LangChain Zero Shot React Agent uses memory or not?

Question

I'm experimenting with LangChain's AgentType.CHAT_ZERO_SHOT_REACT agent. By its name I'd assume this is an agent intended for chat use and I've given it memory but it doesn't seem able to access its memory. What else do I need to do so that this will access its memory? Or have I incorrectly assumed that this agent can handle chats?

Here is my code and sample output:

llm = ChatOpenAI(model_name="gpt-4",
                 temperature=0)

tools = load_tools(["llm-math", "wolfram-alpha", "wikipedia"], llm=llm)
memory = ConversationBufferMemory(memory_key="chat_history")

agent_test = initialize_agent(
    tools=tools, 
    llm=llm, 
    agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION, 
    handle_parsing_errors=True,
    memory=memory, 
    verbose=True
)

>>> agent_test.run("What is the height of the empire state building?")
'The Empire State Building stands a total of 1,454 feet tall, including its antenna.'
>>> agent_test.run("What was the last question I asked?")
"I'm sorry, but I can't provide the information you're looking for."

Answer 1

It does not. that's indicated by zero-shot which means just look at the current prompt. from here

Zero-shot means the agent functions on the current action only — it has no memory. It uses the ReAct framework to decide which tool to use, based solely on the tool’s description.

I think when you work with this agent type, you should add description to the tool. so that based on the description, llm will infer which tool used. that is the description part of ""zero-shot-react-description". example from same link above:

math_tool = Tool(
    name='Calculator',
    func=llm_math.run,
    description='Useful for when you need to answer questions about math.'
)

When the llm sees the prompt, if it infers that prompt is related to math, it will use the math_tool

If you want to use memory, you should be using chat-conversational-react-description

from langchain.memory import ConversationBufferMemory
from langchain.chat_models import ChatOpenAI

memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
llm = ChatOpenAI(openai_api_key=OPENAI_API_KEY, temperature=0)
agent_chain = initialize_agent(tools, llm, agent=AgentType.CHAT_CONVERSATIONAL_REACT_DESCRIPTION, verbose=True, memory=memory)