feat: upgrade memory flush system

- Use LLM to summarize discarded context into concise daily memory entries
- Batch trim to half when exceeding max_turns/max_tokens, reducing flush frequency
- Run summarization asynchronously in background thread, no blocking on replies
- Add daily scheduled flush (23:55) as fallback for low-activity days
- Sync trimmed messages back to agent to keep context state consistent

agent/memory/__init__.py

@@ -9,6 +9,7 @@ from agent.memory.manager import MemoryManager
 from agent.memory.config import MemoryConfig, get_default_memory_config, set_global_memory_config
 from agent.memory.embedding import create_embedding_provider
 from agent.memory.conversation_store import ConversationStore, get_conversation_store
+from agent.memory.summarizer import ensure_daily_memory_file
 
 __all__ = [
     'MemoryManager',
@@ -18,4 +19,5 @@ __all__ = [
     'create_embedding_provider',
     'ConversationStore',
     'get_conversation_store',
+    'ensure_daily_memory_file',
 ]

agent/memory/config.py

@@ -48,9 +48,6 @@ class MemoryConfig:
     enable_auto_sync: bool = True
     sync_on_search: bool = True
     
-    # Memory flush config (独立于模型 context window)
-    flush_token_threshold: int = 50000  # 50K tokens 触发 flush
-    flush_turn_threshold: int = 20  # 20 轮对话触发 flush (用户+AI各一条为一轮)
     
     def get_workspace(self) -> Path:
         """Get workspace root directory"""

agent/memory/manager.py

@@ -363,76 +363,33 @@ class MemoryManager:
             size=stat.st_size
         )
     
-    def should_flush_memory(
+    def flush_memory(
         self,
-        current_tokens: int = 0
-    ) -> bool:
-        """
-        Check if memory flush should be triggered
-        
-        独立的 flush 触发机制，不依赖模型 context window。
-        使用配置中的阈值: flush_token_threshold 和 flush_turn_threshold
-        
-        Args:
-            current_tokens: Current session token count
-            
-        Returns:
-            True if memory flush should run
-        """
-        return self.flush_manager.should_flush(
-            current_tokens=current_tokens,
-            token_threshold=self.config.flush_token_threshold,
-            turn_threshold=self.config.flush_turn_threshold
-        )
-    
-    def increment_turn(self):
-        """增加对话轮数计数（每次用户消息+AI回复算一轮）"""
-        self.flush_manager.increment_turn()
-    
-    async def execute_memory_flush(
-        self,
-        agent_executor,
-        current_tokens: int,
+        messages: list,
         user_id: Optional[str] = None,
-        **executor_kwargs
+        reason: str = "threshold",
+        max_messages: int = 10,
     ) -> bool:
         """
-        Execute memory flush before compaction
-        
-        This runs a silent agent turn to write durable memories to disk.
-        Similar to clawdbot's pre-compaction memory flush.
+        Flush conversation summary to daily memory file.
         
         Args:
-            agent_executor: Async function to execute agent with prompt
-            current_tokens: Current session token count
+            messages: Conversation message list
             user_id: Optional user ID
-            **executor_kwargs: Additional kwargs for agent executor
-            
+            reason: "threshold" | "overflow" | "daily_summary"
+            max_messages: Max recent messages to include (0 = all)
+        
         Returns:
-            True if flush completed successfully
-            
-        Example:
-            >>> async def run_agent(prompt, system_prompt, silent=False):
-            ...     # Your agent execution logic
-            ...     pass
-            >>> 
-            >>> if manager.should_flush_memory(current_tokens=100000):
-            ...     await manager.execute_memory_flush(
-            ...         agent_executor=run_agent,
-            ...         current_tokens=100000
-            ...     )
+            True if content was written
         """
-        success = await self.flush_manager.execute_flush(
-            agent_executor=agent_executor,
-            current_tokens=current_tokens,
+        success = self.flush_manager.flush_from_messages(
+            messages=messages,
             user_id=user_id,
-            **executor_kwargs
+            reason=reason,
+            max_messages=max_messages,
         )
-        
         if success:
-            # Mark dirty so next search will sync the new memories
             self._dirty = True
-        
         return success
     
     def build_memory_guidance(self, lang: str = "zh", include_context: bool = True) -> str:
@@ -460,10 +417,12 @@ class MemoryManager:
 
