Checked other resources

• This is a feature request, not a bug report or usage question.
• I added a clear and descriptive title that summarizes the feature request.
• I used the GitHub search to find a similar feature request and didn't find it.
• I checked the LangChain documentation and API reference to see if this feature already exists.
• This is not related to the langchain-community package.

Package (Required)

• langchain
• langchain-openai
• langchain-anthropic
• langchain-classic
• langchain-core
• langchain-cli
• langchain-model-profiles
• langchain-tests
• langchain-text-splitters
• langchain-chroma
• langchain-deepseek
• langchain-exa
• langchain-fireworks
• langchain-groq
• langchain-huggingface
• langchain-mistralai
• langchain-nomic
• langchain-ollama
• langchain-perplexity
• langchain-prompty
• langchain-qdrant
• langchain-xai
• Other / not sure / general

Feature Description

In streaming mode, only the agent's final output should be retrieved, not any intermediate output, such as tool calls. Note that this is in streaming mode.

Use Case

from langchain.agents import create_agent
from langchain_core.messages import HumanMessage, TextContentBlock
import os
from langchain.chat_models import init_chat_model
from langchain.tools import tool
model = init_chat_model(
model="glm-4.6v",
model_provider="openai",
base_url="https://open.bigmodel.cn/api/paas/v4/",
api_key=os.getenv("ZHIPUAI_API_KEY"),
)

@tool
def query_fabric(name: str) -> str:
"""get fabric info"""
return f"{name} is good."

agent = create_agent(
model=model,
tools=[query_fabric],
)

messages = [HumanMessage(content_blocks=[TextContentBlock(text="How is the linen herringbone fabric?", type="text")])]
async for stream_mode, chunk in agent.astream(
{"messages": messages}, stream_mode=["messages"]
):
token, metadata = chunk
if metadata.get("tool_calls"):
continue
if getattr(token, "tool_calls", []):
continue
if metadata['langgraph_node'] == "tools":
continue
if not token.content_blocks:
continue
if token.content_blocks[0]["type"] == "text":
print(token.content_blocks[0]["text"], end="")

The above is the test code, and the output is as follows：

"""
I'll help you find information about linen herringbone fabric.

Based on the information available, linen herringbone fabric is considered good. Linen herringbone is typically known for its durability, breathability, and classic textured appearance. The herringbone pattern gives it a distinctive V-shaped weave that adds visual interest and texture to the fabric. Linen is a natural fiber that's highly absorbent and gets softer with wear, making it a popular choice for various applications including clothing, home textiles, and upholstery.
"""

Because some LLM have non-empty content when calling the tool, as shown below:

AIMessage(content="\nI'll help you find information about the linen herringbone fabric.\n", additional_kwargs={}, response_metadata={'finish_reason': 'tool_calls', 'model_name': 'glm-4.6v', 'model_provider': 'openai'}, id='lc_run--019b5a5b-1d9d-7a20-962a-247b2901755b', tool_calls=[{'name': 'query_fabric', 'args': {'name': 'linen herringbone'}, 'id': 'call_3506c63fc43f4951854b454b', 'type': 'tool_call'}], usage_metadata={'input_tokens': 165, 'output_tokens': 94, 'total_tokens': 259, 'input_token_details': {'cache_read': 43}, 'output_token_details': {}})

This causes the output of what should be an intermediate process (\nI'll help you find information about the linen herringbone fabric.\n) to also be displayed in streaming mode.
However, I only want to stream the agent's final response, i.e., the following:

"""
Based on the information available, linen herringbone fabric is considered good. Linen herringbone is typically known for its durability, breathability, and classic textured appearance. The herringbone pattern gives it a distinctive V-shaped weave that adds visual interest and texture to the fabric. Linen is a natural fiber that's highly absorbent and gets softer with wear, making it a popular choice for various applications including clothing, home textiles, and upholstery.
"""

I have tried many methods, such as astream_events or middleware, but none of them can achieve this. Are there any other methods to do this, or can this be added as a new feature?

Proposed Solution

No response

Alternatives Considered

No response

Additional Context

No response