Guidance Needed on Reasoning Models Integration for Phoenix

Hey Phoenix team! 👋 I'm instrumenting reasoning models (OpenAI o1, DeepSeek Reasoner, gpt-oss, etc..) that return separate reasoning/thinking content alongside the final answer. Looking for guidance on the right way to represent this in traces. What I tried first: Added reasoning_content as a custom field in output messages:

  output_messages = [{
      "role": "assistant",
      "content": "final answer here",
      "reasoning_content": "internal reasoning/thinking...",  # 
  Custom field
      "tool_calls": [...]
  }]

Issue: Phoenix doesn't display reasoning_content in the formatted Output Messages tab (only visible in Attributes/raw JSON), which makes sense since it's not in the OpenInference semantic conventions. Current workaround im usin: Split into two separate assistant messages:

  output_messages = [
      {
          "role": "assistant",
          "name": "reasoning",
          "content": "internal reasoning/thinking..."
      },
      {
          "role": "assistant",
          "content": "final answer here",
          "tool_calls": [...]
      }
  ]

This displays both in the Output Messages tab, which works well for visibility. Questions: 1. Is adding message.reasoning_content to the OpenInference spec something being considered? (I see llm.token_count.completion_details.reasoning exists for token tracking) 2. Is the two-message workaround the recommended approach, or is there a better pattern? 3. Should I open an issue/PR to discuss adding reasoning content support to the spec? Thanks for the amazing tool! 🙏