Skip to main content

​
Override Span Input/Output

You can explicitly control what appears as a span’s input/output in the trace UI. Use OpenTelemetry AI semconv attributes on the current span, inside your @workflow-decorated function.
  • Set SpanAttributes.TRACELOOP_ENTITY_INPUT and/or SpanAttributes.TRACELOOP_ENTITY_OUTPUT.
  • Pass these via the attributes argument of update_current_span (not keywordsai_params).
  • Values must be JSON‑serializable strings, so use json.dumps(...).
  • Keep any custom metadata in keywordsai_params["metadata"] as needed.
import json
from opentelemetry.semconv_ai import SpanAttributes
from keywordsai_tracing.main import KeywordsAITelemetry
from keywordsai_tracing.decorators import workflow

k_tl = KeywordsAITelemetry()
client = k_tl.get_client()

@workflow(name="update_attributes_test")
def update_attributes_test(input: str):
    # Override what is shown as the span input (and optionally output)
    force_set_attributes = {
        SpanAttributes.TRACELOOP_ENTITY_INPUT: json.dumps({
            "args": [],
            "kwargs": {"text": "hiiiiiii"}
        }),
        # Optional: also override output
        # SpanAttributes.TRACELOOP_ENTITY_OUTPUT: json.dumps("Some desired output")
    }

    client.update_current_span(
        attributes=force_set_attributes,
        name="update_attributes_test",  # optionally rename the current span
        # Keep metadata here if you need it in the trace
        keywordsai_params={"metadata": {"test": "test"}},
    )

    some_desired_output = "Some desired output"
    return some_desired_output

if __name__ == "__main__":
    update_attributes_test("Some input")
Use update_current_span(attributes=...) to override the span’s displayed input/output. keywordsai_params do not map input or output to span attributes; they’re for metadata and KeywordsAI-specific options. After overriding, the trace view will display your custom values for the selected span.