langgraph/tutorials/multi_agent/multi-agent-collaboration/ Bedrock multiagent support

[invalidexplorer]

Hi everyone, I have been trying to implement the tutorial for multiagent on claude using AWS bedrock but I am getting the following error when agent 1 tries to pass information to agent 2.

"ValidationException                       Traceback (most recent call last)
File ~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:681, in BedrockBase._prepare_input_and_invoke(self, prompt, system, messages, stop, run_manager, **kwargs)
    [680](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:680) try:
--> [681](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:681)     response = self.client.invoke_model(**request_options)
    [683](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:683)     (
    [684](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:684)         text,
    [685](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:685)         tool_calls,
   (...)
    [688](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:688)         stop_reason,
    [689](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:689)     ) = LLMInputOutputAdapter.prepare_output(provider, response).values()

File ~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:565, in ClientCreator._create_api_method.<locals>._api_call(self, *args, **kwargs)
    [564](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:564) # The "self" in this scope is referring to the BaseClient.
--> [565](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:565) return self._make_api_call(operation_name, kwargs)

File ~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:1021, in BaseClient._make_api_call(self, operation_name, api_params)
   [1020](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:1020)     error_class = self.exceptions.from_code(error_code)
-> [1021](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:1021)     raise error_class(parsed_response, operation_name)
   [1022](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/botocore/client.py:1022) else:

ValidationException: An error occurred (ValidationException) when calling the InvokeModel operation: Your API request included an `assistant` message in the final position, which would pre-fill the `assistant` response. When using tools, pre-filling the `assistant` response is not supported.

During handling of the above exception, another exception occurred:
...
--> [692](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:692)     raise ValueError(f"Error raised by bedrock service: {e}")
    [694](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:694) if stop is not None:
    [695](https://file+.vscode-resource.vscode-cdn.net/Users/username1/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langchain_aws/llms/bedrock.py:695)     text = enforce_stop_tokens(text, stop)

ValueError: Error raised by bedrock service: An error occurred (ValidationException) when calling the InvokeModel operation: Your API request included an `assistant` message in the final position, which would pre-fill the `assistant` response. When using tools, pre-filling the `assistant` response is not supported."  

Is bedrock not supported by langgraph's multimodel agent? Or am I doing something wrong here?

[hwchase17]

Your API request included an `assistant` message in the final position, which would pre-fill the `assistant` response. When using tools, pre-filling the `assistant` response is not supported.

The graph used in this example seems to send an assistant message as the final message, which Bedrock does not support. So the exact graph used in this example is not supported at the moment, but you can probably modify the graph slightly to avoid this and make it work

[invalidexplorer]

Thanks for the swift reply! Can you give me an example of how I can do that? And I am not able to understand how this multiagent graph is sending assistant message as the final message.

[invalidexplorer]

Hey @hwchase17 in

def agent_node(state, agent, name): result = agent.invoke(state) # We convert the agent output into a format that is suitable to append to the global state if isinstance(result, ToolMessage): pass else: result = AIMessage(**result.dict(exclude={"type", "name"}), name=name) return { "messages": [result], # Since we have a strict workflow, we can # track the sender so we know who to pass to next. "sender": name, }
I tried replacing AIMessage with 'HumanMessage' to fix this issue but I got the error

`ValueError                                Traceback (most recent call last)
Cell In[35], [line 13](vscode-notebook-cell:?execution_count=35&line=13)
      [1](vscode-notebook-cell:?execution_count=35&line=1) get_ipython().system('pip3 install openpyxl')
      [2](vscode-notebook-cell:?execution_count=35&line=2) events = graph.stream(
      [3](vscode-notebook-cell:?execution_count=35&line=3)     {
      [4](vscode-notebook-cell:?execution_count=35&line=4)         "messages": [
   (...)
     [11](vscode-notebook-cell:?execution_count=35&line=11)     {"recursion_limit": 150},
     [12](vscode-notebook-cell:?execution_count=35&line=12) )
---> [13](vscode-notebook-cell:?execution_count=35&line=13) for s in events:
     [14](vscode-notebook-cell:?execution_count=35&line=14)     print(s)
     [15](vscode-notebook-cell:?execution_count=35&line=15)     print("----")

File ~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1073, in Pregel.stream(self, input, config, stream_mode, output_keys, input_keys, interrupt_before, interrupt_after, debug)
   [1070](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1070)         del fut, task
   [1072](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1072) # panic on failure or timeout
-> [1073](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1073) _panic_or_proceed(done, inflight, step)
   [1074](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1074) # don't keep futures around in memory longer than needed
   [1075](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1075) del done, inflight, futures

File ~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1643, in _panic_or_proceed(done, inflight, step)
   [1641](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1641)             inflight.pop().cancel()
   [1642](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1642)         # raise the exception
-> [1643](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/pregel/__init__.py:1643)         raise exc
...
---> [76](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/prebuilt/tool_node.py:76)     raise ValueError("Last message is not an AIMessage")
     [78](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/prebuilt/tool_node.py:78) def run_one(call: ToolCall):
     [79](https://file+.vscode-resource.vscode-cdn.net/Users/nikhilvx/Intern%20Project/~/Library/Python/3.9/lib/python/site-packages/langgraph/prebuilt/tool_node.py:79)     output = self.tools_by_name[call["name"]].invoke(call["args"], config)

ValueError: Last message is not an AIMessage`

Would really appreciate your help with this.

