CopilotKit

Disabling state streaming

Granularly control what is streamed to the frontend.

What is this?#

By default, CopilotKit will stream both your messages and tool calls to the frontend when you use copilotkit_stream. You can disable this by choosing when to use copilotkit_stream vs calling completion directly.

When should I use this?#

Occasionally, you'll want to disable streaming temporarily — for example, the LLM may be doing something the current user should not see, like emitting tool calls or questions pertaining to other employees in an HR system.

Implementation#

Disable all streaming#

You can control whether to stream messages or tool calls by selectively wrapping calls to completion with copilotkit_stream.

from copilotkit.crewai import copilotkit_stream
from typing import cast, Any
from litellm import completion

@start()
async def start(self):

    # 1) Do not emit messages or tool calls, keeping the LLM call private.
    response = completion(
        model="openai/gpt-5.4",
        messages=[
            {"role": "system", "content": "You are a helpful assistant"},
            *self.state.messages
        ],
    )
    message = response.choices[0].message

    # 2) Or wrap the LLM call with `copilotkit_stream` to stream message tokens.
    #    Note that we pass `stream=True` to the inner `completion` call.
    response = await copilotkit_stream(
        completion(
            model="openai/gpt-5.4",
            messages=[
                {"role": "system", "content": "You are a helpful assistant"},
                *self.state.messages
            ],
            stream=True
        )
    )
    message = cast(Any, response).choices[0]["message"]

On this page

What is this?When should I use this?ImplementationDisable all streaming