it looks like Phoenix isn’t tracing the internal tool call when running agents that use the new use_responses_api=True setup.
from langchain.agents import create_agent
from langchain_openai import ChatOpenAI
model = ChatOpenAI(model_name="gpt-4.1", temperature=0.7, use_responses_api=True)
web_search_tool = {"type": "web_search_preview"}
agent = create_agent(model=model, tools=[web_search_tool])
message = {
"role": "user",
"content": [
{"type": "text", "text": "could you check upcoming events in Madrid?"}
],
}
response = agent.invoke({"messages": message})
When this runs, the tool invocation (web_search_preview) executes correctly,
but Phoenix doesn’t trace it — no spans or events are appearing for the tool call.
Could someone confirm if this is expected behavior or if we need to update the tracing integration for use_responses_api?
