Dashboard Overview
The Analytics dashboard gives you a high-level view of how your LLM applications are performing from both a safety and an operational perspective. Four key metrics are displayed at the top of the page:| Metric | Description |
|---|---|
| Total Requests | The number of LLM interactions captured during the selected time range |
| Issue Rate | The percentage of traces where at least one safety issue was detected |
| Average Latency | The mean response time across all traced requests |
| Total Cost | The estimated spend across all models and providers |
Time Ranges
All metrics and charts update based on the selected time range. Choose from:- 24h — Last 24 hours
- 7d — Last 7 days
- 30d — Last 30 days
- 90d — Last 90 days
Switching the time range refreshes every chart and metric on the page. Use shorter ranges for debugging recent incidents and longer ranges for trend analysis and reporting.
Charts
The Analytics dashboard includes six charts that cover request volume, safety, cost, and performance.Request Volume
Request Volume
An area chart showing the number of LLM requests over time. Use this to identify traffic spikes, quiet periods, and overall growth trends.
Clean vs Issues Breakdown
Clean vs Issues Breakdown
A bar chart comparing the number of clean traces to traces with detected issues. This gives you a quick visual indicator of your overall safety posture during the selected period.
Cost by Model
Cost by Model
A breakdown of estimated cost by model. See which models are driving the most spend and how cost distributes across your stack.
Cost is stored internally in cents for precision and displayed in dollars on the dashboard. This avoids floating-point rounding issues in aggregations.
Latency Percentiles
Latency Percentiles
Latency distribution shown at three percentiles:
Monitor p95 and p99 to catch latency regressions before they affect the majority of users.
| Percentile | Meaning |
|---|---|
| p50 | Median latency — half of all requests are faster than this |
| p95 | 95th percentile — only 5% of requests are slower |
| p99 | 99th percentile — tail latency experienced by the slowest 1% |
Issue Distribution by Type
Issue Distribution by Type
A breakdown of detected issues by type (prompt injection, jailbreak, toxicity, PII, bias, hallucination). Use this to understand which safety risks are most prevalent in your application.
Model Comparison
Model Comparison
Compare performance across models on key dimensions: request volume, average latency, cost per request, and issue rate. This helps you evaluate model choices and identify underperforming configurations.
Next Steps
Alerts
Set up automated notifications when safety or performance metrics cross thresholds.
Reports
Generate detailed reports for compliance and stakeholder review.