Skip to main content
Admin › Monitoring › Usage
The usage dashboard visualizes AI resource consumption in real time.
  • Analyze token usage and request count across multiple dimensions (model, user, agent, group, organization).
Usage dashboard

See token, request, and user status at a glance under Admin > Monitoring > Usage

​
Key Metrics

Key metrics appear at the top of the dashboard as cards. Total Tokens, Total Queries, Total Requests, Users in Period, Active Chats, Models Used, and Avg Tokens/Request show the change against the preceding period of the same length, and Total Tokens / Total Requests include a trend sparkline. Online Now, Cache savings, and Estimated Cost have no change indicator, and no change is shown when the period is All or the preceding period has no data. The Estimated Cost card and the Estimated Cost column in the usage-by-model table appear only when at least one value is entered under Model Pricing in Admin › Settings › Models.
If average tokens/request is abnormally high, excessive system prompts or unnecessary context may be included. Review agent settings.

​
Usage Trend Chart

Visualize usage over time as a line chart.
Usage trend chart

Usage trend chart — token usage line graph

​
Period Selection

With Last 1 day selected, the x-axis switches from dates to hours (HH:00).

​
Chart Series

The token usage trend chart plots up to three series. The cache and embedding series appear only when the period has values for them.

​
Multi-dimensional Analysis

Break usage down by model, user, agent, group, and organization.
  • All charts respond to the top filters (period, agent, model, user, group, organization).

​
Usage by Model

Usage by model — request count with input/cache/output composition bars

Usage by model — request count and input/cache/output composition per model

Each model shows its request count and its input · cache · output · total tokens in a table, and the cache column also carries the hit rate (%). The bar is likewise split into input · cache · output — a wider cache segment means the same work was served more cheaply. With model pricing configured, an Estimated Cost column is added, and when embedding-only models are present the table splits into General models and Embedding models sections. The table aggregates by model, so workspace agents and Agent Flows do not appear as their own rows. To see usage per agent, use the Agents tab in Top Rankings.

​
Top Rankings

Switch dimensions with the tabs in the Top Rankings panel. Each tab shows only the top 20 entries, and the share (%) shown next to each entry is measured against the sum of those 20 entries, not against the overall total.
Top Rankings — Users tab
Ranks users by token usage.

​
Usage Type Classification

Classify what work tokens were used for.
  • The chart shows only types with usage records in the selected period.
Usage by type — treemap showing each type's share as area

Usage by type — rectangle area is the token share

Each rectangle’s area broadly reflects its token share, so it is immediately clear which work drives cost. Very small types are drawn at a minimum size so they do not disappear, however, so do not compare small rectangles by area — read the value on the tile or hover over it instead. Response generation (generation) usually dominates; if background work (title/tag generation) or guardrail / auto_eval grows disproportionately large, it is a signal to review the settings.

​
Activity Heatmap

The request count is plotted on a day-of-week × hour (0–23) grid by color intensity. Hours are bucketed in Korea Standard Time (KST).
Activity heatmap — day of week by hour grid

Activity heatmap — usage concentration by day and hour

Darker cells mean more requests. When usage clusters outside business hours or concentrates on particular days, use it as evidence for adjusting batch job schedules or usage limit allocation.
Only types with actual activity in the period appear in the chart.User activity (counts toward the daily limit)Background tasks (not counted toward the limit)System / monitoring

​
Filtering

Combine multiple conditions to filter data.
  • All filters apply simultaneously.
  • Every filter except Period accepts multiple selections and defaults to selecting everything. The selection is applied when you close the dropdown.
While a filter is partially selected, a Clear Filters button appears next to the filters to restore the full selection at once.
The model, user, and agent dropdowns list only items with usage records (across all time, not the selected period). The group and organization dropdowns list every registered group and organization, so items with no usage can also be selected.
Individual conversation details are available in a separate Monitoring > Conversation Logs tab.

General

Configure usage limits and excess actions

Tracing

Track per-request token usage

Auto-Evaluations

Review response quality evaluation results