Analytics
Workspace analytics
One dashboard for every agent and workflow in the workspace. Volume, success rate, latency percentiles, cost breakdown and error analysis — all filterable, all exportable.
What you'll learn
- What every panel on the dashboard means
- How to filter by agent, workflow, status and time range
- How latency percentiles and cost breakdowns are computed
- How to export the view for reporting
The dashboard at a glance
The top row is your KPI strip; everything below is a chart or breakdown that drills into it.
- 1
Total executions
Run count over the selected window, with a sparkline and a delta versus the prior window. - 2
Success rate
Percentage of runs that finished without error or rejection. Trend line shows movement over time. - 3
Average latency
Mean wall-clock duration. Drill into the latency panel for P50, P95 and P99. - 4
Estimated cost
Sum of LLM token spend plus integration billing across all runs. Trend versus prior window.
Charts
- 1
Executions over time
Bar chart by day, week or month. Stack by status to see failures alongside success volume. - 2
Success rate over time
Line chart of pass percentage. Annotated with agent version releases so you can see the effect of a deploy. - 3
Agent and workflow popularity
Top-N agents and workflows by volume. Useful for spotting the workhorse and the long tail. - 4
Error rate
Failure percentage over time, broken out by error class — tool error, LLM error, guardrail block, timeout.
Latency percentiles
The latency panel reports P50, P95 and P99. P50 is your median experience; P95 and P99 are your tail. A widening gap between P50 and P99 usually means slow tool calls or unbounded LLM responses — drill into a P99 run from the panel to see which spans dominate.
Cost breakdown
Cost is decomposed three ways. By agent or workflow tells you which workloads spend the most. By tool surfaces expensive integration patterns (paid APIs, vector search). By model highlights where a cheaper model could substitute without changing behavior — confirm with an A/B in Eval before switching.
Filtering and export
- 1
Time range
Pick 7d, 30d, 90d, or a custom window. The dashboard re-aggregates instantly. - 2
Scope filters
Narrow by agent, workflow, status or channel. Filters apply to every panel together. - 3
Export
CSV for spreadsheet analysis, PDF for board decks. Exports include the active filters and time range.
Frequently asked questions
- How fresh is the dashboard?
- Counts and rates lag live traffic by under a minute. Cost lags by up to five minutes because token billing rolls up in batches.
- Are cost numbers exact?
- They are estimates based on provider pricing and reported tokens. They match within a few percent of your provider invoice — use them for sizing, not for accounting.
- Can I share a dashboard view?
- Yes. Every filtered view has a permalink. Save common views as named tabs so the team always lands on the same starting point.
- Where can I see per-run details?
- Click any chart point or table row to drill into Monitor with the filters pre-applied. From there you can open any individual trace.