Traces

A Trace represents the complete lifecycle of a single request or interaction within the application. It captures every stage of execution, from the initial input to the final output, providing a unified record of how the request was processed.

A trace includes:

• Input received from the user or application.
• Intermediate observations, such as model calls, tool executions, API requests, and workflow activities.
• Outputs generated during execution.
• Execution metadata, including timestamps, latency, token usage, cost, identifiers, tags, and environment information.

Each trace contains one or more observations (also referred to as spans or events) that represent the individual operations performed during execution.

Trace List View

The Trace List View provides a centralized view of all captured traces, enabling users to monitor, analyze, and troubleshoot application executions.

Each trace is displayed as a row in the table with the following information:

Column	Description
Timestamp	Date and time when the trace was created.
Input	Request, prompt, or data submitted to the application.
Output	Response generated by the application or model.
Observation Level	Number and hierarchy of observations associated with the trace.
Latency	Total execution time of the trace.
Tokens	Total number of input and output tokens consumed.
Total Cost	Processing cost incurred for the trace.
Environment	Environment in which the trace was executed (for example, Development, Staging, or Production).
Tags	User-defined labels associated with the trace.
Metadata	Additional contextual information captured during execution.
Score	Evaluation scores assigned to the trace.
Actions	Actions available for the selected trace.

Filter Panel

Use Hide Filters to collapse the filter panel and maximize the space available for viewing trace data.

The panel offers a wide range of filter options to help you precisely narrow down and locate the specific traces relevant to your analysis.

Refresh Options

Configure automatic refresh using one of the following intervals:

• Off
• Every 30 Seconds
• Every 1 Minute
• Every 5 Minutes
• Every 15 Minutes

Trace Detail View

Selecting a trace from the list opens the Trace Detail View, which provides comprehensive information about the selected trace, including execution details, observations, evaluations, and related metadata.

Trace Information

The header section displays key information about the trace.

Field	Description
Trace ID	A unique identifier for the trace that can be used for reference and troubleshooting.
Latency	The total time taken to process the trace from input to output.
Usage Breakdown	A detailed breakdown of token consumption and associated costs.
Session ID	Identifies the session associated with the trace. Selecting the Session ID opens the corresponding session details page.
User ID	Identifies the user associated with the trace. Selecting the User ID opens the corresponding user details page.

Trace Views

The Trace View includes the following tabs for viewing trace information.

• Preview – Displays the trace input, output, and observations in a structured format for quick review.
• Log View – Displays the complete execution sequence, including all observations and workflow activities in chronological order.
• Scores – Displays evaluation scores and assessment metrics applied to the trace and its observations.

Actions

The Trace Detail View provides the following actions.

• Add Item to Dataset – Saves the trace input and output as a reusable dataset item for testing, evaluation, and benchmarking.
• Annotate – Applies predefined scores, labels, or evaluation metrics to a trace or observation.
• Playground – Opens the trace in the Playground for testing and experimentation.
• Fresh Playground – Creates a new Playground using the selected trace without affecting existing configurations.
• Add to Existing – Adds the selected trace to an existing Playground for continued experimentation.
• Add Comment – Adds comments to a trace for collaboration, feedback, or additional context.