1_basic_chatbot_tutorial.md

LangGraph Tutorial Summary: Build a Basic Chatbot

Setup

• pip install -U langgraph langsmith
• Requires access to an LLM (OpenAI, Anthropic, Gemini, etc.)

Core Components

State

• Defined as a TypedDict: class State(TypedDict): messages: Annotated[list, add_messages]
• Uses Annotated[list, add_messages] to append messages (not overwrite)

StateGraph

• Core LangGraph construct for building state machines
• Created with graph_builder = StateGraph(State) using your state schema

Nodes

• Functions that take the current State and return an updated State
• Added with: graph_builder.add_node("chatbot", chatbot)

Edges

• Define control flow between nodes
• Entry edge: graph_builder.add_edge(START, "chatbot")

Types & Reducers

• TypedDict defines state schema
• Annotated[...] attaches a reducer function like add_messages (reducers define how state updates: append, replace, etc.)

LLM Integration

• A node typically invokes the LLM using the current messages: def chatbot(state: State): return {"messages": [llm.invoke(state["messages"])]}

Compile & Run

Compile Graph

• graph = graph_builder.compile()

Run Chatbot

• graph.stream({"messages": [{"role": "user", "content": user_input}]})

Explanation of key lines

return {"messages": [llm.invoke(state["messages"])]}

• state["messages"]: current chat history
• llm.invoke(...): calls the LLM to generate a reply
• [ ... ]: wraps response in a list (required by add_messages)
• {"messages": [...]}: returns a state update with new messages

graph.stream({"messages": [{"role": "user", "content": user_input}]})

• Starts the graph execution
• Sends user_input as a message
• Streams updates as nodes run and state changes