How do I update state from a tool call?

[hwchase17]

I have an agent that I want to give tools, and I want these tools to update the state of the graph. How do I do this?

[hwchase17]

Currently, we do not support LangChain tools updating state. instead, you should just do the "tool call" in a node explicitly. Here's what this would look like:

from langgraph.graph import StateGraph, END, MessagesState
from langgraph.prebuilt import ToolNode
from typing import TypedDict, Literal
from langchain_openai import ChatOpenAI
from langchain_core.tools import tool



@tool
def get_weather(location: str):
    """Get weather for a location"""
    return "sunny"


class Location(TypedDict):
    """Call this to get the current location"""
    get: bool

tool_node = ToolNode([get_weather])

model = ChatOpenAI().bind_tools([get_weather, Location])

class State(MessagesState): 
    """State class for the agent"""
    location: str

def agent(state: State, config: dict) -> State:
    messages = [{"role": "system", "content": f"You are in {state.get('location', 'N/A')}"}] + state['messages']
    result = model.invoke(messages)
    return {"messages": [result]}


def location_node(state: State, config: dict) -> State:
    tool_call = state['messages'][-1].tool_calls[0]
    return {
        "messages": [{"role": "tool", "tool_call_id": tool_call['id'], "content": "Fetched!"}],
        "location": "SF"
    } 

def route_after_agent(state: State, config: dict) -> Literal['location_node', 'tool_node', END]:
    """Function to handle conditional edges for agent"""
    m = state['messages'][-1]
    if len(m.tool_calls) == 0:
        return END
    elif m.tool_calls[0]['name'] == "Location":
        return "location_node"
    else:
        return "tool_node"

workflow = StateGraph(State)

# Add nodes to the graph
workflow.add_node("agent", agent)
workflow.add_node("tool_node", tool_node)
workflow.add_node("location_node", location_node)

# Define edges
workflow.add_conditional_edges("agent", route_after_agent)
workflow.add_edge("location_node", "agent")
workflow.add_edge("tool_node", "agent")
workflow.set_entry_point("agent")

graph=workflow.compile()

[Sagaulf]

Seems like a common case that if you have an agent with access to multiple tools you would like to update the state depending on tool calls and the type of data returned - Currently running into this problem with an agent that has multiple RAG tools. Would be helpful to have a 'standard practice' in the docs on how to approach this.
Is there some reason for state modification not being supported in Langchain's tools?

[vbarda]

We have now added support for updating the state from the tools using the new Command type. Please see this how-to guide for reference https://langchain-ai.github.io/langgraph/how-tos/update-state-from-tools/