumption

Manage → Consumption shows how much the solution consumed, with filters and per-record detail.

Consumption dashboard.

The cards at the top summarise the filtered period:

CardWhat it counts
Filtered recordsHow many measurements fall inside the current slice
Total tokensThe volume of text processed by the models
Total audio timeSeconds of transcription and speech
Total charactersUsed by charges measured in characters
Total bytes, Total files, Total vectorsFiles processed and indexed

A single conversation turn usually produces more than one record — the agent’s answer and the trace narration, for instance, appear separately.

Consumption dashboard filters.

FilterTypical use
PeriodAll time, last 7, 30 or 90 days, or a From/To range
Provider and ModelCompare the cost of different models
AgentFind out which agent concentrates the consumption
Source typeSeparate what came from the agent and what came from trace narration
Charge typeTokens, audio, characters, files
OperationThe kind of call made to the provider
Measurement modeWhether the number was reported by the provider or computed by MarIA
Group byDay, week or month on the timeline

Clear filters returns to the full view.

Below the charts, the Overview, Tokens, Audio, Text, Volume and Records tabs go deeper into each dimension. Records is the most granular one: it lists measurement by measurement, which is what you need when you have to explain a spike.

  1. Filter the period where the rise appeared and group by Day.
  2. Filter by Agent to find where it is concentrated.
  3. Look at Source type: high trace-narration consumption usually points to multi-agent orchestration with too many participants per turn.
  4. Check, in Orchestration, the Max Participant Runs per Turn and the model used for narration.