AI Observability Explorer - Query LLM & AI Trace Data

Use the Explorer tab of AI Observability to query your LLM and agent traces when the Overview dashboard does not answer your question. For prerequisites, see the AI Observability overview.

To open the tab, select More > AI Observability in the side navigation, then select the Explorer tab.

Run a query

The Explorer adds trace fields. Each trace field holds a total for the whole trace. You can use them in the filter, in aggregations, and in Order by. The Trace Explorer does not have these fields.

Field	Value for the trace
`trace.llm_call_count`	Number of LLM calls
`trace.tool_call_count`	Number of tool calls
`trace.distinct_tool_count`	Number of different tools called
`trace.input_tokens`	Sum of input tokens
`trace.output_tokens`	Sum of output tokens
`trace.total_tokens`	Sum of input and output tokens
`trace.estimated_total_cost`	Sum of the estimated cost of all LLM calls
`trace.max_llm_duration_nano`	Duration of the slowest LLM call, in nanoseconds

Type a filter in the filter bar, for example trace.total_tokens > 1000.
Select Run Query, or press Cmd+Enter (Ctrl+Enter on Windows and Linux).

The filter bar suggests only keys and values from GenAI spans that match your current filters. For the filter syntax and the aggregation options, see Query Builder.

To run the query again at an interval, turn on auto-refresh in the time picker.

Views

All views use the same query. The Explorer opens in Trace View. Column changes are saved in your browser only.

trace.* totals depend on the view:

Trace View adds up every span of the trace, even spans outside the selected time range.
List View, Time Series, and Table count only spans inside the selected time range. For example, trace.total_tokens adds up only the tokens of those spans.

Trace View

Trace View shows one row for each AI trace, which is a trace with at least one LLM, tool, or agent span. Each row shows totals such as tokens, cost, and call counts. By default, the table shows the trace fields that exist in your data. Trace View columns and Order by show these fields without the trace. prefix.

Select a row to open the trace in Trace Details. To open it in a new tab, hold Cmd (Ctrl on Windows and Linux) and select the row. To show only the LLM spans of the trace, select the LLM quick filter in Trace Details.

AI Observability Explorer in Trace View, with the quick filters panel on the left and one row for each AI trace — *Trace View in the AI Observability Explorer*

Use the two controls above the table:

Order by: sort the traces by a trace field, in ascending or descending order. The default is last_activity_time (desc). You can sort by fields that the table does not show as columns, such as last_activity_time.
Options: open the Edit columns drawer to show, hide, or reorder columns.

The trace_id column always shows, but you can drag it to a new position.

Edit columns drawer in Trace View, with a search box and the list of added fields — *The Edit columns drawer opens from Options*

List View

List View shows one row for each span, with the newest spans at the top. Select a row to open the trace of that span. The default columns are timestamp, service.name, name, duration_nano, http_method, and response_status_code.

Drag a column header to reorder the column. Drag its edge to resize the column. To remove the column, open its header menu and select Remove column. The timestamp column stays first.

Time Series

Time Series shows the query results as a chart over time. The Y-axis shows milliseconds when every query aggregates span duration, and a plain number otherwise. To change the unit, use the Y-axis unit selector above the chart. To export the chart data, select the download icon at the top right of the chart. To see exact cost values, use the Table view.

Table

Table shows the aggregated results of the query in a grid. To download the results, select the download icon. Choose csv or jsonl. Select Export.

Common tasks

Task	Steps
Find the most expensive traces	In Trace View, set Order by to `estimated_total_cost (desc)`.
Find failed tool calls	In List View, run the filter `gen_ai.tool.name EXISTS AND has_error = true`.
Chart token usage by model	In Time Series, set the aggregation to `sum(gen_ai.usage.input_tokens)` and group by `gen_ai.request.model`.
Compare cost by model	In Table, set the aggregation to `sum(signoz.gen_ai.usage.tokens.cost)` and group by `gen_ai.request.model`.

Next steps

AI Observability overview: see cost, tokens, latency, and tool calls on one dashboard.
Trace Details: inspect the spans of one trace.
Model pricing: set the price for each model.
Attribute Mapping: map other attribute names to the GenAI names.
LLM Observability integrations: instrument OpenAI, Anthropic, LiteLLM, CrewAI, and other frameworks.

Get Help

If you need help with the steps in this topic, please reach out to us on SigNoz Community Slack. If you are a SigNoz Cloud user, please use in product chat support located at the bottom right corner of your SigNoz instance or contact us at cloud-support@signoz.io.