Struct ConversationSummaryBufferMemory 

pub struct ConversationSummaryBufferMemory<M: BaseChatModel> { /* private fields */ }

Conversation Summary Buffer Memory

Combines summary and full conversation:

• Keeps the last k rounds of full conversation (ensuring fluency)
• Summarizes older conversations (saving tokens)

Example

use lc_memory::ConversationSummaryBufferMemory;
use lc_providers::OpenAIChat;

let llm = OpenAIChat::new(config);
let memory = ConversationSummaryBufferMemory::new(llm, 5); // Keep last 5 rounds

// After 20 rounds:
// - First 15 rounds -> summary
// - Last 5 rounds -> full conversation

Implementations

impl<M: BaseChatModel> ConversationSummaryBufferMemory<M>

pub fn new(llm: M, max_token_limit: usize) -> Self

使用给定的 LLM 与 token 预算创建新的摘要缓冲记忆。

pub fn with_input_key(self, key: impl Into<String>) -> Self

设置输入 key,save_context 用它从 inputs 中取出用户输入。

pub fn with_output_key(self, key: impl Into<String>) -> Self

设置输出 key,save_context 用它从 outputs 中取出 AI 输出。

pub fn with_memory_key(self, key: impl Into<String>) -> Self

设置记忆 key,加载的历史将暴露在该 key 下。

pub fn with_summary_prompt(self, prompt: impl Into<String>) -> Self

设置用于生成摘要的提示词模板。

pub fn with_return_messages(self, return_messages: bool) -> Self

设置加载的历史是否以消息列表(而非文本)形式返回。

pub fn with_counter(self, counter: Arc<dyn TokenCounter>) -> Self

注入自定义 token 计数器。

默认 TiktokenCounter(BPE 口径,与 ContextWindow 一致);需要零依赖 快路径时注入 CharRatioCounter::new(4)。

pub fn set_summary(&mut self, summary: impl Into<String>)

P1-3: 从持久化存储回灌摘要状态,保证续写会话摘要链连续。

pub fn set_max_token_limit(&mut self, max_token_limit: usize)

P1-3: 设置 token 预算(持久化 config 的单一来源)。

pub fn chat_memory(&self) -> &ChatMessageHistory

返回底层聊天消息历史的不可变引用。

pub fn chat_memory_mut(&mut self) -> &mut ChatMessageHistory

返回底层聊天消息历史的可变引用。

pub fn max_token_limit(&self) -> usize

返回当前配置的 token 预算。

pub async fn buffer(&self) -> String

返回当前的摘要缓冲内容。

pub fn last_summary_error(&self) -> Option<&str>

P2-4: 最近一次摘要失败的原因(无失败则 None)。

Trait Implementations

impl<M: BaseChatModel + Send + Sync + 'static> BaseChatMemory for ConversationSummaryBufferMemory<M>

where <M as Runnable<Vec<Message>, LLMResult>>::Error: Display,

P0-1: ConversationSummaryBufferMemory 实现 BaseChatMemory。

fn messages(&self) -> &[Message]

Get chat message list

fn add_message(&mut self, message: Message)

Add message

fn add_user_message(&mut self, content: &str)

Add user message

fn add_ai_message(&mut self, content: &str)

Add AI message

impl<M: BaseChatModel + Send + Sync + 'static> BaseMemory for ConversationSummaryBufferMemory<M>

where <M as Runnable<Vec<Message>, LLMResult>>::Error: Display,

fn memory_variables(&self) -> Vec<&str>

Get memory variable names

fn load_memory_variables<'life0, 'life1, 'async_trait>( &'life0 self, _inputs: &'life1 HashMap<String, String>, ) -> Pin<Box<dyn Future<Output = Result<HashMap<String, Value>, MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Load memory variables

fn save_context<'life0, 'life1, 'life2, 'async_trait>( &'life0 mut self, inputs: &'life1 HashMap<String, String>, outputs: &'life2 HashMap<String, String>, ) -> Pin<Box<dyn Future<Output = Result<(), MemoryError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait,

Save context

fn clear<'life0, 'async_trait>( &'life0 mut self, ) -> Pin<Box<dyn Future<Output = Result<(), MemoryError>> + Send + 'async_trait>>

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

Clear memory