 **背景知识**: 下方包含核心长期记忆，可直接使用。需要查找历史时，用 memory_search 搜索（搜索一次即可，不要重复）。
 
-**存储记忆**: 当用户分享重要信息时（偏好、决策、事实等），主动用 write 工具存储：
-- 长期信息 → MEMORY.md
+**主动存储**: 遇到以下情况时，主动用 edit/write 工具存储（无需告知用户）：
+- 用户要求记住的信息、个人偏好、重要决策
+- 对话中产生的重要结论、方案、约定
+- 完成复杂任务后的关键步骤和结果
+- 长期信息 → MEMORY.md（保持精简）
 - 当天笔记 → memory/{today_file}
-- 静默存储，仅在明确要求时确认
 
 **使用原则**: 自然使用记忆，就像你本来就知道。不需要生硬地提起或列举记忆，除非用户提到。"""
         else:
@@ -471,10 +430,12 @@ class MemoryManager:
 
 **Background Knowledge**: Core long-term memories below - use directly. For history, use memory_search once (don't repeat).
 
-**Store Memories**: When user shares important info (preferences, decisions, facts), proactively write:
-- Durable info → MEMORY.md  
+**Proactive Storage**: Store memories silently when:
+- User asks to remember something, shares preferences or decisions
+- Important conclusions, plans, or agreements emerge in conversation
+- Complex tasks are completed (record key steps and results)
+- Durable info → MEMORY.md (keep concise)
 - Daily notes → memory/{today_file}
-- Store silently; confirm only when explicitly requested
 
 **Usage**: Use memories naturally as if you always knew. Don't mention or list unless user explicitly asks."""
         
@@ -490,10 +451,10 @@ class MemoryManager:
         """
         Load bootstrap memory files for session start
         
-        Following clawdbot's design:
-        - Only loads MEMORY.md from workspace root (long-term curated memory)
-        - Daily files (memory/YYYY-MM-DD.md) are accessed via memory_search tool, not bootstrap
-        - User-specific MEMORY.md is also loaded if user_id provided
+        Loads:
+        1. MEMORY.md from workspace root (long-term curated memory)
+        2. User-specific MEMORY.md if user_id provided
+        3. Recent daily memory files (today + yesterday) for continuity
         
         Returns memory content WITHOUT obvious headers so it blends naturally
         into the context as background knowledge.
@@ -502,7 +463,7 @@ class MemoryManager:
             user_id: Optional user ID for user-specific memories
             
         Returns:
-            Memory content to inject into system prompt (blends naturally as background context)
+            Memory content to inject into system prompt
         """
         workspace_dir = self.config.get_workspace()
         memory_dir = self.config.get_memory_dir()
@@ -510,7 +471,6 @@ class MemoryManager:
         sections = []
         
         # 1. Load MEMORY.md from workspace root (long-term curated memory)
-        # Following clawdbot: only MEMORY.md is bootstrap, daily files use memory_search
         memory_file = Path(workspace_dir) / "MEMORY.md"
         if memory_file.exists():
             try:
@@ -518,7 +478,8 @@ class MemoryManager:
                 if content:
                     sections.append(content)
             except Exception as e:
-                print(f"Warning: Failed to read MEMORY.md: {e}")
+                from common.log import logger
+                logger.warning(f"[MemoryManager] Failed to read MEMORY.md: {e}")
         
         # 2. Load user-specific MEMORY.md if user_id provided
         if user_id:
@@ -530,15 +491,80 @@ class MemoryManager:
                     if content:
                         sections.append(content)
                 except Exception as e:
-                    print(f"Warning: Failed to read user memory: {e}")
+                    from common.log import logger
+                    logger.warning(f"[MemoryManager] Failed to read user memory: {e}")
+        
+        # 3. Load recent daily memory files (today + yesterday) for context continuity
+        recent_daily = self._load_recent_daily_memories(
+            memory_dir, user_id, days=2, max_tokens=2000
+        )
+        if recent_daily:
+            sections.append(recent_daily)
         
         if not sections:
             return ""
         
-        # Join sections without obvious headers - let memories blend naturally
-        # This makes the agent feel like it "just knows" rather than "checking memory files"
         return "\n\n".join(sections)
     
