V2.5

README.md

@@ -111,6 +111,7 @@ cortex://resources/{resource_name}/
 - <strong>Session Management:</strong> Track conversation timelines, participants, and message history with automatic indexing and event-driven processing.
 - <strong>Multi-Tenancy Support:</strong> Isolated memory spaces for different users and agents within a single deployment via tenant-aware collection naming.
 - <strong>Event-Driven Automation:</strong> File watchers and auto-indexers for background processing, synchronization, and profile enrichment.
+- <strong>LLM Result Caching:</strong> Intelligent caching with LRU eviction and TTL expiration reduces redundant LLM API calls by 50-75%, with cascade layer debouncing for 70-90% reduction in layer updates.
 - <strong>Agent Framework Integration:</strong> Built-in support for Rig framework and Model Context Protocol (MCP).
 - <strong>Web Dashboard:</strong> Svelte 5 SPA (Insights) for monitoring, tenant management, and semantic search visualization.
 

README_zh.md

@@ -111,6 +111,7 @@ cortex://resources/{resource_name}/
 - <strong>会话管理：</strong> 跟踪对话时间线、参与者和消息历史，具有自动索引和事件驱动处理。
 - <strong>多租户支持：</strong> 通过租户感知集合命名，在单个部署中为不同用户和代理提供隔离的内存空间。
 - <strong>事件驱动自动化：</strong> 文件监视器和自动索引器用于后台处理、同步和配置文件丰富。
+- <strong>LLM结果缓存：</strong> 智能缓存采用LRU淘汰和TTL过期机制，减少50-75%的冗余LLM API调用，级联层防抖可减少70-90%的层更新调用。
 - <strong>代理框架集成：</strong> 内置支持Rig框架和模型上下文协议（MCP）。
 - <strong>Web仪表板：</strong> Svelte 5 SPA（Insights）用于监控、租户管理和语义搜索可视化。
 

cortex-mem-cli/src/commands/get.rs

@@ -7,10 +7,20 @@ pub async fn execute(
     operations: Arc<MemoryOperations>,
     uri: &str,
     abstract_only: bool,
+    overview: bool,
 ) -> Result<()> {
     println!("{} Getting memory: {}", "🔍".bold(), uri.cyan());
 
-    if abstract_only {
+    if overview {
+        // Get overview (L1 layer)
+        let overview_result = operations.get_overview(uri).await?;
+
+        println!("\n{}", "─".repeat(80).dimmed());
+        println!("{} Overview (L1)", "📝".bold());
+        println!("{}\n", "─".repeat(80).dimmed());
+        println!("{}", overview_result.overview_text);
+        println!("{}\n", "─".repeat(80).dimmed());
+    } else if abstract_only {
         // Get abstract (L0 layer)
         let abstract_result = operations.get_abstract(uri).await?;
 

cortex-mem-cli/src/commands/search.rs

@@ -50,9 +50,9 @@ pub async fn execute(
         recursive: true,
     };
 
-    // Perform vector search
+    // Perform layered vector search (L0/L1/L2 hierarchical search)
     let results = operations.vector_engine()
-        .semantic_search(query, &options)
+        .layered_semantic_search(query, &options)
         .await?;
 
     println!("\n{} Found {} results\n", "✓".green().bold(), results.len());

cortex-mem-cli/src/commands/session.rs

@@ -49,5 +49,30 @@ pub async fn create(
         println!("  {}: {}", "Title".cyan(), t);
     }
 
+    Ok(())
+}
+
+/// Close a session and trigger memory extraction, layer generation, and indexing
+pub async fn close(operations: Arc<MemoryOperations>, thread: &str) -> Result<()> {
+    println!("{} Closing session: {}", "🔒".bold(), thread.cyan());
+
+    // Close the session (triggers SessionClosed event → MemoryEventCoordinator)
+    operations.close_session(thread).await?;
+
+    println!("{} Session closed successfully", "✓".green().bold());
+    println!("  {}: {}", "Thread ID".cyan(), thread);
+    println!();
+    println!("{} Waiting for memory extraction, L0/L1 generation, and indexing to complete...", "⏳".yellow().bold());
+
+    // Wait for background tasks to complete (max 60 seconds)
+    // This ensures memory extraction, layer generation, and vector indexing finish before CLI exits
+    let completed = operations.flush_and_wait(Some(1)).await;
+
+    if completed {
+        println!("{} All background tasks completed successfully", "✓".green().bold());
+    } else {
+        println!("{} Background tasks timed out (some may still be processing)", "⚠".yellow().bold());
+    }
+
     Ok(())
 }

