This article is auto-synced from its in-app version in Tai.
Analytics shows how your agents are being used — how often they run, how long they take, and how many skill, connector, and knowledge calls they make — across a configurable time window.
When you'd use this
Use Analytics to see where agent activity is coming from and whether it's growing. For example: check which agents ran most last week, see whether a new connector is actually being called, or find out who in your company is using agents the most.
Steps
View activity charts
In the Workbench sidebar, click Analytics (under the Monitoring section).
Set the date range using the start and end date pickers. The default is the last 30 days.
Choose a granularity — Hourly, Daily, Weekly, or Monthly — to control how the over-time charts bucket data.
The cards at the top show totals for the selected period: Agent uses, Total agent time, Skill calls, Connector calls, and Knowledge calls.
Over time charts these five measures across the period, including both total and average agent time per run.
Breakdowns ranks activity by agent, by skill and by connector, and ranks your knowledge bases by how many searches found something in them.
A breakdown chart shows the top 10 entries plus a single Other bar for everything below them, so the bars stay comparable however many agents, skills, connectors, or knowledge bases you have. When a chart is capped this way it gains a View all button — click it for the full ranked list as a table you can search and sort.
Filter by agent
On the Analytics page, use the Agent multi-select to show data for specific agents only.
Agents are listed by name, and the charts update immediately.
Filter by user
On the Analytics page, use the User multi-select to show data for specific callers only.
People are listed by email address and API keys by the key's name, so you can pick the caller you mean without looking up an ID.
The By user section lower down the page ranks callers the same way, in Requests by user and Agent time by user.
Tips and limits
The page reports activity, not spend. For credit consumption, see the Billing center.
Dates are bucketed in UTC, so an evening run west of UTC counts towards the next day.
Average agent time is the mean per run, so a single slow run moves it much more on a quiet day than on a busy one.
Knowledge calls counts how many searches your agents ran. Searches by knowledge base counts, per base, how many of those searches found something in it — not how many results came back, so a search that returns ten passages from one base still counts once. A single search can find something in several bases, so the bars can total more than the tile.
Searches by knowledge base starts empty. It only covers activity from the release that introduced it onwards, so give it a few days of agent runs before reading anything into it. The Knowledge calls tile and the over-time series include your earlier activity as usual.
If a caller can't be matched to a record in your company, the charts and filter show their raw identifier instead. Deactivated users and revoked API keys still show their name.
Everyone sees their own activity. The User filter, the By user charts, and the Overview tiles (active users and sessions over 24h, 7d, and 28d) require the Company Admin role.
Related