Module attr 

OpenTelemetry GenAI semantic-convention attribute keys.

Constants

AGENT_ID

AGENT_NAME

CACHE_CREATION_INPUT_TOKENS

Input tokens written to a provider-managed cache.

CACHE_READ_INPUT_TOKENS

Input tokens served from a provider-managed cache.

CONVERSATION_ID

ERROR_TYPE

FINISH_REASONS

INPUT_MESSAGES

INPUT_TOKENS

OPERATION

OTEL_NAME

The human-readable span name override consumed by OTel bridges.

OTEL_STATUS_CODE

tracing-opentelemetry special field: sets the OTel span status code ("ERROR" or "OK") when a bridge is attached.

OTEL_STATUS_MESSAGE

tracing-opentelemetry special field: sets the OTel span status description. Used here to carry the exception message, mirroring Python’s capture_exception (observability.py:1407-1411, which calls set_status(description=repr(exception))).

OUTPUT_MESSAGES

OUTPUT_TOKENS

PROMPT_NAME

Low-cardinality prompt name (e.g. for a named/templated prompt).

PROVIDER_NAME

Semantic-convention replacement for SYSTEM, introduced above v1.36.0. Exactly one of the two is emitted, selected by super::ObservabilityConfig::use_latest_experimental_gen_ai_semconv.

REASONING_OUTPUT_TOKENS

Output tokens spent on reasoning.

REQUEST_FREQUENCY_PENALTY

REQUEST_MAX_TOKENS

REQUEST_MODEL

REQUEST_PRESENCE_PENALTY

REQUEST_SEED

REQUEST_STOP_SEQUENCES

REQUEST_TEMPERATURE

REQUEST_TOP_P

RESPONSE_ID

RESPONSE_MODEL

SYSTEM

The provider/system tag, e.g. "openai". Supplied by the client.

SYSTEM_INSTRUCTIONS

Content-capture-gated: the system/instructions prompt, JSON-encoded as [{"type":"text","content":...}] (mirrors observability.py:1444-1448).

TOOL_CALL_ARGUMENTS

Content-capture-gated tool-call arguments (JSON-encoded).

TOOL_CALL_ID

TOOL_CALL_RESULT

Content-capture-gated tool-call result (JSON-encoded).

TOOL_DEFINITIONS

Content-capture-gated: the request’s tool list, JSON-encoded (mirrors _tools_to_dict, observability.py:1388-1391).

TOOL_DESCRIPTION

TOOL_NAME

TOOL_TYPE