cortex-mem-cli/src/main.rs

@@ -87,6 +87,10 @@ enum Commands {
         /// Show abstract (L0) instead of full content
         #[arg(short, long)]
         abstract_only: bool,
+
+        /// Show overview (L1) instead of full content
+        #[arg(short, long)]
+        overview: bool,
     },
 
     /// Delete a memory
@@ -131,6 +135,12 @@ enum SessionAction {
         #[arg(short, long)]
         title: Option<String>,
     },
+
+    /// Close a session and trigger memory extraction, layer generation, and indexing
+    Close {
+        /// Thread ID to close
+        thread: String,
+    },
 }
 
 #[derive(Subcommand)]
@@ -252,8 +262,8 @@ async fn main() -> Result<()> {
         } => {
             list::execute(operations, uri.as_deref(), include_abstracts).await?;
         }
-        Commands::Get { uri, abstract_only } => {
-            get::execute(operations, &uri, abstract_only).await?;
+        Commands::Get { uri, abstract_only, overview } => {
+            get::execute(operations, &uri, abstract_only, overview).await?;
         }
         Commands::Delete { uri } => {
             delete::execute(operations, &uri).await?;
@@ -265,6 +275,9 @@ async fn main() -> Result<()> {
             SessionAction::Create { thread, title } => {
                 session::create(operations, &thread, title.as_deref()).await?;
             }
+            SessionAction::Close { thread } => {
+                session::close(operations, &thread).await?;
+            }
         },
         Commands::Stats => {
             stats::execute(operations).await?;

cortex-mem-core/Cargo.toml

@@ -23,6 +23,7 @@ tracing-subscriber = { workspace = true }
 walkdir = { workspace = true }
 rig-core = { workspace = true }
 reqwest = { workspace = true }
+log = "0.4"
 
 # Additional dependencies
 regex = "1.10"
@@ -31,7 +32,7 @@ cortex-mem-config = { path = "../cortex-mem-config" }
 schemars = "0.8"
 
 # Vector search dependencies (mandatory)
-qdrant-client = "1.11"
+qdrant-client = "1.17"
 dyn-clone = "1.0"
 
 [features]

cortex-mem-core/src/automation/auto_extract.rs

@@ -1,12 +1,10 @@
 use crate::{
     Result,
-    extraction::MemoryExtractor,
     filesystem::CortexFilesystem,
     llm::LLMClient,
-    session::SessionManager,
 };
 use std::sync::Arc;
-use tracing::{info, warn};
+use tracing::info;
 
 /// 会话自动提取配置
 #[derive(Debug, Clone)]
@@ -38,18 +36,15 @@ pub struct AutoExtractStats {
 
 /// 会话自动提取器
 ///
-/// 🔧 简化版本：移除了profile.json相关代码
-/// 现在所有记忆统一由SessionManager的MemoryExtractor处理
+/// v2.5: 此结构体已被简化，记忆提取现在由 SessionManager 通过 MemoryEventCoordinator 处理。
+/// 保留此结构体仅用于向后兼容。
 pub struct AutoExtractor {
     #[allow(dead_code)]
     filesystem: Arc<CortexFilesystem>,
     #[allow(dead_code)]
     llm: Arc<dyn LLMClient>,
     #[allow(dead_code)]
-    extractor: MemoryExtractor,
-    #[allow(dead_code)]
     config: AutoExtractConfig,
-    /// 用户ID（保留用于兼容性）
     user_id: String,
 }
 
