🛠️ Building a LangChain ReAct Agent with Memory and Tools

LangChain's structured agents enable powerful interaction patterns, especially when paired with memory and tools. In this post, we'll create a conversational AI assistant that can:

Remember past interactions
Answer real-time questions
Use tools like current time and Wikipedia search

🛠️ Features

Uses ConversationBufferMemory to track conversation history
Supports external tools for dynamic responses
Handles parsing errors gracefully

1from dotenv import load_dotenv
2from langchain import hub
3from langchain.agents import AgentExecutor, create_structured_chat_agent
4from langchain.memory import ConversationBufferMemory
5from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
6from langchain_core.tools import Tool
7from langchain_openai import ChatOpenAI
8import datetime
9from wikipedia import summary
10
11load_dotenv()
12
13def get_current_time(*args, **kwargs):
14  now = datetime.datetime.now()
15  return now.strftime("%I:%M %p")
16
17def search_wikipedia(query):
18  try:
19      return summary(query, sentences=2)
20  except:
21      return "I couldn't find any information on that."
22
23tools = [
24  Tool(name="Time", func=get_current_time, description="Useful for when you need to know the current time."),
25  Tool(name="Wikipedia", func=search_wikipedia, description="Useful for when you need to know information about a topic."),
26]
27
28prompt = hub.pull("hwchase17/structured-chat-agent")
29llm = ChatOpenAI(model="gpt-4o")
30
31memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
32
33agent = create_structured_chat_agent(llm=llm, tools=tools, prompt=prompt)
34
35agent_executor = AgentExecutor.from_agent_and_tools(
36  agent=agent,
37  tools=tools,
38  verbose=True,
39  memory=memory,
40  handle_parsing_errors=True,
41)
42
43initial_message = "You are an AI assistant that can provide helpful answers using available tools.
44If you are unable to answer, you can use the following tools: Time and Wikipedia."
45memory.chat_memory.add_message(SystemMessage(content=initial_message))
46
47while True:
48  user_input = input("User: ")
49  if user_input.lower() == "exit":
50      break
51
52  memory.chat_memory.add_message(HumanMessage(content=user_input))
53  response = agent_executor.invoke({"input": user_input})
54  print("Bot:", response["output"])
55  memory.chat_memory.add_message(AIMessage(content=response["output"]))

🧠 Key Components

Tools

Two simple Python functions:

get_current_time: gets the current time.
search_wikipedia: searches and returns a brief summary using Wikipedia.

Memory

1ConversationBufferMemory(memory_key="chat_history", return_messages=True)

This allows the agent to maintain memory across multiple turns.

Agent Setup

We use create_structured_chat_agent() for cleaner tool interaction:

1agent = create_structured_chat_agent(llm=llm, tools=tools, prompt=prompt)

Executor

The AgentExecutor connects everything and maintains state:

1agent_executor = AgentExecutor.from_agent_and_tools(...)

💬 Example Conversation

User: What time is it?
Bot: It is 04:17 PM.

User: Tell me about Python programming.
Bot: Python is a high level programming language... (Wikipedia summary)

User: What did I ask earlier?
Bot: You asked about the current time and Python programming.

← Previous: ReAct Agent with Tool

Next: MCP →