Skip to main content

Assistant Activity

The Assistant Activity page offers a real-time, end-to-end view of every assistant and tool execution in Deepdesk. It provides a comprehensive audit trail, capturing inputs, outputs, timing, status, and resource usage for all assistant and tool calls.

With this feature, teams can:

  • Monitor, debug, and optimize assistant-driven processes.
  • Trace the full lifecycle of every invocation.
  • Drill down into each step to see how assistants, operators, and tools interact.
  • Identify bottlenecks, errors, or inefficiencies in complex workflows.

Key Concepts​

Trace​

A trace is a single execution flow triggered by a user or system event (such as a new message, agent acceptance, or API call). It includes all assistant and tool calls made as part of that event, forming a complete picture from start to finish.

Step​

A step is an individual operation within a trace, such as:

  • An assistant or operator call.
  • A tool invocation (e.g., call_api, call_knowledge_assist).
  • Nested operations (e.g., an operator calling multiple assistants, which then call tools).

Each step records its inputs, outputs, timing, status, and is linked to its parent operation in the trace.

Example​

  1. An assistant is triggered, starting the trace.
  2. The assistant calls two tools as the first step.
    • Each tool responds individually as a step.
  3. The assistant re-evaluates and calls another tool, send-user-notification.
    • The notification tool sends a message to the user.
  4. The assistant concludes the trace with a response to the original trigger.

Using the Assistant Activity Page​

The Assistant Activity page displays all traces in a chronological list. Each trace shows:

  • Timestamp: When the trace started
  • Assistant/Operator: Which assistant or operator initiated the trace
  • Status: Success, failure, or in progress
  • Duration: Total execution time
  • Steps: Number of individual operations performed

Click on any trace to expand and view its detailed step-by-step breakdown.

Step Details Side Panel​

When you select a specific step within a trace, the side panel displays:

  • Input Parameters: The data passed to the step
  • Output Response: The result returned by the step
  • Timing Information: Start time, end time, and duration
  • Status: Whether the step succeeded or failed
  • Error Details: If the step failed, detailed error information
  • Resource Usage: Token consumption, API calls, or other resource metrics

This granular view helps identify exactly where issues occur and how data flows through your assistant workflows.

Why Use Assistant Activity?​

  • Debugging: Quickly identify where and why assistant workflows fail
  • Performance Optimization: Find slow steps and optimize execution times
  • Audit Trail: Maintain complete records of all assistant interactions
  • Quality Assurance: Verify that assistants are performing as expected
  • Compliance: Track all automated actions for regulatory requirements

Best Practices​

  • Regular Monitoring: Check the Activity page regularly for failed traces
  • Performance Analysis: Look for patterns in slow-performing steps
  • Error Investigation: Use step details to understand and fix recurring errors
  • Workflow Optimization: Identify unnecessary steps or redundant tool calls

Anonymization of Traces​

How Anonymization Works​

To protect sensitive information while maintaining the utility of trace data for debugging and optimization, Deepdesk can anonymize traces by:

  • Replacing personal identifiers with generic placeholders
  • Masking sensitive data fields while preserving data structure
  • Maintaining trace relationships and timing information

Anonymization Complexities & Results​

What Gets Anonymized:

  • Customer names, email addresses, and phone numbers
  • Account numbers and sensitive identifiers
  • Personal information in conversation content

What Remains:

  • Trace structure and step relationships
  • Timing and performance data
  • Error types and system responses
  • Assistant and tool execution patterns

This approach allows teams to analyze assistant performance and debug issues without exposing sensitive customer information.