Struct RunnableWithMessageHistory 

pub struct RunnableWithMessageHistory<L> { /* private fields */ }

带记忆的 LLM 封装,作为单个 Runnable 参与 LCEL 组合。

Implementations

impl<L> RunnableWithMessageHistory<L>

pub fn new(llm: L, memory: impl BaseMemory + 'static) -> Self

用 LLM + 单个记忆对象构造封装(所有调用共享这一份记忆)。

pub fn with_session_history<F>(llm: L, factory: F) -> Self

where F: Fn(&str) -> SharedMemory + Send + Sync + 'static,

用 LLM + session 回调构造封装(对齐 Python RunnableWithMessageHistory(llm, get_session_history))。

factory(session_id) 为一个 session 槽建出记忆对象;每次调用按 config.configurable["session_id"] 选槽,同一 session 复用已建槽。

Example

let pipe = RunnableWithMessageHistory::with_session_history(llm, |session_id| {
    Arc::new(Mutex::new(Box::new(
        ConversationBufferMemory::new().with_return_messages(true),
    ) as Box<dyn BaseMemory>))
});
let cfg = RunnableConfig::new().with_configurable("session_id", json!("s1"));
pipe.invoke("我叫小明".into(), Some(cfg)).await?;

pub fn with_max_sessions(self, max: usize) -> Self

设置 session 缓存上限(Sessions 模式)。超过上限后,新 session 会淘汰最旧 的槽,防 session_id 无限增长的内存 DoS(M2a)。默认 [DEFAULT_MAX_SESSIONS]。

pub fn memory(&self) -> SharedMemory

暴露内部记忆句柄,便于读取已保存的历史(调试、展示、验证写回等)。

Sessions 模式下槽不唯一,返回的是占位记忆(不会参与管道读写); 要检查真实历史请从管道内部或业务侧记忆对象读取。

Trait Implementations

impl<L> Runnable<String, LLMResult> for RunnableWithMessageHistory<L>

where L: BaseChatModel + 'static, L::Error: Into<LcelError>,

type Error = LcelError

Error type.

fn invoke<'life0, 'async_trait>( &'life0 self, input: String, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<LLMResult, LcelError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Transforms single input to output.

fn batch<'life0, 'async_trait>( &'life0 self, inputs: Vec<Input>, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<Vec<Output>, Self::Error>> + Send + 'async_trait>>

where 'life0: 'async_trait, Self: 'async_trait,

Batch processing - transforms multiple inputs to outputs.

fn batch_as_completed<'life0, 'async_trait>( &'life0 self, inputs: Vec<Input>, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<Vec<(usize, Output)>, Self::Error>> + Send + 'async_trait>>

where 'life0: 'async_trait, Self: 'async_trait,

Batch processing that returns results in completion order.

fn stream<'life0, 'async_trait>( &'life0 self, input: Input, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<Pin<Box<dyn Stream<Item = Result<Output, Self::Error>> + Send>>, Self::Error>> + Send + 'async_trait>>

where 'life0: 'async_trait, Self: 'async_trait,

Streaming output - for real-time responses (LLM, etc).

fn transform<'life0, 'async_trait>( &'life0 self, input: Pin<Box<dyn Stream<Item = Result<Input, Self::Error>> + Send>>, config: Option<RunnableConfig>, ) -> Pin<Box<dyn Future<Output = Result<Pin<Box<dyn Stream<Item = Result<Output, Self::Error>> + Send>>, Self::Error>> + Send + 'async_trait>>

where 'life0: 'async_trait, Self: 'async_trait,

Stream-to-stream transformation - the core of LCEL streaming.