-
Notifications
You must be signed in to change notification settings - Fork 167
Closed
Labels
Description
Hi, I'm currently using LangSmith with a langgraph.prebuilt.create_react_agent to capture traces. I'm using the Anthropic Claude Sonnet 4.5 model with interleaved thinking enabled. I am able to see user messages, assistant messages, tool calls and tool responses in the traces. However, I'm unable to see any thinking blocks in the traces. I know they exist because I see them in my Streamlit app.
Setup details:
- Tracing enabled via
LANGCHAIN_TRACING_V2=trueenvironment variable - Model initialized with
init_chat_model(model_id, model_provider="bedrock_converse", ...) - Thinking enabled via
additional_model_request_fields={"thinking": {"type": "enabled", "budget_tokens": N}} - Agent created with
create_react_agent(model, tools)
austinmw