Defining nodes and edges for flexible function calling

Designing Agentic Systems with LangChain

Dilini K. Sumanapala, PhD

Founder & AI Engineer, Genverv, Ltd.

Building a multi-tool workflow

Multiple available tools
- Palindrome
- Historical events
- Wikipedia

Full chatbot workflow

Define workflow functions

Create a stopping function

Chatbot and tools nodes of chatbot workflow.

Define workflow functions

Create a stopping function
- Check for tool calls

Stopping function rules highlighted.

Define workflow functions

Create a stopping function
- Check for tool calls
- End conversation if none

END rules highlighted in workflow

Define workflow functions

Create a stopping function
- Check for tool calls
- End conversation if none
Create a dynamic tool caller
- Return a tool response if tool call present

Call_model rules highlighted.

Define workflow functions

Create a stopping function
- Check for tool calls
- End conversation if none
Create a dynamic tool caller
- Return a tool response if tool call present
- Invoke the LLM with just the chatbot node if no tool calls

Call_model rules highlighted.

Define workflow functions

Create a stopping function
- Check for tool calls
- End conversation if none
Create a dynamic tool caller
- Return a tool response if tool call present
- Invoke the LLM with just the chatbot node if no tool calls
Compile the full graph

Call_model rules highlighted.

Create a stop condition function

from langgraph.graph import MessagesState, START, END


# Use MessagesState to define the state of the stopping function
def should_continue(state: MessagesState):


    # Get the last message from the state
    last_message = state["messages"][-1]


    # Check if the last message includes tool calls
    if last_message.tool_calls:
        return "tools"


    # End the conversation if no tool calls are present
    return END

Create a dynamic tool caller

# Extract the last message from the history
def call_model(state: MessagesState):

    last_message = state["messages"][-1]


    # If the last message has tool calls, return the tool's response
    if isinstance(last_message, AIMessage) and last_message.tool_calls:


        # Return the messages from the tool call
        return {"messages": [AIMessage(content=last_message.tool_calls[0]["response"])]}


    # Otherwise, proceed with a regular LLM response
    return {"messages": [model_with_tools.invoke(state["messages"])]}

Create the graph

workflow = StateGraph(MessagesState)

Create the graph

workflow = StateGraph(MessagesState)


# Add nodes for chatbot and tools
workflow.add_node("chatbot", call_model)
workflow.add_node("tools", tool_node)

Chatbot and tools nodes.

Create the graph

workflow = StateGraph(MessagesState)


# Add nodes for chatbot and tools
workflow.add_node("chatbot", call_model)
workflow.add_node("tools", tool_node)


# Connect the START node to the chatbot
workflow.add_edge(START, "chatbot")

Add START node to chatbot.

Create the graph

workflow = StateGraph(MessagesState)


# Add nodes for chatbot and tools
workflow.add_node("chatbot", call_model)
workflow.add_node("tools", tool_node)


# Connect the START node to the chatbot
workflow.add_edge(START, "chatbot")


# Define conditions, then loop back to chatbot
workflow.add_conditional_edges("chatbot", 
         should_continue, ["tools", END])

Graph with conditions added.

Create the graph

workflow = StateGraph(MessagesState)


# Add nodes for chatbot and tools
workflow.add_node("chatbot", call_model)
workflow.add_node("tools", tool_node)


# Connect the START node to the chatbot
workflow.add_edge(START, "chatbot")


# Define conditions, then loop back to chatbot
workflow.add_conditional_edges("chatbot", 
         should_continue, ["tools", END])

workflow.add_edge("tools", "chatbot")

Full graph workflow.

Adding memory

# Set up memory and compile the workflow
memory = MemorySaver()

app = workflow.compile(
               checkpointer=memory)


display(Image(app.get_graph()
        .draw_mermaid_png()))

Full graph workflow.

Let's practice!

Designing Agentic Systems with LangChain

Defining nodes and edges for flexible function calling

Building a multi-tool workflow

Multiple available tools

Palindrome

Historical events

Wikipedia

Define workflow functions

Create a stopping function

Define workflow functions

Create a stopping function

Define workflow functions

Create a stopping function

Define workflow functions

Create a stopping function

Create a dynamic tool caller

Define workflow functions

Create a stopping function

Create a dynamic tool caller

Define workflow functions

Create a stopping function

Create a dynamic tool caller

Compile the full graph

Create a stop condition function

Create a dynamic tool caller

Create the graph

Create the graph

Create the graph

Create the graph

Create the graph

Adding memory

Let's practice!