Skip to main content

Overview

The Tracing feature in AgentOS provides deep observability into your agent and team executions. Every run generates a trace containing detailed information about what happened, how long it took, and what resources were used.
  • Traces capture the complete execution flow of agents and teams
  • Each trace contains spans representing individual operations (LLM calls, tool executions, etc.)
  • Each span contains token usage (when relevant), latency, and error information at a glance
  • You can filter and search traces to debug issues or analyze performance
Prerequisites: Your AgentOS must be connected and active with tracing enabled. If you see “Disconnected” or “Inactive,” review your connection settings.

Accessing Tracing

  • Open the Tracing section in the sidebar.
  • Click the Refresh button to sync the latest traces.

Filtering Traces

Use the filtering options to quickly find the traces you need:

View Mode

  • By Session - Group traces by conversation session to see all activity within a session
  • By Run - View individual execution runs to analyze specific interactions

Time Range

  • Quick filters - Select from preset ranges: last hour, last 24 hours, or last 7 days
  • Custom range - Pick specific start and end dates for precise time-based filtering
Traces date filter options

Trace Views

Click on any trace to view its execution breakdown. You can switch between two visualization modes:

Tree View

The Tree view displays spans in a hierarchical structure, showing the parent-child relationships between operations. This is useful for understanding how teams delegate to agents and how agents invoke tools.

Waterfall View

The Waterfall view visualizes spans on a time axis, showing when each operation started, how long it took, and which operations ran in parallel. This helps identify bottlenecks and optimize performance.