Usage analytics
The Usage analytics view gives you aggregate metrics across all runs of a flow — how much it costs, how reliably it completes, how fast it responds, and which sub-agents are driving the most activity.
Accessing usage data
There are two entry points:
- Usage tab in the agent graph's right panel — a summary of the last period's key metrics, always visible while you're building
- Full analytics page — click Open Full Analytics from the Usage tab to open the dedicated analytics view with complete breakdowns and charts
Key metrics
Four headline numbers appear at the top of both views, with a time range filter (last 24 hours, 7 days, or 30 days):
| Metric | What it measures |
|---|---|
| Total calls | The number of runs in the selected period |
| Success rate | Percentage of runs that completed without error — the same figure shown on the Flows list card |
| Estimated cost | Aggregated LLM token cost across all runs and sub-agents in the period |
| P95 latency | The 95th percentile end-to-end run time — 95% of runs completed within this duration |
Breakdowns
The full analytics page breaks down usage across three dimensions:
By sub-agent — cost and call volume per sub-agent. Use this to identify which specialists are running most often and whether any single sub-agent is disproportionately expensive. A sub-agent with unexpectedly high cost may be receiving more transfers than intended, or may be making redundant tool calls.
By model — cost and call volume per LLM model used across the flow. Useful when sub-agents use different model overrides and you want to understand the cost contribution of each.
By source — call volume broken down by how runs were triggered: chat, API, scheduled runs, or A2A calls. Use this to understand where your agent's traffic actually comes from.
The time series chart
The full analytics page includes an area chart showing run volume or cost over the selected period. Use it to spot trends — a spike in failures, a sudden increase in cost, or a drop in traffic after a configuration change.
Cost estimates
Estimated costs are based on token usage reported by each LLM provider and standard per-token pricing. Actual billing depends on your provider agreement and may differ from the estimate shown here.
Success rate and the Flows list
The success rate shown here is the same figure displayed on the agent's card in the Flows list — calculated from recent runs in the same rolling window. A flow with a low success rate (shown in amber below 90% and red below 60%) warrants a review of recent executions to identify the failure pattern.