ReqLLM.
View Source
Builds the scalar gen_ai.* / server.* / error.* attribute maps
emitted on GenAI client spans, from ReqLLM request lifecycle metadata.
Shared between the live bridge (ReqLLM.OpenTelemetry) and the
dependency-free mapper (ReqLLM.Telemetry.OpenTelemetry) so both emit
identical payloads. See ReqLLM.OpenTelemetry for the end-to-end flow.
start/1 runs on [:req_llm, :request, :start] and returns roughly:
%{
"gen_ai.provider.name" => "openai",
"gen_ai.operation.name" => "chat",
"gen_ai.request.model" => "gpt-5",
"gen_ai.request.temperature" => 0.7,
"server.address" => "api.openai.com",
"server.port" => 443,
"req_llm.request_id" => "2184"
}terminal/1 is merged on top at :stop (response id, finish reasons,
usage, embedding dims, OpenAI extension fields, streaming TTFC).
exception/1 and exception_event/1 cover the error path.
Keys are binary spec names; the live bridge atomizes them at the adapter
boundary. nil and empty-list values are dropped so OTel backends never
see blank-but-present attributes.
Summary
Functions
Returns the normalized cost breakdown map for langfuse.observation.cost_details,
or nil when no cost info is present. Result is suitable for Jason.encode!/1.
Returns the span error-status hint for stop events. nil for success.
Resolves a error.type string from request metadata. Falls back to
"_OTHER" when nothing is recognizable.
Builds attributes added on :exception events. Includes error.type.
Returns the exception event payload (exception.type, exception.message).
Returns nil for empty strings and nil, otherwise returns the value.
Used by metric and attribute builders to drop blank-but-present fields.
Resolves the GenAI provider name from request metadata, falling back to the
model's provider when metadata.provider is absent.
Returns the requested model id, e.g. "gpt-5".
Builds GenAI span start attributes from request lifecycle metadata.
Returns gen_ai.response.time_to_first_chunk as seconds, or nil for
non-streaming requests or streams that never observed a content chunk.
Builds the additional attributes that become available at request stop.
Functions
Returns the normalized cost breakdown map for langfuse.observation.cost_details,
or nil when no cost info is present. Result is suitable for Jason.encode!/1.
Returns the span error-status hint for stop events. nil for success.
Resolves a error.type string from request metadata. Falls back to
"_OTHER" when nothing is recognizable.
Builds attributes added on :exception events. Includes error.type.
Returns the exception event payload (exception.type, exception.message).
Returns nil for empty strings and nil, otherwise returns the value.
Used by metric and attribute builders to drop blank-but-present fields.
Resolves the GenAI provider name from request metadata, falling back to the
model's provider when metadata.provider is absent.
Returns the requested model id, e.g. "gpt-5".
Builds GenAI span start attributes from request lifecycle metadata.
The returned map uses binary attribute names as defined by the OpenTelemetry GenAI semantic conventions. Callers that need atom-keyed maps (e.g. the live bridge) atomize at the adapter boundary.
Returns gen_ai.response.time_to_first_chunk as seconds, or nil for
non-streaming requests or streams that never observed a content chunk.
Builds the additional attributes that become available at request stop.
These are merged on top of the start attributes already on the span.