Exiting the agent loop

After your agent has finished a workflow, you'll usually want to explicitly end that loop by calling the copilotkit_exit() method in your Python code.

Exiting the agent has different effects depending on mode:

Router Mode: Exiting the agent hands responsibility for handling input back to the router, which can initiate chat, call actions, other agents, etc. The router can return to this agent later (starting a new loop) to satisfy a user request.
Agent Lock Mode: Exiting the agent restarts the workflow loop for the current agent.

In this example from our email-sending app, the send_email node explicitly exits, then manually sends a response back to the user as a ToolMessage:

Install the CopilotKit SDK#

Any LangGraph agent can be used with CopilotKit. However, creating deep agentic experiences with CopilotKit requires our LangGraph SDK.

uv add copilotkit

poetry add copilotkit

pip install copilotkit --extra-index-url https://copilotkit.gateway.scarf.sh/simple/

conda install copilotkit -c copilotkit-channel

npm npm install @copilotkit/sdk-js

Exit the agent loop#

This will exit the agent session as soon as the current CrewAI run is finished, either by a breakpoint or by reaching the END node.

from litellm import completion
from crewai.flow.flow import start
from copilotkit.crewai import copilotkit_exit
# ...
@start()
async def send_email(self):
    """Send an email."""


    # get the last message and cast to ToolMessage
    last_message = self.state["messages"][-1]
    if last_message["content"] == "CANCEL":
        text_message = "❌ Cancelled sending email."
    else:
        text_message = "✅ Sent email."
    self.state["messages"].append({"role": "assistant", "content": text_message, "id": str(uuid.uuid4())})
    # Exit the agent loop after processing
    await copilotkit_exit()