[tuanvdo]

I ran into the same error which is preventing us from moving forward with our multi-agent design using langgraph. Can you give us a hint how can we work around the issue of assistant message @hwchase17 ? Thanks a lot.

[tuanvdo]

I am able to continue with the following "work-around" - instead of always wrapping the result as AIMessage, we wrap it only when it's a result with tool_calls. Can you try on your end @invalidexplorer ?

        if result.tool_calls:
            result = AIMessage(**result.dict(exclude={"type", "name"}), name=name)
        else:
            result = HumanMessage(**result.dict(exclude={"type", "name"}), name=name)

[invalidexplorer]

Thanks @tuanvdo ! seems to be working fine with this!

[jhachirag7]

`from langchain_core.messages import (
BaseMessage,
HumanMessage,
ToolMessage,
)
from inputs import description,links
import operator
from typing import Annotated, Sequence, TypedDict
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from tools.uber_suggest import search_seo_keywords
from tools.scraper import scrape
from tools.serper import get_serper_results
from langgraph.graph import END, StateGraph, START
from langgraph.prebuilt import ToolNode
from langchain_aws import ChatBedrock,ChatBedrockConverse
import os

llm = ChatBedrockConverse(
model_id="anthropic.claude-3-sonnet-20240229-v1:0",
temperature=0,
credentials_profile_name="bedrock",
region_name='us-west-2'
)

llm = ChatOpenAI(model="gpt-4-1106-preview")

def create__generation_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)

def create_reviewer_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."
" prefix your response with FINAL REVIEW or CONTINUE so the team knows to stop or continue."
" prefix will be FINAL REVIEW if nothing major to review."
" prefix will be CONTINUE if you have a critique"
" 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)

This defines the object that is passed between each node

in the graph. We will create different nodes for each agent and tool

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

import functools

from langchain_core.messages import AIMessage

Helper function to create a node for a given agent

def agent_node(state, agent, name):
result = agent.invoke(state)
# We convert the agent output into a format that is suitable to append to the global state
if result.tool_calls:
result = AIMessage(**result.dict(exclude={"type", "name"}), name=name)
else:
result = HumanMessage(**result.dict(exclude={"type", "name"}), name=name)
return {
"messages": [result],
# Since we have a strict workflow, we can
# track the sender so we know who to pass to next.
"sender": name,
}

Research agent and node

generator_agent = create__generation_agent(
llm,
tools,
system_message="""
""",
)
generator_node = functools.partial(agent_node, agent=generator_agent, name="generator")

chart_generator

reviewer_agent = create_reviewer_agent(
llm,
tools,
system_message="""""",
)
reviewer_node = functools.partial(agent_node, agent=reviewer_agent, name="reviewer")
tool_node = ToolNode(tools)

from typing import Literal

def router(state) -> Literal["call_tool", "end", "continue"]:
# This is the router
messages = state["messages"]
last_message = messages[-1]
if last_message.tool_calls:
# The previous agent is invoking a tool
return "call_tool"
if "FINAL ANSWER" in last_message.content:
return "continue"
elif "FINAL REVIEW" in last_message.content:
return "end"
else:
return "continue"

workflow = StateGraph(AgentState)

workflow.add_node("generator", generator_node)
workflow.add_node("reviewer", reviewer_node)
workflow.add_node("call_tool", tool_node)

workflow.add_conditional_edges(
"generator",
router,
{"continue": "reviewer", "call_tool": "call_tool", "end": END},
)
workflow.add_conditional_edges(
"reviewer",
router,
{"continue": "generator", "call_tool": "call_tool", "end": END},
)

workflow.add_conditional_edges(
"call_tool",
# Each agent node updates the 'sender' field
# the tool calling node does not, meaning
# this edge will route back to the original agent
# who invoked the tool
lambda x: x["sender"],
{
"generator": "generator",
"reviewer": "reviewer",
},
)
workflow.add_edge(START, "generator")
graph = workflow.compile()

events = graph.stream(
{
"description":description,
"links":links,
"messages": [
HumanMessage(
content=f"""""
)
],
},
# Maximum number of steps to take in the graph
{"recursion_limit": 150},
)
for s in events:
print(s)
print("----")`

when calling the Converse operation: A conversation must alternate between user and assistant roles. Make sure the conversation alternates between user and assistant roles and try again.

i got this error ?