Monitoring and feedback
Track how your agents are used, and review the ratings people give their answers.
SPARK's Monitoring area, under Settings, is where you keep an eye on usage across the workspace. It has three views: Agent Usage, Data Usage and Workflow Usage. This page focuses on Agent Usage, the dashboard for how much your agents are used, what they are doing, and how people rate their answers.
Open the Agent Usage page
- In the left sidebar, open the Settings tab.
- In the settings sidebar, under Monitoring, select Agent Usage.
The page opens with the subtitle Monitor agent consumption, tokens, tools, skills and models. Everything on it is scoped to the current workspace.
Set the time range
A row of time-range tabs at the top right drives every chart and table on the page: Last 7d, Last 30d, Last 90d, Last 6m, Last year, YTD and All time. It opens on Last 30d. Trend badges compare the selected range against the previous period of the same length.
Agent consumption
The first section, Agent consumption, leads with three KPI cards:
- Agent calls: how many times your agents ran in the period, with a vs previous period trend.
- Raw tokens: the total input and output tokens your agents processed, also with a trend.
- MTD calls: calls so far this month, against a progress bar reading of 1k / month.
Below the cards, an Agent calls chart plots Calls per day with a running MTD calls line and a dashed 1k / month line. The Agent breakdown section that follows has a chart stacking the top agents by call volume (the rest are grouped as Other) and a paged per-agent table listing each agent with its Calls, Tool calls, Skill calls, Raw tokens, Cache read (tokens served from the model provider's prompt cache) and Rated bad, the share of its rated answers that got a thumbs down.
Organisation-wide limits and quotas, for tokens, runs, documents, seats and storage, are set and tracked under Settings > Organisation > Limits & Quotas, not on this page.
Model breakdown
The Model breakdown section shows which models your agents used. A Daily model calls chart tracks usage over time, a Model breakdown chart compares models, and a table lists each Model with its Tier, Calls and Raw tokens. Calls on automatic model selection are grouped under NGT Auto rather than the model each turn ran on (see Models). A turn that failed is counted under the model it was meant to run on, or under Other when no model is known.
Tools and skills
The Tools & skills section ranks the most-used capabilities: Tool calls (top 10) and Skill calls (top 10), each as a chart and a table of Name and Calls.
Feedback
The Feedback section, folded until you open it, lists the ratings people gave agent replies in this workspace with the thumbs up and thumbs down under each answer (see Start a thread). Each row shows the Date, User, Agent, Rating (Good or Bad), Category, Comment and Status. Filter by rating (All, Good, Bad) or by agent (All agents), and use Load more for older ratings. With none yet, it reads No feedback yet.
Click a row to open Feedback details, which shows who gave the rating, why, and whether NGT has reviewed it, with Open thread to jump to the conversation. The status reads New until NGT has Captured it.
Feedback is reviewed by the NGT team and shapes how SPARK improves. Whether your users' feedback, and the thread it relates to, is shared with NGT is an organisation setting, Help improve Spark, under Settings > Organisation > General.
Activity log
The Activity log at the bottom is a per-call record, collapsed by default. Expand it to browse individual agent calls, newest first. A search box (Search agent, user, status or model) filters the list, and Load more pages through older entries.
Each row is one agent call, with Time, Agent, User (calls made by the system rather than a person show as System), Model, Status and Raw tokens. Click a row to open Log details (One agent call: who made it, the model it used and what it consumed.), which adds the model Tier, Prompt mode, a Thread link back to the conversation when there is one, the Input tokens, Output tokens, Reasoning and Cache read tokens, the Tool calls and Skill calls the call made, and its Duration.
The other Monitoring views
The same Monitoring area holds two more dashboards:
- Data Usage: totals and growth for records and storage across your data objects.
- Workflow Usage: workflow run counts, success and failure totals, and monthly limits.
See Monitoring in the administration guide.