Analytics
Track a project's request volume, latency, errors, tokens, and cost over time.
The Overview page shows how a project is doing over a period: how many requests it served, how fast, how many failed, and what the model calls cost. It combines trace data with the OTLP metrics that your SDKs report.
Select Overview in the project sidebar. In telemetry.dev, a request is one trace.
Choose a period
Select 7d, 30d, or 90d. The default is 30 days. Select the refresh button to load new data.
The period cannot be longer than the project’s retention. For example, a project with 14-day retention shows 14 days when you select 30d, and the page says “Retention limits this view to 14 days.”
Read the summary cards
| Card | Value |
|---|---|
| Requests | Number of traces |
| LLM cost | Total calculated model cost in USD |
| p95 latency | 95th percentile of trace duration |
| LLM tokens | Input tokens plus output tokens |
| Error rate | Traces with an error status, as a percentage of all traces |
Each card compares the period with the period of the same length just before it. For p95 latency and Error rate, a decrease counts as an improvement.
A comparison appears only if the project’s retention covers both periods. For example, the 30-day view needs at least 60 days of retention. With 14-day retention, the 7-day view has a comparison and the 30-day view does not.
Read the charts
The page has these charts:
- Requests over time: traces per day.
- LLM cost over time: model cost per day, in USD.
- LLM tokens over time: input and output tokens per day, with cache-read, reasoning, and text, image, or audio totals when your telemetry reports them.
- p95 latency and Error rate: daily values for the period.
- Latency distribution: how many traces fall into each duration range, with the p50, p95, and p99 durations marked.
- LLM cost by provider: each provider’s share of model cost.
- Slowest operations: the 5 operations with the highest p95 span duration. Select one to list its spans.
- Top users: the 5 users with the most traces. Select one to list their traces.
Days with no data show as zero, so each chart covers the full period.
Break traces down by a dimension
The breakdown chart shows traces by model at first. Select By Model, By Provider, By Environment, By User, or By Framework to change the grouping.
The chart shows up to 8 groups, ordered by cost and then by number of traces.
Read metric series
Metric series shows a small chart for each OTLP metric series that the project received in the period. The page shows the first three. Select View all metrics to see the rest; if the list was cut short, the button reads View top 100 metrics.
telemetry.dev picks how to combine values from the metric type:
| Metric type | Value shown |
|---|---|
Histogram with unit s or ms |
p95 |
| Gauge | Average |
| Other types | Sum |
For a duration histogram, telemetry.dev estimates the p95 from the histogram’s bucket bounds and counts.
The charts use hourly points for the 7-day period and daily points for longer periods.
The page shows at most 100 series. If the project has more, it shows the series with the most data points and says that some series were omitted.
When the page is empty
If the period has no traces and no metrics, the page shows Go to Setup. Select it to connect an application. Read Connect an application.