Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.bentolabs.ai/llms.txt

Use this file to discover all available pages before exploring further.

A run is one raw recorded agent run, exactly as it landed in Bento: the bottom of the stack, before any analysis or issue. The dashboard Runs page lists them with charts on top. The traces group reads the runs; the analytics group reads the totals and timeline above them. Once Bento analyzes a run it becomes a trajectory, which is where findings and issues come from. This page is the layer underneath. A run is a trace: a request your agent handled, captured whole. It’s made of spans, the steps inside it like an LLM call or a tool call, and carries a full body (the complete payload) and links to any issues it contributed to. A run’s status is error if at least one span reported an error, otherwise ok. The Runs table shows, per run:
ColumnWhat it is
Run IDThe run’s identifier, what you pass to traces get.
Top level spanThe outermost step that names the run.
Model(s)The models the run called.
Linked issuesIssues this run contributed to.
# spans · Tokens · Cost · Total timeThe run’s rollups.
The charts come from analytics: a summary of totals (total runs and errors) and a timeline histogram of total/error counts bucketed by hour or day.

​
Commands

bentolabs traces --help        # the runs themselves
bentolabs analytics --help     # the totals and timeline over them
CommandWhat it does
traces listList runs. Filter by status, tag, model, finding kind, and a lookup query.
traces get <run-id>Open one run: metadata, rollups, status.
traces get-body <run-id>The run’s full payload.
traces list-spans <run-id>The steps inside the run.
traces get-span <span-id>Open one span by id.
traces list-issues <run-id>The issues this run contributed to.
analytics trace-summaryTotals over a window: total runs, errors.
analytics trace-timelineBucketed total/error counts for the chart.

​
Browse runs

List the most recent runs: 
bentolabs traces list --limit 20 --output table
Filter to what matters. --status is ok or error; --tag, --model, and --finding-kind repeat to pass several: 
bentolabs traces list --status error --finding-kind tool_error --finding-kind timeout --output table
Look up a run by its run ID or trace UUID with --q, and bound the window with --start-time / --end-time: 
bentolabs traces list --q <run-id> --start-time 2026-05-01T00:00:00Z --end-time 2026-05-30T00:00:00Z
Open one run: 
bentolabs traces get <run-id>
Read its steps, then open the one you care about. To list only failed steps, pass the span-level --status ERROR (uppercase: span status, not run status): 
bentolabs traces list-spans <run-id> --status ERROR --output table

bentolabs traces get-span <span-id>
Read the full payload, or see what the run is linked to: 
bentolabs traces get-body <run-id>

bentolabs traces list-issues <run-id>

​
Read the charts

The two analytics commands power the totals and the histogram. Both require --start and --end (ISO datetimes) and take the same filters as the list, so the numbers match the table. Totals over a window: 
bentolabs analytics trace-summary --start 2026-05-01T00:00:00Z --end 2026-05-30T00:00:00Z
The timeline histogram. --bucket is hour (default) or day: 
bentolabs analytics trace-timeline --start 2026-05-01T00:00:00Z --end 2026-05-30T00:00:00Z --bucket day
Scope the charts with the same filters as the table: 
bentolabs analytics trace-summary --start 2026-05-01T00:00:00Z --end 2026-05-30T00:00:00Z --status error --model gpt-4o

​
See also

Trajectories

What a run becomes after Bento analyzes it.

Issues

The tracked problems your runs roll up into.

Findings

The observations Bento extracts from these runs.