About the problem of langgraph return parameter clipping

[kissycn]

I implemented a langgraph agent with reference Launch Local LangGraph Server and deployed it through the langgraph platform. When calling it through the langgraph open api, the data flow can be returned normally, but some fields are too redundant. How can I trim them?

from typing import Annotated
from langchain_openai import ChatOpenAI
from typing_extensions import TypedDict
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages
from langgraph.types import StreamWriter
from typing import Dict, Any
from langchain_core.output_parsers import StrOutputParser
from langchain_core.messages import AIMessage

class State(TypedDict):
    messages: str

graph_builder = StateGraph(State)

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

def chatbot(state: State):
    chain = llm | StrOutputParser()
    return {"messages": [chain.invoke(state["messages"])]}

graph_builder.add_node("chatbot", chatbot)
graph_builder.set_entry_point("chatbot")
graph_builder.add_edge("chatbot", END)

graph = graph_builder.compile()

run command:

poetry run langgraph dev

curl -X POST 'http://127.0.0.1:2024/threads/1a9bbbe1-3a15-452d-bdb1-f0ca0dcddfd0/runs/stream' \
-H 'Content-Type: application/json' \
-d '{
    "input": {
        "messages": "hello"
    },
    "config": {
        "tags": [],
        "recursion_limit": 100,
        "configurable": {
            "thread_id": "4b76e733-7806-4f02-aa8a-4cbcde6b1503"
        }
    },
    "metadata": {},
    "stream_mode": ["messages-tuple"],
    "stream_subgraphs": false,
    "assistant_id": "fe096781-5601-53d2-b2f6-0d3403f7e9ca",
    "interrupt_before": [],
    "interrupt_after": [],
    "multitask_strategy": "rollback"
}'

response:
For front-end applications, only the content in content is needed. The other content seems redundant. How can I make langgraph return only the content of {"content":"Hello"}?

