LangGraph Swarm: Missing ToolMessage for Tool Calls Before Agent Handoff

[jainamnahar14surat]

I'm using langgraph_swarm with multiple agents and tool calls. The issue is that when an agent makes a tool call (e.g., transfer_to_reporting_agent), LangGraph expects a corresponding ToolMessage, but sometimes the agent switches before the ToolMessage is recorded, causing this error:

ValueError: Found AIMessages with tool_calls that do not have a corresponding ToolMessage.

What I've Observed:

• The tool call is triggered correctly by the assistant.
• The handoff happens too soon, before the ToolMessage is appended.
• If multiple tool calls happen in sequence, some ToolMessages are skipped.
• Setting max_parallel_agents=1 does not fully resolve the issue.

What I’ve Tried:

• Debugging shows the ToolMessage is generated but not always stored in history.
• Explicitly appending ToolMessage before agent handoff in handoff_to_agent().
• Ensuring ToolMessage is added to state["messages"] before switching agents.

Question:
How can I ensure that LangGraph properly records ToolMessages before switching agents? Has anyone else faced this issue?

[daavoo]

Any updates on this?

[jainamnahar14surat]

no, nothing updates on it

[chamanthmvs]

@vbarda could you please respond to this?

[zehuiHuang]

I encountered the same problem

[vbarda]

there are several issues in the langgraph-swarm github repository with the answer https://github.com/langchain-ai/langgraph-swarm-py/issues?q=is%3Aissue%20state%3Aclosed%20found%20aimessage%20with%20tool_calls

this occurs on parallel handoffs, which is not supported in the library - you would need to explicitly disable it in your individual agents via model.bind_tools(..., parallel_tool_calls=False)

[jainamnahar14surat]

assistant_runnable = create_react_agent(
model_name,
tools_name,
prompt ,
name
)

here where we put the parallel tool = False

[manojkumarvohra]

@jainamnahar14surat do it like this:

def book_hotel(hotel_name: str):
"""Book a hotel"""
return f"Successfully booked a stay at {hotel_name}."

tools=[book_hotel, transfer_to_flight_assistant]
llm_with_tools = llm.bind_tools(tools, parallel_tool_calls=False)

hotel_assistant = create_react_agent(
model=llm_with_tools,
tools=tools,
prompt="You are a hotel booking assistant",
name="hotel_assistant"
)

[NithinXD]

this works for openai and anthropic models but what about gemini models?

[manojkumarvohra]

You can use langchain google integration sdk with ChatGoogleGenerativeAI class to interact with Google's chat models.

Refer:
https://python.langchain.com/docs/integrations/chat/google_generative_ai/#tool-calling

[NithinXD]

hi,
Can we disable parallel tool calling using ChatGoogleGenerativeAI?

Im using Gemini ChatGoogleGenerativeAI create_react_agent in a swarm without an orchestrator.
Each agent has multiple tools including handoffs and when i pass queries like “add task to db and write a poem” the model tries to call all the tools in parallel. Thats what causes this error. Ive tried AgentExecutor just to get the same error.

So I need a solution to disable parallel tool calls with gemini models.