How to properly route tool in supervisor multi agent if some of the agent have tools?

[cris-m]

I was trying to route the following but if the agent use it tools, I got an error. here is the graph code:

def supervisor_agent(state):
    supervior_chain = prompt | llm.with_structured_output(routeResponse)
    return supervior_chain.invoke(state)

class routeResponse(BaseModel):
    next: Literal["Researcher", "Coder", "FINISH"]

def agent_node(state, agent, name):
    result = agent.invoke(state)

    return {
        "messages": [HumanMessage(content=result["messages"][-1].content, name=name)]
    }

team_members = ["Researcher", "Coder", "FINISH"]

research_agent = create_react_agent(llm, tools=[tavily_tool])
research_node = functools.partial(agent_node, agent=research_agent, name="Researcher")

code_agent = create_react_agent(llm, tools=[python_repl_tool])
code_node = functools.partial(agent_node, agent=code_agent, name="Coder")

class AgentState(TypedDict):
    messages: Annotated[Sequence[BaseMessage], operator.add]
    next: str

workflow = StateGraph(AgentState)
workflow.add_node("Researcher", research_node)
workflow.add_node("Coder", code_node)
workflow.add_node("supervisor", supervisor_agent)

workflow.add_node("Researcher_Tool", ToolNode([tavily_tool]))
workflow.add_node("Coder_Tool", ToolNode([python_repl_tool]))

conditional_map = {k: k for k in team_members}
conditional_map["FINISH"] = END
workflow.add_conditional_edges("supervisor", lambda x: x["next"], conditional_map)

workflow.add_conditional_edges("Researcher", tools_condition, ["Researcher_Tool", "supervisor"])
workflow.add_conditional_edges("Coder", tools_condition, ["Coder_Tool", "supervisor"])

workflow.add_edge(START, "supervisor")
workflow.add_edge("Researcher_Tool", "Researcher")
workflow.add_edge("Coder_Tool", "Coder")

graph = workflow.compile()

Error:

KeyError('__end__')Traceback (most recent call last):


  File "/usr/local/lib/python3.10/dist-packages/langgraph/pregel/__init__.py", line 1612, in stream
    for _ in runner.tick(


  File "/usr/local/lib/python3.10/dist-packages/langgraph/pregel/runner.py", line 104, in tick
    run_with_retry(t, retry_policy, writer=writer)


  File "/usr/local/lib/python3.10/dist-packages/langgraph/pregel/retry.py", line 40, in run_with_retry
    task.proc.invoke(task.input, config)


  File "/usr/local/lib/python3.10/dist-packages/langgraph/utils/runnable.py", line 412, in invoke
    input = context.run(step.invoke, input, config)


  File "/usr/local/lib/python3.10/dist-packages/langgraph/utils/runnable.py", line 184, in invoke
    ret = context.run(self.func, input, **kwargs)


  File "/usr/local/lib/python3.10/dist-packages/langgraph/graph/graph.py", line 96, in _route
    return self._finish(writer, input, result, config)


  File "/usr/local/lib/python3.10/dist-packages/langgraph/graph/graph.py", line 131, in _finish
    destinations: Sequence[Union[Send, str]] = [


  File "/usr/local/lib/python3.10/dist-packages/langgraph/graph/graph.py", line 132, in <listcomp>
    r if isinstance(r, Send) else self.ends[r] for r in result


KeyError: '__end__'

[cris-m]

I changed everything and fixed the issue

import functools
import operator
from typing import Annotated, Literal, Sequence

from langchain_anthropic import ChatAnthropic
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_core.messages import BaseMessage
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_experimental.tools import PythonREPLTool
from langgraph.graph import END, START, StateGraph
from langgraph.prebuilt import create_react_agent
from pydantic import BaseModel
from typing_extensions import TypedDict
from langchain_core.messages import AIMessage
from langgraph.prebuilt import ToolNode


# initial tools
tavily_tool = TavilySearchResults(max_results=5)
python_repl_tool = PythonREPLTool()

def agent_node(state, agent, name):
    result = agent.invoke(state)
    
