Explore logs

Use Logs → Spans when you need to inspect one recorded operation, such as an LLM request, tool call, task, or workflow step. The table combines time and identity fields with model, token, cost, latency, status, and custom-property data so you can move from a broad production question to the exact span behind it.

Choose the right Logs surface

The Logs section contains three related views:

Start with Spans when a dashboard, evaluator, customer report, or known log identifier points to one request. Use View and debug traces when the cause depends on the surrounding workflow or conversation.

Find the spans that matter

1

Set the project and time range

Confirm that you are in the intended project, then choose a range that covers the event. The table displays the timezone beside the Time column.

If no rows appear, widen the range before changing instrumentation or removing filters.

2

Choose useful columns

Open the table settings control to choose the sort field, sort direction, and visible columns. A focused investigation usually starts with:

• Time and Log ID to identify the record
• Customer ID, Thread ID, or another stable identifier for context
• Model, Provider, and Status for routing and failures
• Latency, TTFT, Tokens, and Cost for performance and usage
• Span name, Workflow name, and Custom properties for application context

Hide customer data, API keys, prompt content, and other sensitive columns before sharing screenshots.

3

Filter the table

Select Filter, choose a field, then set its operator and value. Add another filter group when the investigation needs a separate set of conditions.

Starting question	Useful first filters
Which request failed?	Status or error fields, plus model or workflow context
Why did latency increase?	Latency fields, then model, provider, or workflow
Which customer reported this?	Customer, thread, or custom identifiers and the incident time
Which model call was expensive?	Token or cost fields, then model and provider
Where did this application step run?	Span, task, or workflow names, or a custom property

The available fields depend on the surface and the data your application records. Check the returned rows before saving or exporting the result.

4

Save a reusable view

Use the view control to create a named view, add an optional description, apply the filters that define the cohort, and select Save view. Names such as Production errors — checkout or Slow calls — model X make the purpose clear to other investigators.

See Views and saved filters for how to update and reuse views across the platform.

Filter by custom properties

Use custom properties for application-specific dimensions such as environment, tenant, prompt version, release, region, or feature flag. Choose Custom properties in the filter menu, then select the property, operator, and value.

Record stable dimensions during instrumentation so the same cohort can be found in Spans, Traces, Metrics, and Monitors. See Metadata and tags.

Inspect one span

Select a row to open its detail panel. Read the header first: it identifies the span, timestamp, status, request type, and model context.

The available tabs can include:

• Messages for recorded input and output
• Tools for tool definitions, arguments, and results
• Metrics for latency, token, cost, and other recorded measurements
• Metadata for identifiers and custom properties
• Config for model and request configuration

The tabs and fields vary with the span type and what your application recorded.

Change the content format

Use the content-format selector to switch among Pretty, JSON, and YAML:

• Pretty is easiest for reading a conversation or response.
• JSON is useful when checking exact fields and nested values.
• YAML can make long structured content easier to scan.

Changing the representation does not change the stored span.

Act on a production example

The span detail header and overflow menu expose actions for the selected record. Depending on the span, you can:

• Select Run to open the run workflow available for that span.
• Use Assign to to route the span for review.
• Add positive or negative feedback.
• Select Show nearby requests to inspect records around the same time.
• Select Save as Prompt to turn useful content into a reusable prompt.

Work with several spans

Select rows with the table checkboxes to open the bulk-action bar. You can assign the selection, add it to a dataset, or download it. Use filters before a large selection so every selected span belongs to the intended cohort.

For an export with its own name, columns, sampling, and job history, use Retention and export.

Troubleshooting

No spans appear

Confirm the project and time range, clear filters, and send a test request. If the request still does not appear, follow the Tracing quickstart.

Input or output is missing

The application may have disabled content logging, the selected span type may not contain messages, or the content may be outside the project’s retention window. Check the Metadata and Config tabs, then review Retention and export.

Related operations appear as separate rows

Spans are the individual-operation view. Open Traces to read their parent-child execution order. If related operations are not grouped into one trace, add or propagate a parent workflow span.

Next steps