observability-pydantic-ai-tracing-pydantic-ai-traces-show-empty-input-output-values--f5fca189

THEN Upgrade to Langfuse >=2.57.3. Avoid manually creating or passing a parent trace/span when using the `observeOpenAI` wrapper to allow automatic population of input/output data. After generation, explicitly set the output using `langfuse.trace()`. Customize span names by providing a `name` parameter when creating a span via `langfuse.span()`.

from langfuse import Langfuse
langfuse = Langfuse()

# Instead of manually creating a parent span/span:
# with langfuse.trace(..., name="manual") as parent:
#     parent.span(...)

# Use the observeOpenAI wrapper (automatically creates traces):
from openai import OpenAI
from langfuse.openai import observeOpenAI
client = observeOpenAI(OpenAI())

# After the LLM call, explicitly set the output on the root trace:
response = client.chat.completions.create(...)
langfuse.trace().update(output=response.choices[0].message.content)

# For custom span naming: