AI Agent Workbench
Agent Activity View
Learn how to track, debug, and analyze your agent’s operations using the Activity view
What is the Activity View?
The Activity view provides real-time visibility into your agent’s operations, allowing you to track execution details, debug issues, and analyze performance. This powerful monitoring interface helps you understand exactly how your agent is processing tasks and interacting with external systems.
Key Components of the Activity View
Thread History
Thread History
The left panel shows your agent’s conversation threads, including:
- User information: Who initiated the conversation
- Creation timestamp: When the thread started
- Action count: Number of actions the agent took
- Token usage: How many tokens were consumed
- Session ID: Unique identifier for the conversation
Operation Graph
Operation Graph
The central area displays a visual representation of the operations your agent has executed:
- Nodes: Individual operations like “Search Profiles” or “Retrieve Company Posts”
- Connections: Shows the sequence and relationship between operations
- Visual hierarchy: Displays how operations build upon each other
Execution Logs
Execution Logs
The right panel provides detailed execution logs, including:
- API calls: Exact parameters sent to external services
- Response data: Full JSON responses from operations
- LLM interactions: Complete model inputs and outputs
- Error messages: Details when operations fail
Using the Activity View
When your agent isn’t behaving as expected, the Activity view helps you pinpoint exactly where things went wrong:
1
Select the Problematic Thread
In the left panel, find and click on the thread where the issue occurred
2
Examine the Operation Graph
Look at which operations were executed and in what order to identify unexpected paths
3
Inspect Detailed Logs
In the right panel, expand the log entries to see exact inputs, outputs, and any error messages
4
Identify Root Causes
Look for incorrect parameters, unexpected responses, or logic errors in the agent’s decision-making
Pay special attention to LLM Response logs, which show how the AI interpreted instructions and decided what operations to use.
Understanding Log Types
The Activity view displays several types of log entries, each providing different insights:Operation Calls
Operation Calls
When you see “Calling [Operation]” in the logs (like “Calling Search Profiles by Criteria”), you’re seeing:
- The exact operation being executed
- Input parameters the agent provided
- The response data returned
LLM Responses
LLM Responses
LLM Response logs show you the raw model output, including:
- The model’s reasoning process
- Tool selection decisions
- Parameter choices
- Interpretation of data
System Events
System Events
System events capture agent lifecycle events like:
- Thread creation
- Message additions
- Task completion
- Error events
Best Practices for Using the Activity View
- Review regularly: Check the Activity view after making changes to your agent to ensure it’s working as expected
- Store important threads: Keep track of threads that demonstrate either successful patterns or problematic issues
- Compare before and after: Use the Activity view to compare agent behavior before and after optimization
- Share specific examples: When collaborating with team members, reference specific thread IDs to discuss exact behavior
- Document findings: Record insights from the Activity view to inform future agent improvements