@@ -60,13 +55,9 @@ impl AutoExtractor {
         llm: Arc<dyn LLMClient>,
         config: AutoExtractConfig,
     ) -> Self {
-        let extraction_config = crate::extraction::ExtractionConfig::default();
-        let extractor = MemoryExtractor::new(filesystem.clone(), llm.clone(), extraction_config);
-
         Self {
             filesystem,
             llm,
-            extractor,
             config,
             user_id: "default".to_string(),
         }
@@ -79,13 +70,9 @@ impl AutoExtractor {
         config: AutoExtractConfig,
         user_id: impl Into<String>,
     ) -> Self {
-        let extraction_config = crate::extraction::ExtractionConfig::default();
-        let extractor = MemoryExtractor::new(filesystem.clone(), llm.clone(), extraction_config);
-
         Self {
             filesystem,
             llm,
-            extractor,
             config,
             user_id: user_id.into(),
         }
@@ -96,51 +83,19 @@ impl AutoExtractor {
         self.user_id = user_id.into();
     }
 
-    /// 🔧 简化:extract_session现在只需要直接使用SessionManager处理即可
-    /// AutoExtractor不再负责用户记忆提取(由MemoryExtractor统一处理)
+    /// 提取会话记忆
+    ///
+    /// v2.5: 此方法已被废弃。记忆提取现在由 SessionManager::close_session 通过
+    /// MemoryEventCoordinator 异步处理。此方法返回空统计用于向后兼容。
     pub async fn extract_session(&self, _thread_id: &str) -> Result<AutoExtractStats> {
-        info!("AutoExtractor::extract_session is deprecated - all memory extraction is now handled by SessionManager::close_session");
-        warn!("Use SessionManager::close_session instead. This method returns empty stats for compatibility.");
-
+        info!(
+            "AutoExtractor::extract_session is deprecated - memory extraction is handled by MemoryEventCoordinator"
+        );
         Ok(AutoExtractStats::default())
     }
-}
-
-/// 增强SessionManager支持自动提取
-pub struct AutoSessionManager {
-    session_manager: SessionManager,
-    #[allow(dead_code)]
-    auto_extractor: AutoExtractor,
-}
-
-impl AutoSessionManager {
-    /// 创建新的自动会话管理器
-    pub fn new(
-        session_manager: SessionManager,
-        auto_extractor: AutoExtractor,
-    ) -> Self {
-        Self {
-            session_manager,
-            auto_extractor,
-        }
-    }
-
-    /// 获取内部的 SessionManager
-    pub fn session_manager(&self) -> &SessionManager {
-        &self.session_manager
-    }
-
-    /// 获取可变的 SessionManager
-    pub fn session_manager_mut(&mut self) -> &mut SessionManager {
-        &mut self.session_manager
-    }
 
-    /// 关闭会话并自动提取（增强版）
-    pub async fn close_session(&mut self, thread_id: &str) -> Result<()> {
-        // 先通过SessionManager关闭会话(触发timeline和记忆提取)
-        self.session_manager.close_session(thread_id).await?;
-
-        info!("Session {} closed with automatic memory extraction via SessionManager", thread_id);
-        Ok(())
+    /// 获取用户ID
+    pub fn user_id(&self) -> &str {
+        &self.user_id
     }
 }

cortex-mem-core/src/automation/indexer.rs

@@ -96,7 +96,7 @@ impl AutoIndexer {
                 run_id: Some(thread_id.to_string()),
                 actor_id: None,
                 role: Some(format!("{:?}", message.role)),
-                memory_type: crate::types::MemoryType::Conversational,
+                layer: "L2".to_string(),
                 hash: self.calculate_hash(&message.content),
                 importance_score: 0.5,
                 entities: vec![],
@@ -189,7 +189,7 @@ impl AutoIndexer {
                                 run_id: Some(thread_id.to_string()),
                                 actor_id: None,
                                 role: Some(format!("{:?}", message.role)),
-                                memory_type: crate::types::MemoryType::Conversational,
+                                layer: "L2".to_string(),
                                 hash: self.calculate_hash(&message.content),
                                 importance_score: 0.5,
                                 entities: vec![],
@@ -592,7 +592,11 @@ impl AutoIndexer {
                 run_id: None,
                 actor_id: None,
                 role: None,
-                memory_type: crate::types::MemoryType::Conversational,
+                layer: match layer {
+                    ContextLayer::L0Abstract => "L0",
+                    ContextLayer::L1Overview => "L1",
+                    ContextLayer::L2Detail => "L2",
+                }.to_string(),
                 hash: self.calculate_hash(content),
                 importance_score: 0.5,
                 entities: vec![],