Runs & Traces
Browse and analyze individual agent executions with full trace details.
What is a Run?
A run (also called a trace) represents a single execution of your AI agent. Each run contains:
- Trace ID - Unique identifier
- Workflow - Which agent/workflow executed
- User Query - The input that triggered the run
- Response - The agent's output
- Duration - Total execution time
- Status - Success, Error, or Running
- Spans - Nested operations within the run
Runs List
Navigate to Operations → Runs to see all executions:
- Flat list view with workflow column
- Sortable by time, duration, status
- Pagination for large datasets
- Quick filters for status and workflow
Filtering Runs
Use filters to find specific runs:
- Time Range - Last hour, 24h, 7d, custom
- Workflow - Filter by specific workflow
- Status - Success, Error, Running
- Search - Search by trace ID or user query
Run Detail View
Click on any run to see the full trace:
Overview
- Run metadata (ID, workflow, timestamp)
- Input and output
- Duration and token usage
- Error details (if failed)
Span Tree
- Hierarchical view of all operations
- Timing waterfall visualization
- Click spans to see details
Span Types
| Type | Description |
|---|---|
llm | LLM API calls (OpenAI, Anthropic, etc.) |
tool | Tool/function invocations |
retrieval | Vector search and RAG operations |
agent | Agent decision-making steps |
chain | Sequential chain executions |
custom | User-defined operations |
Span Details
Each span shows:
- Input - What was passed to this operation
- Output - What it returned
- Duration - How long it took
- Metadata - Model, tokens, cost, etc.
- Errors - Exception details if failed
Bookmarking Runs
Click the bookmark icon to save important runs for later reference. Access bookmarked runs from Operations → Bookmarks.