Skip to main content

Dashboard - Monitor Agentic Flow Executions and Trends in ConvyAI

Overview​

The Dashboard of Agentic Orchestrator shows how the Agentic Flows of your tenant are performing. Use it to spot executions stuck in a wait, to check whether your processes are completing, and to see how duration, token consumption and errors change from one period to the next. Every figure opens Monitoring already filtered on the executions it counts, so you can go from a number to the runs behind it.

Open it from the Dashboard entry of the Agentic Orchestrator menu in ConvyAdmin. It is available on tenants with the Agentic Orchestrator module enabled, to the roles that access Monitoring: agenticorchestrator_admin, agenticorchestrator_user and the higher user levels.

The figures cover all the flows of the tenant. To analyze a single flow, open Flow Report.

The page has two zones: Active, about what is running now, and Trends, about what happened in a period.

Active zone​

The Active zone shows the current state of the executions. It does not depend on the selected period.

CardWhat it shows
In progressExecutions started and not yet ended, including those waiting.
WaitingExecutions paused in a Delay or Wait For Event node, split by how long they have been waiting: < 1 h, 1-24 h, > 24 h.

Waiting is a subset of In progress. The > 24 h band is highlighted when it is greater than zero: use it to spot executions waiting for an event that may never arrive.

The zone refreshes automatically every minute, and a chip shows the time left before the next refresh. There is no manual refresh.

Click In progress to open Monitoring on the running and waiting executions. Click Waiting to open Monitoring on the waiting executions, whatever their band.

The Trends zone analyzes the executions that ended in the selected period, as stated by its subtitle. Executions still open are not counted here: you find them in the Active zone.

Choose the period​

Use the selector in the header of the zone. All the cards show the same period.

PresetPeriodCompared with
Last 24 hoursThe last 24 hours.The 24 hours before.
Last 7 daysThe last 7 days, today included.The 7 days before.
Last 30 daysThe last 30 days, today included.The 30 days before.
Last 90 daysThe last 90 days, today included.The 90 days before.
This monthFrom the first day of the month until now.The same interval of the previous month.
Last monthThe previous calendar month.The month before it.

The default is Last 7 days. Next to the selector, the page shows the exact dates of the period and of the comparison period. Custom periods are not available on the Dashboard.

The comparison period always has the same duration as the current one, so the day or hour in progress does not skew the comparison. Dates follow the time zone of your browser.

Charts group executions by hour for Last 24 hours, by week for Last 90 days and by day for the other presets. The current hour, day or week is marked as partial.

The zone refreshes automatically every 5 minutes. To refresh it earlier, use the refresh action: at least one minute must pass between two refreshes. The period and flow you choose are kept when you refresh the page, and are lost when you close the browser tab.

Read the cards​

Each card compares the period with the previous one through a delta. The delta is marked as favorable or unfavorable only when a direction has a meaning, as indicated below. When the comparison period is zero, the delta cannot be calculated and shows -. When no execution ended in the period, cards show 0.

CardWhat it showsDelta
ExecutionsExecutions ended in the period.Direction and size only: more executions is neither good nor bad.
Completion ratePercentage of executions that completed, with the percentage ended in error (Errors) and stopped by a user (Stopped) below. The three add up to 100%.Percentage points. Favorable when it increases.
DurationMedian duration (p50) of completed executions, with the p95 below. Includes the time spent waiting for external events.On the p50. Favorable when it decreases.

A stop is a voluntary interruption and does not count as an error. Duration is calculated only on completed executions, because for runs that ended in error or were stopped it does not represent the time the process takes.

What p50 and p95 mean​

The Dashboard uses two percentiles to describe durations and token consumption, because an average can be distorted by a few very long or very heavy executions.

  • p50 is the median: half of the executions are below this value and half above. It describes the typical execution.
  • p95 is the value that 95% of the executions stay below: only 1 execution in 20 is longer or heavier. It describes the slow or heavy cases.

For example, with a p50 of 2m and a p95 of 10m, a typical execution takes about 2 minutes, while the slowest ones take 10 minutes or more. A p95 much higher than the p50 means that a few executions behave very differently from the others, and it is worth looking into them in Monitoring.

Tokens per execution​

Shows how many Input, Output and Thinking tokens an execution consumes on average, with the delta, the p95 per execution and the total of the period for each type. The three types are not summed, because input and output tokens are priced differently and the billing of thinking depends on the model. Totals are abbreviated (for example 412k): hover to see the exact value. For tokens, a decrease is favorable.

Only executions that ran at least one AI Agent node enter the average, so flows without agents do not lower it. Executions that ended in error or were stopped count too, because they consumed tokens anyway. If no execution ran an AI Agent node in the period, the card shows No AI Agent executions in this period.

Executions by flow​

A table of the 10 flows that rank highest on the criterion you choose in Top 10 by. Use it, for example, to find the flows that consume the most tokens or fail the most.

ColumnDescription
FlowName of the flow. Renamed or deleted flows with executions in the period stay visible.
ExecutionsExecutions ended in the period.
Completion ratePercentage of completed executions.
ErrorsExecutions ended in error, not counting stopped ones.
Duration p50, Duration p95Duration of completed executions. - if there are none.
Input per execution, Output per executionAverage tokens of executions that ran an AI Agent node. - if there are none.

Top 10 by offers Executions (default), Errors, Duration p50, Duration p95, Input per execution and Output per execution. Use the control next to it to switch between highest first (default) and lowest first. A flow with no value on the chosen criterion is not listed. The table shows no delta and its columns cannot be sorted.

When more than 10 flows have executions, the last row, Other flows, sums up the remaining ones (executions, completion rate and errors) and the footer reads Top 10 of N flows.

Each row has two actions: open Monitoring filtered on that flow, or open Flow Report for that flow. The footer also has a Flow Report link to open the page without selecting a flow. Click the number in Errors to open Monitoring on the executions of that flow that ended in error.

Executions over time​

A stacked bar chart of the executions ended in each interval, split into Completed, Errors and Stopped. Use it to spot peaks or drops. Hover a bar to see the interval, the executions for each outcome and the total. Empty intervals appear as zero. The chart does not show the comparison period.

Click a bar to open Monitoring on that interval. Click a segment to also filter on the corresponding outcome.

Errors by node type​

A pie chart of the executions ended in error, grouped by the type of node where they failed, with the total in the header. Use it to understand whether your weak point is integrations, AI or something else. Node types use the names of the designer, for example HTTP Request, Wait For Event and AI Agent.

  • The chart shows the 5 most frequent types. The others are grouped in Others.
  • Not attributed groups executions in error with no associated node, and always comes last.
  • Errors that do not stop the execution, such as those of a node set to continue on error, are not counted.
  • The chart shows no comparison. When there are no errors, it shows No errors in this period.

Click the total number of errors to open Monitoring on the executions that ended in error. Slices are not clickable.

Open Monitoring from the Dashboard​

The Dashboard counts executions by end date, while Monitoring filters by start date by default. Links from the Trends zone therefore open Monitoring with Filter period by set to End time, so the number you clicked matches the executions listed.

ElementExecutions listed in Monitoring
In progressRunning and waiting.
WaitingWaiting.
Monitoring action of a flow rowExecutions of the flow in the selected period.
Errors of a flow rowExecutions of the flow in the selected period that ended in error.
Bar of Executions over timeExecutions that ended in that interval.
Segment of a barAs the bar, limited to the corresponding outcome.
Total of Errors by node typeExecutions that ended in error in the selected period.