LangGraph Chatbot with External Tools (Groq & LangChain)

This repository contains a Jupyter Notebook implementing an agentic chatbot using LangGraph, LangChain, and the Groq API (llama-3.3-70b-versatile). The chatbot dynamically decides whether it needs to fetch information from external sources (such as Wikipedia) to answer user questions, executes those tools, and feeds the results back to the LLM to construct a final response.

🚀 Features

• Stateful Conversational Agent: Utilizes LangGraph's state management with an message-reducing state schema to preserve conversation history.
• Dynamic Tool Invocation: Implements conditional routing (tools_condition and ToolNode) allowing the LLM to run external search queries autonomously.
• Groq Integration: Powered by Groq's low-latency, high-performance Llama-3.3-70b model.
• Payload Optimization: Automatically strips out reasoning tokens from intermediate messages to reduce prompt sizes and context window consumption.
• Search Capabilities: Configured to query external knowledge via Wikipedia.

---

📊 Graph Architecture

The conversational agent is structured as a cyclic directed graph (DAG):

graph TD
    START((START)) --> Chatbot[Chatbot Node]
    Chatbot --> Action{Call Tool?}
    Action -- Yes --> Tools[Tool Node: Wikipedia]
    Tools --> Chatbot
    Action -- No --> END((END))

1. START: Receives the user message and appends it to the state.
2. Chatbot Node: Evaluates the messages, prepares the payload (filtering reasoning metadata), and invokes the ChatGroq model bound with tools.
3. Action (Conditional Edge): Evaluates whether the LLM generated tool calls.
   • If yes, routes to the Tools Node.
   • If no, routes to END.
4. Tools Node: Executes the specified tool (Wikipedia search) and routes the output back to the Chatbot Node to continue the loop.

---

🛠️ Prerequisites & Setup

1. Installation

Install the required packages in your Python environment:

pip install langgraph langsmith langchain langchain_groq langchain_community wikipedia "arxiv<2.0.0"

(Note: arxiv is pinned to <2.0.0 in the notebook to ensure compatibility with API wrappers).

2. Environment Variables & API Keys

The notebook is configured to run in a Google Colab environment and expects a Groq API Key stored in Colab Secrets under the key name gorq_api_key.

You can set this up as follows:

• Colab User Data: Add gorq_api_key to your Colab Secrets (the key icon in the sidebar).
• Local Environment: Alternatively, you can set it directly as an environment variable in Python:

  import os
  os.environ["GROQ_API_KEY"] = "your-groq-api-key"
  

---

🔍 Detailed Code Walkthrough

1. Defining the State

The state keeps track of the conversation history. We use the add_messages reducer to append new messages instead of overwriting the state list.

from typing import Annotated
from typing_extensions import TypedDict
from langgraph.graph.message import add_messages

class State(TypedDict):
    messages: Annotated[list, add_messages]

2. Integrating Tools

We wrap the Wikipedia API as a LangChain query run tool:

from langchain_community.utilities import WikipediaAPIWrapper
from langchain_community.tools import WikipediaQueryRun

wiki_wrapper = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=300)
wiki_tool = WikipediaQueryRun(api_wrapper=wiki_wrapper)

tools = [wiki_tool]

3. Setting Up the LLM

We initialize ChatGroq using Llama 3.3 and bind the tools to it so the model knows they are available.

from langchain_groq import ChatGroq

llm = ChatGroq(groq_api_key=groq_api_key, model_name="llama-3.3-70b-versatile")
llm_with_tools = llm.bind_tools(tools=tools)

4. Defining the Chatbot Node

The chatbot function cleanses the messaging payloads by removing reasoning_content keys from previous model turns before invoking the LLM.

def chatbot(state: State):
    messages = []
    for m in state["messages"]:
        if hasattr(m, "additional_kwargs"):
            # Strip reasoning content to keep payloads minimal
            m.additional_kwargs.pop("reasoning_content", None)
            messages.append(m)
    return {"messages": [llm_with_tools.invoke(messages)]}

5. Compiling the Graph

We build the state graph, connect nodes, add conditional routing, and compile the application.

from langgraph.graph import StateGraph, START, END
from langgraph.prebuilt import ToolNode, tools_condition

graph_builder = StateGraph(State)

# Add nodes
graph_builder.add_node("chatbot", chatbot)
graph_builder.add_node("tools", ToolNode(tools=tools))

# Add edges
graph_builder.add_edge(START, "chatbot")
graph_builder.add_conditional_edges("chatbot", tools_condition)
graph_builder.add_edge("tools", "chatbot")
graph_builder.add_edge("chatbot", END)

graph = graph_builder.compile()

---

🏃 Running the Chatbot

You can stream messages through the compiled graph to observe nodes firing in real-time:

user_input = "Who is Arwin"

events = graph.stream(
    {"messages": [("user", user_input)]}, 
    stream_mode="values"
)

for event in events:
    event["messages"][-1].pretty_print()

Sample Output Log:

================================ Human Message =================================
Who is Arwin

================================== Ai Message ==================================
Tool Calls:
  wikipedia (7kw6dabsx)
  Args:
    query: Arwin

================================= Tool Message =================================
Name: wikipedia
Page: Martin Melcher
Summary: Martin Melcher... series of business ventures named Arwin...

================================== Ai Message ==================================
Tool Calls:
  wikipedia (6r7veghsa)
  Args:
    query: Arwin Productions

...

================================== Ai Message ==================================
Arwin is likely referring to Arwin Hawkhauser, a character from the Disney Channel Original Series "The Suite Life of Zack & Cody". He is the hotel's engineer and a friend of the main characters...

---

📄 License

This project is open-source. Feel free to copy, modify, and use it in your own applications!