Skip to main content

observability.prompt_turn_metrics

Prompt-turn completion metrics emission for agent-runtimes.

Emits OTEL metrics to the Datalayer OTEL service on every completed prompt turn. The exporter is configured lazily and reads:

  • per-request user JWT token from transport route Authorization headers
  • DATALAYER_OTEL_SERVICE_NAME for service.name resource attribute
  • DATALAYER_OTLP_METRICS_URL / OTEL_EXPORTER_OTLP_METRICS_ENDPOINT (explicit metrics endpoint override)
  • DATALAYER_OTLP_URL / OTEL_EXPORTER_OTLP_ENDPOINT / DATALAYER_OTEL_RUN_URL (base OTLP endpoint)

extract_bearer_token​

def extract_bearer_token(auth_header: str | None) -> str | None

Extract a bearer token from an Authorization header value.

extract_jwt_token​

def extract_jwt_token(*header_values: str | None) -> str | None

Extract JWT token from common auth header value forms.

Supports:

  • Authorization: Bearer <jwt>
  • Authorization: token <jwt>
  • raw JWT values (e.g., X-External-Token)

decode_user_uid​

def decode_user_uid(token: str | None) -> str | None

Extract the Datalayer user_uid from a JWT token.

Follows the platform convention: payload.user.uid first, then payload.sub.

extract_user_id_from_jwt​

def extract_user_id_from_jwt(user_jwt_token: str | None) -> str | None

Best-effort JWT claim extraction without signature verification.

extract_identity_hints​

def extract_identity_hints(
    identities: list[dict[str, Any]] | None
) -> tuple[str | None, str | None, str | None]

Return (user_id, provider, jwt_token) from identities payload.

PromptTurnMetricsEmitter Objects​

class PromptTurnMetricsEmitter()

Emits prompt-turn metrics through OTLP HTTP exporter.

record_prompt_turn_completion​

def record_prompt_turn_completion(*,
                                  prompt: str,
                                  response: str,
                                  duration_ms: float,
                                  protocol: str,
                                  stop_reason: str,
                                  success: bool,
                                  model: str | None,
                                  tool_call_count: int,
                                  input_tokens: int | None = None,
                                  output_tokens: int | None = None,
                                  total_tokens: int | None = None,
                                  user_id: str | None = None,
                                  user_provider: str | None = None,
                                  identities_count: int | None = None,
                                  user_jwt_token: str | None = None,
                                  agent_id: str | None = None) -> None

Emit prompt-turn completion metrics.

This function is safe to call from request paths: failures are swallowed and logged without affecting prompt execution.