Send multi-modal input to Logfire

[DouweM]

Description

Logfire (and other OTel consumers) currently only receive kind: image/video/audio/document-url (but not the URL) or kind: binary-content (but not the bytes):

pydantic-ai/pydantic_ai_slim/pydantic_ai/messages.py

Lines 329 to 335 in f955890

	def otel_event(self) -> Event:
	if isinstance(self.content, str):
	content = self.content
	else:
	# TODO figure out what to record for images and audio
	content = [part if isinstance(part, str) else {'kind': part.kind} for part in self.content]
	return Event('gen_ai.user.message', body={'content': content, 'role': 'user'})

To render the LLM conversation accurately and aid in debugging, we should send the actual data along.

Note that URLs are typically not passed directly to the LLM, but downloaded by PydanticAI and sent as bytes. We could consider sending both the URL and the data to Logfire, especially as the URL may not always be available. We may need a size limit though.

References

No response

[Kludex]

This is something we'll likely be doing on our offsite in 2 weeks. :)

[DouweM]

Partially addressed in #1592. We don't send binary content yet, and don't follow the URLs to send the actual image/etc (not that we definitely should...).

@Kludex I'll assign you to decide whether to close this or do a follow-up.

[Kludex]

This is done.