基于Runnable封装记忆链实现记忆自动管理

一、实现思路

通过​​配置注入+生命周期钩子​​实现记忆组件的无侵入式集成：

1. ​​配置化传递​​

   chain = my_chain.configure({"configurable": {"memory": chat_history}})

将 Memory 实例通过 configurable 字段注入运行时上下文

​​2. 执行时获取​​
在 invoke/stream 执行方法中通过第二个参数访问上下文:

def invoke(input, config):
    memory = config.get("configurable", {}).get("memory")
    # 获取历史对话记录
    print(memory.get_history()) 

3. 生命周期存储​​
   通过 on_end 钩子自动持久化对话记录：

chain.with_listeners(on_end=lambda output: 
    memory.save(input=output["input"], response=output["output"]))

二、运行流程

三、资料推荐

• 官方文档参考
• 大模型中转API推荐
• ✨中转使用教程

四、示例代码

@classmethod
def _load_memory_variables(cls, input: Dict[str, Any], config: RunnableConfig) -> Dict[str, Any]:
    """加载记忆变量"""
    configurable = config.get("configurable", {})
    configurable_memory = configurable.get("memory", None)
    if configurable_memory is not None and isinstance(configurable_memory, BaseMemory):
        return configurable_memory.load_memory_variables(input)
    return {"history": []}

@classmethod
def _save_context(cls, run: Run, config: RunnableConfig) -> None:
    configurable = config.get("configurable", {})
    configurable_memory = configurable.get("memory", None)
    if configurable_memory is not None and isinstance(configurable_memory, BaseMemory):
        configurable_memory.save_context(run.inputs, run.outputs)

def debug(self, app_id: UUID):
    """聊天接口"""
    ...
    chain = (
        (RunnablePassthrough.assign(
            history=RunnableLambda(self._load_memory_variables) | itemgetter("history")
        ) | prompt | llm | StrOutputParser()).
        with_listeners(on_end=self._save_context)
    )

    # 5.调用链生成内容
    chain_input = {"query": req.query.data}
    content = chain.invoke(chain_input, config={"configurable": {"memory": memory}})
	...