ai-agents-output-parsing-error-llm-output-for-a-react-agent-contains-both-a-final-3d06623b

THEN Set the stop token to 'Observation:' when initializing the language model used with the agent. This prevents the LLM from continuing to generate text beyond the tool observation, ensuring it only produces one action or final answer per step. Example: if using ChatOpenAI, pass stop=["Observation:"] in the model constructor or call. For other LLMs, use the appropriate stop parameter.

from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-4o", stop=["Observation:"])
agent = create_react_agent(llm, tools, prompt)