Monitoring agent analytics

Evaluate agent performance by monitoring analytics such a message volume, latency, and failure rates. These metrics help identify issues early and provide actionable insights into agent behavior and efficiency.

Monitoring all activities

To see a summary dashboard with agent analysis, you must go to the Build agents and tools page. A dashboard with agent analysis is shown on the top.

Agent analytics dashboard
Figure 1. Agent analytics dashboard on the build agents and tools page.

The agent analytics dashboard contains the following data:

  • Total messages: Total number of messages processed by all agents, including messages that failed.
  • Failed messages: Failed messages that are tracked when the agent returns error responses. For example, service connection issues or 500 errors.
  • Latency average: Average time agents take to process a message. This information helps to identify performance issues.

To view detailed analytics for each agent, click View all at the bottom of the dashboard.

Agent analytics page
Figure 2. Agent analytics page.

In addition to the tiles with total messages, failed messages, and latency average of all agents, you can also see a table that shows details about each agent:

  • Name: The name of the agent.
  • Description: The description of the agent.
  • Messages: The total number of messages that were processed by the agent.
  • Failed messages: The number of messages that failed.
  • Latency avg: The average time to process a message by the agent.

Viewing agent traces

To view the status of each message for an agent, click the ellipsis button Overflow menu icon and select View analytics.

Agent analytics trace table
Figure 3. Trace table on agent analytics page.

The trace table shows the following details:

  • Timestamp: The time the message was processed.
  • Trace ID: The unique identifier of the message.
  • Status: The message status, which can be Success or Error.
  • Model: The large language model (LLM) that is used.
  • Latency: The time that was taken to process the message.

Viewing trace details

Analyze traces is important to understand how an agent behaves over time. You can do this analysis through the message status, which can be either Success or Error. When you click the status, a window opens by showing the flow of control and providing insights to help you identify and resolve issues.

Agent analytics trace details
Figure 4. Trace details of the message.

You can access the trace details, for example, to analyze a high latency, which might require adjusting parameters or using faster models.