Understanding Agent Logs

What Are Agent Logs?

Agent Logs provide a centralized location to monitor and analyze the performance of your AI agents within the CRM. This feature consolidates execution activity, conversation context, and step-level details into a single interface, making it easier to understand agent behavior, troubleshoot issues, and enhance performance over time.

Key Benefits of Using Agent Logs

Agent Logs offer several advantages for managing AI agents:

• Centralized Visibility: Review all supported AI agent activity in one place instead of relying on isolated testing views.
• Faster Troubleshooting: Navigate from a broad activity overview into specific conversations and individual execution steps to quickly identify where issues occurred.
• Enhanced Debugging Context: Compare agent responses with the execution timeline to understand how decisions were made.
• Performance Monitoring: Use step-level execution timing to identify slow areas that may need optimization.
• Increased Confidence: Inspect technical details closely when troubleshooting prompts, logic, variables, or tool behavior.

Global Activity Overview

The Global Activity Overview provides a high-level view of recent agent activity, allowing you to quickly identify where to focus your attention. This top-level perspective is ideal for monitoring activity at scale, narrowing down problems, or locating specific interactions for closer inspection.

Common items and actions in this view include:

• Timestamp: Shows when the interaction occurred.
• Agent Name: Identifies which agent handled the interaction.
• Agent ID: Displays the unique ID for the agent involved.
• AI Product: Indicates which supported AI product the activity belongs to.
• Channel: Shows where the interaction took place.
• Status: Displays the current result or state of the interaction.
• Filters: Narrow results by product, agent name, or search by contact.
• Sorting: Reorder results to find the most relevant interactions faster.

Conversational Context & Timeline

After selecting a specific interaction, you can review the conversation alongside an execution timeline. This view helps you understand both the user experience and the logic behind it, answering questions like why the agent responded a certain way or when a tool was involved.

Key features include:

• Reading the interaction in a familiar conversation format.
• Comparing the agent’s response with the steps that led to it.
• Tracing the flow from the original user message through the final response.

In some cases, the execution timeline includes a task checklist that tracks multi-step work, using descriptive labels instead of numeric step indexes. The interface indicates which task is active and which are completed, improving clarity when the agent adjusts plans mid-run.

Granular Step Execution

For precise troubleshooting, the Granular Step Execution view allows you to inspect individual execution steps. This is useful when a conversation appears incorrect or incomplete, helping reveal issues related to logic, timing, data handling, or tool-related actions.

This detailed view may include:

• The model used for the step.
• Latency and execution timing.
• Timestamp details.
• Input and output for the selected step.
• Prompt details, where applicable.
• Additional technical metadata for debugging.

Step output can be reviewed in different formats, such as a raw JSON-style view or a more readable parsed view, and can be copied as JSON for deeper troubleshooting or sharing.