Skip to main content
While the API Logs tab shows individual atomic requests, Workflow Runs are designed for complex applications that combine multiple AI tasks into a single workflow. Navigate to Logs in the sidebar and select the Workflow Runs tab.

​
When to Use Workflow Runs

Use Workflow Runs when building autonomous agents, logic routers, or multi-step chains — scenarios where multiple LLM prompts execute to fulfill a single end-user request.

​
Workflow Detail View

Clicking a workflow entry opens a detail panel showing:
  • Workflow name and execution timestamp
  • Total duration and aggregate cost across all steps
  • Execution steps list with per-step status, model, tokens, and cost

​
Trace Debugger

Click View Full Trace to open the interactive trace debugger, which provides a rich visual representation of your workflow execution: The trace debugger includes:
A top-down visual graph showing when each span (step) occurred and how long it took. This helps diagnose blocking code vs fast async operations.
A hierarchical tree of all spans in the workflow. The root node represents the total workflow, and child nodes represent individual steps like API calls, retrieval functions, or model completions.
Click any span to see its detailed metrics: duration, model used, input/output tokens, cost, and any metadata attributes.
A metadata panel showing key-value pairs, node types, routing expressions, and other contextual information for each span.
The trace debugger is especially useful for identifying bottlenecks in your AI workflows — look for spans with unexpectedly high durations in the timeline view.