Langserve & AgentExecutor Output on streaming events using beta API

[rhlarora84]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

async for event in self.agent_executor.astream_events(
                input=input,
                config=config,
                version=version,
                include_names=include_names,
                include_types=include_types,
                include_tags=include_tags,
                exclude_names=exclude_names,
                exclude_types=exclude_types,
                exclude_tags=exclude_tags,
                **kwargs
        ):
            kind = event["event"]

            if kind.startswith("on_chain"):
               ## Delete history here

Description

Hi,
Is there a way to control the output and remove history & agent_scratchpad from AgentExecutor when using the astream_events method? I am using Langserve. I can successfully do that with setting the output_type and it works on invoke/ainvoke.

System Info

System Information

OS: Windows
OS Version: 10.0.19045
Python Version: 3.11.4 (tags/v3.11.4:d2340ef, Jun 7 2023, 05:45:37) [MSC v.1934 64 bit (AMD64)]

Package Information

langchain_core: 0.2.28
langchain: 0.2.12
langchain_community: 0.2.11
langsmith: 0.1.84
langchain_anthropic: 0.1.22
langchain_aws: 0.1.15
langchain_experimental: 0.0.64
langchain_ollama: 0.1.0rc0
langchain_openai: 0.1.20
langchain_text_splitters: 0.2.1
langserve: 0.2.2

[dosubot[bot]]

Yes, you can control the output and remove history and agent_scratchpad from AgentExecutor when using the astream_events method in Langserve by customizing the prompt and the agent's behavior. Here is an example of how you can achieve this:

from langchain.agents import AgentExecutor, create_tool_calling_agent
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.tools import tool
from langchain_openai import ChatOpenAI

model = ChatOpenAI(model="gpt-4o")

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant."),
        ("human", "{input}"),
        # Remove the placeholder for agent_scratchpad
    ]
)

@tool
def magic_function(input: int) -> int:
    """Applies a magic function to an input."""
    return input + 2

tools = [magic_function]

agent = create_tool_calling_agent(model, tools, prompt=prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools)

async for event in agent_executor.astream_events({"input": query}):
    if event["event"] == "on_chat_model_stream":
        event["data"].pop('history', None)
        event["data"].pop('agent_scratchpad', None)
        print(event["data"])

In this example, the prompt is defined without the agent_scratchpad placeholder, and during the streaming of events, the history and agent_scratchpad are removed from each event before printing. This approach leverages the astream_events method to filter and process the streamed output effectively ^[1][2].

To continue talking to Dosu, mention @dosu.