# Runs and traces

> Explore an execution and inspect its model requests, tool calls and recorded data.

The execution view brings an agent’s operations and their recorded data into one workspace. Its hierarchy shows which agent, model or tool operation called another; selecting an operation opens its details alongside that hierarchy.

In **Logs**, select a row in **All agent runs** to open an invocation with its model requests and tool calls. In **All traces**, selecting an operation opens that operation and its descendants.

One run covers one agent invocation, including its model requests, tool calls and retries. Further user turns can start new runs in the same conversation. Use [Conversations](/docs/platform/logs/search-and-filters.md#conversations) to inspect those runs together.

[![Support agent invocation beside the selected lookup_order tool's recorded details](/docs/images/platform/run-and-tool.png)](/docs/images/platform/run-and-tool.png)

The run remains visible while the selected tool’s Details tab shows its timing and reported error.

## Inspect an operation

The selected operation’s **Details** tab summarizes its timing, model usage, context and recorded errors. The available sections depend on the operation. **Attributes** contains its recorded attributes, including application-specific fields and structured payloads.

For prompts, responses and tool calls inside a model exchange, open [Thread](/docs/platform/logs/threads.md). For timestamped annotations or captured inputs and outputs, see [Events and inspection](/docs/platform/logs/events-and-inspection.md). These tabs appear when the selected operation contains the corresponding data.

## Show timing and usage

Choose **Show Timeline** in the execution pane to display values beside the operation hierarchy. Switch the timeline between **Duration**, **Token Usage** and **Cost (USD)**. Expanding or collapsing an operation updates the tree and timeline together.

Usage can be reported on an operation or aggregated from its children. The [telemetry reference](/docs/platform/reference/telemetry.md#tokens-and-costs) explains how nested agents and already-aggregated values are handled.

A run’s outcome is derived from its agent-invocation span. A handled error in a child operation does not, by itself, mark the run as failed.

## Continue an investigation

Use **Ask Meta Agent** on the execution or selected operation to ask a question with that activity as context. Its answer links back to the records it used.

To follow connections between operations within a trace or across grouped activity, open [Trajectory View](/docs/platform/logs/trajectories.md).