metadata data: {"run_id":"1efb1509-82fc-6ac0-af40-a9aaf4034512","attempt":1}
16:28:38
messages data: [{"content":"","additional_kwargs":{},"response_metadata":{},"type":"AIMessageChunk","name":null,"id":"run-72e23f05-1d89-449f-956a-9aae4b539b83","example":false,"tool_calls":[],"invalid_tool_calls":[],"usage_metadata":null,"tool_call_chunks":[]},{"created_by":"system","graph_id":"agent","run_attempt":1,"langgraph_version":"0.2.53","langgraph_plan":"developer","langgraph_host":"self-hosted","thread_id":"1a9bbbe1-3a15-452d-bdb1-f0ca0dcddfd0","run_id":"1efb1509-82fc-6ac0-af40-a9aaf4034512","assistant_id":"fe096781-5601-53d2-b2f6-0d3403f7e9ca","user_id":"","langgraph_step":394,"langgraph_node":"chatbot","langgraph_triggers":["start:chatbot"],"langgraph_path":["__pregel_pull","chatbot"],"langgraph_checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","ls_provider":"openai","ls_model_name":"gpt-4o-mini","ls_model_type":"chat","ls_temperature":0.7}]
16:28:40
messages data: [{"content":"Hello","additional_kwargs":{},"response_metadata":{},"type":"AIMessageChunk","name":null,"id":"run-72e23f05-1d89-449f-956a-9aae4b539b83","example":false,"tool_calls":[],"invalid_tool_calls":[],"usage_metadata":null,"tool_call_chunks":[]},{"created_by":"system","graph_id":"agent","run_attempt":1,"langgraph_version":"0.2.53","langgraph_plan":"developer","langgraph_host":"self-hosted","thread_id":"1a9bbbe1-3a15-452d-bdb1-f0ca0dcddfd0","run_id":"1efb1509-82fc-6ac0-af40-a9aaf4034512","assistant_id":"fe096781-5601-53d2-b2f6-0d3403f7e9ca","user_id":"","langgraph_step":394,"langgraph_node":"chatbot","langgraph_triggers":["start:chatbot"],"langgraph_path":["__pregel_pull","chatbot"],"langgraph_checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","ls_provider":"openai","ls_model_name":"gpt-4o-mini","ls_model_type":"chat","ls_temperature":0.7}]
16:28:40
messages data: [{"content":"!","additional_kwargs":{},"response_metadata":{},"type":"AIMessageChunk","name":null,"id":"run-72e23f05-1d89-449f-956a-9aae4b539b83","example":false,"tool_calls":[],"invalid_tool_calls":[],"usage_metadata":null,"tool_call_chunks":[]},{"created_by":"system","graph_id":"agent","run_attempt":1,"langgraph_version":"0.2.53","langgraph_plan":"developer","langgraph_host":"self-hosted","thread_id":"1a9bbbe1-3a15-452d-bdb1-f0ca0dcddfd0","run_id":"1efb1509-82fc-6ac0-af40-a9aaf4034512","assistant_id":"fe096781-5601-53d2-b2f6-0d3403f7e9ca","user_id":"","langgraph_step":394,"langgraph_node":"chatbot","langgraph_triggers":["start:chatbot"],"langgraph_path":["__pregel_pull","chatbot"],"langgraph_checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","ls_provider":"openai","ls_model_name":"gpt-4o-mini","ls_model_type":"chat","ls_temperature":0.7}]
16:28:40
messages data: [{"content":" How","additional_kwargs":{},"response_metadata":{},"type":"AIMessageChunk","name":null,"id":"run-72e23f05-1d89-449f-956a-9aae4b539b83","example":false,"tool_calls":[],"invalid_tool_calls":[],"usage_metadata":null,"tool_call_chunks":[]},{"created_by":"system","graph_id":"agent","run_attempt":1,"langgraph_version":"0.2.53","langgraph_plan":"developer","langgraph_host":"self-hosted","thread_id":"1a9bbbe1-3a15-452d-bdb1-f0ca0dcddfd0","run_id":"1efb1509-82fc-6ac0-af40-a9aaf4034512","assistant_id":"fe096781-5601-53d2-b2f6-0d3403f7e9ca","user_id":"","langgraph_step":394,"langgraph_node":"chatbot","langgraph_triggers":["start:chatbot"],"langgraph_path":["__pregel_pull","chatbot"],"langgraph_checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","ls_provider":"openai","ls_model_name":"gpt-4o-mini","ls_model_type":"chat","ls_temperature":0.7}]
16:28:40
messages data: [{"content":" can","additional_kwargs":{},"response_metadata":{},"type":"AIMessageChunk","name":null,"id":"run-72e23f05-1d89-449f-956a-9aae4b539b83","example":false,"tool_calls":[],"invalid_tool_calls":[],"usage_metadata":null,"tool_call_chunks":[]},{"created_by":"system","graph_id":"agent","run_attempt":1,"langgraph_version":"0.2.53","langgraph_plan":"developer","langgraph_host":"self-hosted","thread_id":"1a9bbbe1-3a15-452d-bdb1-f0ca0dcddfd0","run_id":"1efb1509-82fc-6ac0-af40-a9aaf4034512","assistant_id":"fe096781-5601-53d2-b2f6-0d3403f7e9ca","user_id":"","langgraph_step":394,"langgraph_node":"chatbot","langgraph_triggers":["start:chatbot"],"langgraph_path":["__pregel_pull","chatbot"],"langgraph_checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","checkpoint_ns":"chatbot:1cab0926-f9af-0f6b-abe5-be7b24fd1696","ls_provider":"openai","ls_model_name":"gpt-4o-mini","ls_model_type":"chat","ls_temperature":0.7}]
16:28:40

[kissycn]

This is an engineering problem. My agent is published through LangGraph Platform. Then I want to expose the stream to our IOS App. Then there will be some business logic processing between App and our LangGraph PlatformServe (for example, not limited to formatting parameters, parameter verification, permission verification, etc.). Then there are three paths in front of us:

1. Directly expose the API of LangGraph PlatformServe to IOS APP. The problem is that the API parameters are complex, there is too much redundant information, and it is not flexible and there is no way to customize the input and output parameters.

2. Use LangGraph Python SDK to call LangGraph Platform, and then customize some input parameters and output parameters through FastAPI. The data flow becomes: APP —> FastApiAdatper —> LangGraph Platform Serve. The advantage of this is flexibility and customization, but the disadvantage is that the data has an extra layer of transfer, which feels a bit useless;

3. There is no need to publish the Agent to LangGraph PlatformServe. Directly define a set of FfastAPI and then call langgraph to complete the business logic operation. The advantage of this is flexibility and one less layer of data transfer. The disadvantage is that some historical messages and context messages need to be processed;