langgraph tool如何获取上下文和RunableConfig

在构造langraph工具函数时，常需要使用到运行时的配置RunableConfig以及对话的上下文数据（常命名为State）。文章记录下在工具函数中该如何获取上述配置

获取运行时配置RunableConfig

获取运行时配置，比较简单，只需要将RunableConfig作为函数的一个入参即可，一些确定的元数据常常用RunableConfig进行传递。

from langchain_core.runnables.config import RunnableConfig
from langchain_core.tools import tool

@tool
def get_fligh_info(config: RunnableConfig):
    """
    获取乘客的航班信息
    """
    configuration = config.get("configurable", {})
    passenger_id = configuration.get("passenger_id")
    print(f"传入的passenger_id:{passenger_id}")
    return {"name":"老谢", "flight": "AX123"}

传入的passenger_id是123

获取对话上下文

tools中获取对话上下文(state)，需要使用InjectedState来注解。使用该注解后，函数的入参将对模型调用隐藏，意味着该参数不会有大模型的funcation calling生成，而是由state直接注入。可选注入整个state或者state中的某个字段，改写上面的函数

from typing import List
from typing_extensions import Annotated, TypedDict

from langgraph.graph.message import AnyMessage, add_messages
from langgraph.prebuilt import InjectedState
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import tool


class State(TypedDict):
    messages: Annotated[list[AnyMessage], add_messages]
    foo: str


@tool
def get_fligh_info(config: RunnableConfig,
                       state: Annotated[dict, InjectedState],
                       foo: Annotated[str, InjectedState("foo")]):
    """
    获取乘客的航班信息
    """
    configuration = config.get("configurable", {})
    passenger_id = configuration.get("passenger_id")
    print(f"传入的passenger_id:{passenger_id}")
    print(f"传入的state:{state}")
    print(f"传入的foo:{foo}")
    return {"name":"老谢", "flight": "AX123"}

测试代码

• graph定义

定义一个工具用于获取乘客的航班信息，定义一个助手Agent用于工具的调用，同时创建graph如下图

from typing_extensions import Annotated, TypedDict

from langchain_core.prompts import ChatPromptTemplate
from langchain_core.runnables import Runnable, RunnableConfig
from langchain_core.tools import tool
from langgraph.graph.message import AnyMessage, add_messages
from langgraph.graph import StateGraph, START
from langgraph.prebuilt import ToolNode, InjectedState, tools_condition
from langgraph.checkpoint.memory import MemorySaver

class State(TypedDict):
    messages: Annotated[list[AnyMessage], add_messages]
    foo: str

@tool
def get_fligh_info(config: RunnableConfig,
                       state: Annotated[dict, InjectedState],
                       foo: Annotated[str, InjectedState("foo")]):
    """
    获取乘客的航班信息
    """
    configuration = config.get("configurable", {})
    passenger_id = configuration.get("passenger_id")
    print(f"传入的passenger_id:{passenger_id}")
    print(f"传入的state:{state}")
    print(f"传入的foo:{foo}")
    return {"name":"老谢", "flight": "AX123"}

class Assistant:
    def __init__(self, runnable: Runnable):
        self.runnable = runnable

    def __call__(self, state: State, config: RunnableConfig):
        result = self.runnable.invoke(state)
        return {"messages": result, "foo": "bar"}

assistant_system_prompt = "你是航空公司的专业助手，你需要协助用户完成查询操作并解答相关问题"
assistant_prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            assistant_system_prompt,
        ),
        ("placeholder", "{messages}"),
    ],
)
assistant_tool = [get_fligh_info]
assistant_runnable = assistant_prompt | llm.bind_tools(assistant_tool)
assistant = Assistant(assistant_runnable)

# node定义
builder = StateGraph(State)
builder.add_node("assistant", assistant)
builder.add_node("tools", ToolNode(assistant_tool))

# edge定义
builder.add_edge(START, "assistant")
builder.add_conditional_edges("assistant", tools_condition)
builder.add_edge("tools", "assistant")
memory = MemorySaver()
graph = builder.compile(checkpointer=memory)

• graph调用

import uuid
thread_id = str(uuid.uuid4())
config = {"configurable": {"thread_id": thread_id, "passenger_id": 1234}}
events= graph.stream({"messages": ("user", "帮我查询一下航班信息")}, config, stream_mode="updates")
for event in events:
    continue

传入的passenger_id:1234
传入的state:{'messages': [HumanMessage(content='帮我查询一下航班信息', id='ceaf9e61-0141-4a4a-af93-2973c4a6f6ec'), AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_dMELZC2UM7gWMatoPwTZQwpx', 'function': {'arguments': '{}', 'name': 'get_fligh_info'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 29, 'prompt_tokens': 86, 'total_tokens': 115}, 'model_name': 'gpt-4', 'system_fingerprint': 'fp_e49e4201a9', 'finish_reason': 'tool_calls', 'logprobs': None, 'content_filter_results': {}}, id='run-d68e4eaa-cd94-49fa-9200-2360a786a4fe-0', tool_calls=[{'name': 'get_fligh_info', 'args': {}, 'id': 'call_dMELZC2UM7gWMatoPwTZQwpx', 'type': 'tool_call'}], usage_metadata={'input_tokens': 86, 'output_tokens': 29, 'total_tokens': 115})], 'foo': 'bar'}
传入的foo:bar

参考文档

https://github.com/webup/notebooks/blob/main/langgraph-tool-node.ipynb