    # Return the complete messages array including the AIMessage
    return {
        "messages": state["messages"] + result["messages"]
    }

members = ["Researcher", "Coder"]

system_prompt = (
    """You are a supervisor tasked with managing a conversation between the following workers: {members}.
    Given the following user request, respond with the worker to act next. Each worker will perform a task and respond with their results and status.
    If you see 'FINAL ANSWER' in any response or the task is complete, respond with FINISH.
    Otherwise, choose the most appropriate worker to continue the task."""
)

options = ["FINISH"] + members

class routeResponse(BaseModel):
    next: Literal["Researcher", "Coder", "FINISH"]

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", system_prompt),
        MessagesPlaceholder(variable_name="messages"),
        (
            "user",
            "Given the conversation above, who should act next?"
            " Or should we FINISH? Select one of: {options}",
        ),
    ]
).partial(options=str(options), members=", ".join(members))

llm = ChatAnthropic(model="claude-3-5-sonnet-20240620", temperature=0, max_tokens=1000)

def supervisor_agent(state):
    supervisor_chain = prompt | llm.with_structured_output(routeResponse)
    result = supervisor_chain.invoke(state)
    return {
        "messages": state["messages"],
        "next": result.next 
    }

def tools_condition(state):
    messages = state["messages"]
    last_message = messages[-1]
    
    if isinstance(last_message, AIMessage) and last_message.additional_kwargs.get("tool_calls"):
        return "tool" 
    return "supervisor"


class AgentState(TypedDict):
    messages: Annotated[Sequence[BaseMessage], operator.add]
    next: str

research_agent = create_react_agent(llm, tools=[tavily_tool])
research_node = functools.partial(agent_node, agent=research_agent, name="Researcher")

code_agent = create_react_agent(llm, tools=[python_repl_tool])
code_node = functools.partial(agent_node, agent=code_agent, name="Coder")

workflow = StateGraph(AgentState)
workflow.add_node("Researcher", research_node)
workflow.add_node("Coder", code_node)
workflow.add_node("supervisor", supervisor_agent)

workflow.add_node("Researcher_Tool", ToolNode([tavily_tool]))
workflow.add_node("Coder_Tool", ToolNode([python_repl_tool]))

# from langgraph.prebuilt import tools_condition

conditional_map = {k: k for k in members}
conditional_map["FINISH"] = END
workflow.add_conditional_edges("supervisor", lambda x: x["next"], conditional_map)


workflow.add_conditional_edges(
    "Researcher", 
    tools_condition, 
    {
        "tool": "Researcher_Tool",
        "supervisor": "supervisor"
    }
)

workflow.add_conditional_edges(
    "Coder", 
    tools_condition, 
    {
        "tool": "Coder_Tool",
        "supervisor": "supervisor"
    }
)

workflow.add_edge(START, "supervisor")
workflow.add_edge("Researcher_Tool", "Researcher")
workflow.add_edge("Coder_Tool", "Coder")


graph = workflow.compile()

[vbarda]

yes, this works, and you can also just use create_react_agent (https://langchain-ai.github.io/langgraph/reference/prebuilt/#langgraph.prebuilt.chat_agent_executor.create_react_agent) for both agents (researcher/coder) and avoid defining separate tool nodes altogether (as each react agent has a tool node inside). we're actually updating our tutorials to reflect that

[cris-m]

If I want to specify how the agent should use tools, I might need the agent to rely on its internal knowledge first. The agent should only use specific tools if its internal knowledge is insufficient to provide an answer.

I was thinking to use solution like the one in Multi-agent network instead of using create_react_agent

def create_agent(llm, tools, system_message: str):
    """Create an agent."""
    prompt = ChatPromptTemplate.from_messages(
        [
            (
                "system",
                "You are a helpful AI assistant, collaborating with other assistants."
                " Use the provided tools to progress towards answering the question."
                " If you are unable to fully answer, that's OK, another assistant with different tools "
                " will help where you left off. Execute what you can to make progress."
                " If you or any of the other assistants have the final answer or deliverable,"
                " prefix your response with FINAL ANSWER so the team knows to stop."
                " You have access to the following tools: {tool_names}.\n{system_message}",
            ),
            MessagesPlaceholder(variable_name="messages"),
        ]
    )
    prompt = prompt.partial(system_message=system_message)
    prompt = prompt.partial(tool_names=", ".join([tool.name for tool in tools]))
    return prompt | llm.bind_tools(tools)