all_message_events truncation

[goradawid]

Question

Unable to prevent truncation of all_messages_events in Pydantic Logfire

Issue Description

I'm using PydanticAI with Logfire for monitoring my LLM conversations. I'm working with very long user messages and need to see them completely in Logfire, but I'm consistently encountering truncation:

"logfire.truncated" : [ 0: "all_messages_events", ]

What I've Tried

• Set OTEL_ATTRIBUTE_VALUE_LENGTH_LIMIT and OTEL_SPAN_ATTRIBUTE_VALUE_LENGTH_LIMIT environment variables to large values (10M characters)
• Used InstrumentationSettings(event_mode="logs") as recommended in the documentation:

  from pydantic_ai import Agent
  from pydantic_ai.agent import InstrumentationSettings
  import logfire
  
  # Configure logfire
  logfire.configure()
  
  # Create instrumentation settings with event_mode='logs'
  instrumentation_settings = InstrumentationSettings(event_mode="logs")
  
  # Create agent with these settings
  agent = Agent(
      'google-gla:gemini-2.5-flash-preview-04-17',
      output_type=KnowledgeGraph,
      deps_type=KnowledgeGraphDeps,
      instrument=instrumentation_settings,
  )

pydantic-ai==0.1.8
logfire==3.14.1

Is there any configuration option or approach I'm missing that would prevent the truncation of these large messages? I need to capture the complete conversation for debugging purposes.
Thanks for any help or guidance!

[alexmojaki]

There was a bug here which meant that all_messages_events was being dropped entirely when it was too long, instead of being truncated. That's fixed, meaning the situation is somewhat better now. The JSON string is the thing that gets truncated, so it will sometimes leave valid JSON, sometimes not. We know this isn't ideal and we do want to work towards (1) smarter truncation and (2) allowing storing the full value without truncation.