Derived metrics are aggregated time-series counters and histograms Dash0 computes from the raw spans your agents emit. These are not tags on individual spans—they are computed aggregations you query over time. Use them to build dashboards, track token consumption, and measure session characteristics.

For the raw span attributes you can group these metrics by, see Span Attributes.

Metric	Type	What it measures
`dash0.gen_ai.cache_tokens.input`	Counter	Input tokens served from the prompt cache.
`dash0.gen_ai.chats.duration`	Histogram	Per-turn response latency.
`dash0.gen_ai.conversations`	Counter	Sessions started.
`dash0.gen_ai.conversations.duration`	Histogram	End-to-end session duration.
`dash0.gen_ai.tokens.input`	Counter	Input tokens consumed.
`dash0.gen_ai.tokens.output`	Counter	Output tokens generated.
`dash0.gen_ai.tool_calls`	Counter	Tool invocations, with an `error` label for failures.
`dash0.gen_ai.tool_calls.duration`	Histogram	Tool execution time, exposing `_sum`, `_count`, and `_bucket` for percentiles.

Because the metrics carry the same labels as the spans, you can group any of them by repository, user, model, branch, harness, or team. See Span Attributes for the complete list of grouping dimensions.

Note

These metrics are derived from harness telemetry only. The pull request measurements on the Overview and Productivity tabs, such as time to merge, time to first review, and review depth, come from the GitHub integration rather than from spans, and are not exposed as derived metrics you can query here.

Derived Metrics

Further Reading