How to add an initial greeting with no input

[francisjervis]

Checked other resources

• I added a very descriptive title to this issue.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.

Example Code

from langgraph.graph import StateGraph, END
from langgraph.graph.message import add_messages
from langchain_core.messages import AIMessage, HumanMessage, ToolMessage
from typing import Annotated

from typing_extensions import TypedDict

from langchain_core.runnables import chain, RunnableLambda

class State(TypedDict):
    # Messages have the type "list". The `add_messages` function
    # in the annotation defines how this state key should be updated
    # (in this case, it appends messages to the list, rather than overwriting them)
    messages: Annotated[list, add_messages]


graph_state = StateGraph(State)

welcome_message = AIMessage(content="Welcome to the hello, world! state machine.")


def emit_greeting() -> AIMessage:
    print("emit_greeting")
    return welcome_message




def handle_greeting_node(state):
    print("handle_greeting_node")
    print(state)
    state.add_message(welcome_message)
    print(state)
    return state




def handle_user_input(state):
    print("handle_user_input")
    print(state)


graph_state.add_node("handle_greeting_node", lambda state: handle_greeting_node(state))
# graph_state.add_node("handle_user_input", handle_user_input)
graph_state.set_entry_point("handle_greeting_node")
# graph_state.add_edge("handle_greeting_node", "handle_user_input")
graph_state.add_edge("handle_greeting_node", END)
app = graph_state.compile()

app.invoke()

Error Message and Stack Trace (if applicable)

Traceback (most recent call last):
File "scratch_4.py", line 51, in
app.invoke()
TypeError: Pregel.invoke() missing 1 required positional argument: 'input'

Description

It appears to be impossible to set up a LangGraph agent to send an initial message to the user from the start node. This means that the user is always presented with a "blank screen" which is unacceptable in production. This is due to "input" being required when it should be optional.

System Info

langgraph 0.0.66
python 3.11
mac os 14.5

[hinthornw]

If you want to invoke it with an empty state, I think you can invoke it with an empty version of your state:

app.invoke({}) # Empty input

If this doesn't satisfy your requirements, could you clarify why? Thanks!

[francisjervis]

If you want to invoke it with an empty state, you can invoke it with an empty version of your state:

app.invoke({}) # Empty input

If this doesn't satisfy your requirements, could you clarify why? Thanks!

This does not work. I get "langgraph.errors.InvalidUpdateError: Must write to at least one of ['messages']"

[hinthornw]

app.invoke({"messages": []}) # Empty input

then?

It depends on your graph state.

Or could add a state variable

class State:
    said_hello: bool
    ...

that you toggle on once you've sent the intro sequence

[francisjervis]

AttributeError: 'dict' object has no attribute 'add_message'

Of course, that error is being introduced because there is no way to set up a LangGraph agent to send an initial message to the user from the start node. If there were, I wouldn't be trying to change the state manually.

[hinthornw]

@francisjervis , the error you are seeing is because "state" in handle_greeting_node is of type State, which is a typed dict.

It doesn't have a method called add_message.

The add_messages (plural) function is a reducer

The following code seems to be what you're looking for?

from langgraph.graph import StateGraph, END
from langgraph.graph.message import add_messages
from langchain_core.messages import AIMessage, HumanMessage, ToolMessage
from typing import Annotated

from typing_extensions import TypedDict

from langchain_core.runnables import chain, RunnableLambda

class State(TypedDict):
    # Messages have the type "list". The `add_messages` function
    # in the annotation defines how this state key should be updated
    # (in this case, it appends messages to the list, rather than overwriting them)
    messages: Annotated[list, add_messages]


graph_state = StateGraph(State)

welcome_message = AIMessage(content="Welcome to the hello, world! state machine.")


def handle_greeting_node(state):
    if state["messages"]:
           # We've already greeted the user
        return {"messages": []}
    # Return a greeting ,since no message has been sent
    return {"messages": [welcome_message]}


graph_state.add_node("handle_greeting_node", lambda state: handle_greeting_node(state))
# graph_state.add_node("handle_user_input", handle_user_input)
graph_state.set_entry_point("handle_greeting_node")
# graph_state.add_edge("handle_greeting_node", "handle_user_input")
graph_state.add_edge("handle_greeting_node", END)
app = graph_state.compile()

app.invoke({"messages": []})