Last updated: August 15, 2026
Gen AI Metrics
dash0.gen_ai.cache_tokens.input
Total GenAI input tokens served from cache.
View the dash0.gen_ai.cache_tokens.input metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {token} |
| Stability |  |
dash0.gen_ai.cache_tokens.input is a synthetic metric computed
on-the-fly from the ingested span data.
Each span contributes the value of its
gen_ai.usage.cache_read.input_tokens attribute to the per-bucket
sum.
Deprecated PromQL alias: dash0_gen_ai_cache_tokens_input_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.cache_tokens.output
Total GenAI output tokens served from cache.
View the dash0.gen_ai.cache_tokens.output metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {token} |
| Stability | |
dash0.gen_ai.cache_tokens.output is a synthetic metric computed
on-the-fly from the ingested span data.
Each span contributes the value of its
gen_ai.usage.cache_read.output_tokens attribute to the per-bucket
sum.
Deprecated PromQL alias: dash0_gen_ai_cache_tokens_output_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.chats
Counts the number of GenAI chat operations ingested by Dash0.
View the dash0.gen_ai.chats metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {chat} |
| Stability | |
dash0.gen_ai.chats is a synthetic metric computed on-the-fly
from the ingested span data.
Each span with gen_ai.operation.name=chat contributes a single count.
Deprecated PromQL alias: dash0_gen_ai_chats_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.chats.duration
Distribution of GenAI chat operation durations.
View the dash0.gen_ai.chats.duration metric in Dash0
| Property | Value |
|---|---|
| Instrument | Histogram |
| Unit | s |
| Stability | |
dash0.gen_ai.chats.duration is a synthetic metric computed on-the-fly
from the ingested span data.
The duration is the duration of each individual span with
gen_ai.operation.name=chat.
Deprecated PromQL aliases: dash0_gen_ai_chats_duration_seconds_bucket,
dash0_gen_ai_chats_duration_seconds_sum, dash0_gen_ai_chats_duration_seconds_count.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
dash0.gen_ai.conversations
Counts the number of GenAI conversations ingested by Dash0.
View the dash0.gen_ai.conversations metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {conversation} |
| Stability | |
dash0.gen_ai.conversations is a synthetic metric computed on-the-fly
from the ingested span data.
Deprecated PromQL alias: dash0_gen_ai_conversations_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.conversations.duration
Distribution of GenAI conversation durations.
View the dash0.gen_ai.conversations.duration metric in Dash0
| Property | Value |
|---|---|
| Instrument | Histogram |
| Unit | s |
| Stability | |
dash0.gen_ai.conversations.duration is a synthetic metric computed
on-the-fly from the ingested span data.
Deprecated PromQL aliases: dash0_gen_ai_conversations_duration_seconds_bucket,
dash0_gen_ai_conversations_duration_seconds_sum, dash0_gen_ai_conversations_duration_seconds_count.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
dash0.gen_ai.conversations.info
Identity / metadata mapping for a GenAI conversation.
View the dash0.gen_ai.conversations.info metric in Dash0
| Property | Value |
|---|---|
| Instrument | Gauge |
| Unit | 1 |
| Stability | |
dash0.gen_ai.conversations.info is a synthetic metric computed on-the-fly
from the ingested span data. The value is always 1; the info-style
series exists to carry the conversation's metadata labels
(gen_ai.conversation.id, gen_ai.agent.name, gen_ai.provider.name,
gen_ai.request.model, service.name) so display-layer queries can
join other GenAI metrics against it via PromQL group_left.
Mirrors the pattern established by dash0.slo.info (ENG-7637). Not
intended for alerting or trending — use the per-span counters /
histograms (dash0.gen_ai.chats.*, dash0.gen_ai.tokens.*, etc.) for
that.
Deprecated PromQL alias: dash0_gen_ai_conversations_info.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
dash0.gen_ai.subagents
Counts the number of GenAI subagent invocations ingested by Dash0.
View the dash0.gen_ai.subagents metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {subagent} |
| Stability | |
dash0.gen_ai.subagents is a synthetic metric computed on-the-fly
from the ingested span data.
Each span with gen_ai.operation.name=invoke_agent contributes a single count.
Deprecated PromQL alias: dash0_gen_ai_subagents_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.subagents.duration
Distribution of GenAI subagent invocation durations.
View the dash0.gen_ai.subagents.duration metric in Dash0
| Property | Value |
|---|---|
| Instrument | Histogram |
| Unit | s |
| Stability | |
dash0.gen_ai.subagents.duration is a synthetic metric computed on-the-fly
from the ingested span data.
The duration is the duration of each individual span with
gen_ai.operation.name=invoke_agent.
Deprecated PromQL aliases: dash0_gen_ai_subagents_duration_seconds_bucket,
dash0_gen_ai_subagents_duration_seconds_sum, dash0_gen_ai_subagents_duration_seconds_count.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
dash0.gen_ai.tokens.input
Total GenAI input tokens consumed.
View the dash0.gen_ai.tokens.input metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {token} |
| Stability | |
dash0.gen_ai.tokens.input is a synthetic metric computed on-the-fly
from the ingested span data.
Each span contributes the value of its gen_ai.usage.input_tokens
attribute to the per-bucket sum.
Deprecated PromQL alias: dash0_gen_ai_tokens_input_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.tokens.output
Total GenAI output tokens generated.
View the dash0.gen_ai.tokens.output metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {token} |
| Stability | |
dash0.gen_ai.tokens.output is a synthetic metric computed on-the-fly
from the ingested span data.
Each span contributes the value of its gen_ai.usage.output_tokens
attribute to the per-bucket sum.
Deprecated PromQL alias: dash0_gen_ai_tokens_output_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.tool_calls
Counts the number of GenAI tool call operations ingested by Dash0.
View the dash0.gen_ai.tool_calls metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | {tool_call} |
| Stability | |
dash0.gen_ai.tool_calls is a synthetic metric computed on-the-fly
from the ingested span data.
Each span with gen_ai.operation.name=execute_tool contributes a single count.
Deprecated PromQL alias: dash0_gen_ai_tool_calls_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.
dash0.gen_ai.tool_calls.duration
Distribution of GenAI tool call operation durations.
View the dash0.gen_ai.tool_calls.duration metric in Dash0
| Property | Value |
|---|---|
| Instrument | Histogram |
| Unit | s |
| Stability | |
dash0.gen_ai.tool_calls.duration is a synthetic metric computed on-the-fly
from the ingested span data.
The duration is the duration of each individual span with
gen_ai.operation.name=execute_tool.
Deprecated PromQL aliases: dash0_gen_ai_tool_calls_duration_seconds_bucket,
dash0_gen_ai_tool_calls_duration_seconds_sum, dash0_gen_ai_tool_calls_duration_seconds_count.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
dash0.gen_ai.usage.cost
Total USD cost attributed to GenAI spans.
View the dash0.gen_ai.usage.cost metric in Dash0
| Property | Value |
|---|---|
| Instrument | Counter |
| Unit | USD |
| Stability | |
dash0.gen_ai.usage.cost is a synthetic metric computed on-the-fly from
the ingested span data.
Each span contributes the value of its dash0.gen_ai.usage.cost
attribute (USD, float64) to the per-bucket sum.
Deprecated PromQL alias: dash0_gen_ai_usage_cost_total.
This synthetic metric does not expose attributes by default. You can group by span attributes and resource attributes using PromQL aggregations.
It is computed from the telemetry that reached Dash0, so sampling applied before ingestion also reduces what it reports.
Counters always start from zero regardless of query timeframe and represent only the increase over time, providing a clear view of trends without retaining absolute counts.