+    def _load_recent_daily_memories(
+        self,
+        memory_dir: Path,
+        user_id: Optional[str],
+        days: int = 2,
+        max_tokens: int = 2000
+    ) -> str:
+        """
+        Load recent daily memory files for bootstrap context.
+        Loads the most recent N days that have non-empty content.
+        
+        Args:
+            memory_dir: Memory directory path
+            user_id: Optional user ID
+            days: Number of recent days to load
+            max_tokens: Approximate max tokens to include (rough char estimate)
+        """
+        from common.log import logger
+        
+        daily_sections = []
+        total_chars = 0
+        max_chars = max_tokens * 4  # rough token-to-char ratio
+        
+        for i in range(days):
+            date = (datetime.now() - timedelta(days=i)).strftime("%Y-%m-%d")
+            
+            # Check user-specific daily file first, then shared
+            candidates = []
+            if user_id:
+                candidates.append(memory_dir / "users" / user_id / f"{date}.md")
+            candidates.append(memory_dir / f"{date}.md")
+            
+            for daily_file in candidates:
+                if not daily_file.exists():
+                    continue
+                try:
+                    content = daily_file.read_text(encoding='utf-8').strip()
+                    if not content or len(content) < 30:
+                        continue
+                    
+                    # Truncate if adding this would exceed limit
+                    remaining = max_chars - total_chars
+                    if remaining <= 0:
+                        break
+                    if len(content) > remaining:
+                        content = content[:remaining] + "\n...(truncated)"
+                    
+                    label = "Today" if i == 0 else "Yesterday" if i == 1 else date
+                    daily_sections.append(f"### {label} ({date})\n{content}")
+                    total_chars += len(content)
+                    break  # only load one file per date (user-specific takes priority)
+                except Exception as e:
+                    logger.warning(f"[MemoryManager] Failed to read daily memory {daily_file}: {e}")
+        
+        if not daily_sections:
+            return ""
+        
+        return "### Recent Activity\n\n" + "\n\n".join(daily_sections)
+    
     def get_status(self) -> Dict[str, Any]:
         """Get memory status"""
         stats = self.storage.get_stats()
@@ -568,6 +594,37 @@ class MemoryManager:
         content = f"{path}:{start_line}:{end_line}"
         return hashlib.md5(content.encode('utf-8')).hexdigest()
     
+    @staticmethod
+    def _compute_temporal_decay(path: str, half_life_days: float = 30.0) -> float:
+        """
+        Compute temporal decay multiplier for dated memory files.
+        
+        Inspired by OpenClaw's temporal-decay: exponential decay based on file date.
+        MEMORY.md and non-dated files are "evergreen" (no decay, multiplier=1.0).
+        Daily files like memory/2025-03-01.md decay based on age.
+        
+        Formula: multiplier = exp(-ln2/half_life * age_in_days)
+        """
+        import re
+        import math
+        
+        match = re.search(r'(\d{4})-(\d{2})-(\d{2})\.md$', path)
+        if not match:
+            return 1.0  # evergreen: MEMORY.md, non-dated files
+        
+        try:
+            file_date = datetime(
+                int(match.group(1)), int(match.group(2)), int(match.group(3))
+            )
+            age_days = (datetime.now() - file_date).days
+            if age_days <= 0:
+                return 1.0
+            
+            decay_lambda = math.log(2) / half_life_days
+            return math.exp(-decay_lambda * age_days)
+        except (ValueError, OverflowError):
+            return 1.0
+    
     def _merge_results(
         self,
         vector_results: List[SearchResult],
@@ -575,8 +632,7 @@ class MemoryManager:
         vector_weight: float,
         keyword_weight: float
     ) -> List[SearchResult]:
-        """Merge vector and keyword search results"""
-        # Create a map by (path, start_line, end_line)
+        """Merge vector and keyword search results with temporal decay for dated files"""
         merged_map = {}
         
         for result in vector_results:
@@ -598,7 +654,6 @@ class MemoryManager:
                     'keyword_score': result.score
                 }
         
-        # Calculate combined scores
         merged_results = []
         for entry in merged_map.values():
             combined_score = (
@@ -606,7 +661,11 @@ class MemoryManager:
                 keyword_weight * entry['keyword_score']
             )
             
+            # Apply temporal decay for dated memory files
             result = entry['result']
+            decay = self._compute_temporal_decay(result.path)
+            combined_score *= decay
+            
             merged_results.append(SearchResult(
                 path=result.path,
                 start_line=result.start_line,
@@ -617,6 +676,5 @@ class MemoryManager:
                 user_id=result.user_id
             ))
         
-        # Sort by score
         merged_results.sort(key=lambda r: r.score, reverse=True)
         return merged_results

agent/memory/summarizer.py

@@ -1,225 +1,324 @@
 """
 Memory flush manager
 
-Triggers memory flush before context compaction (similar to clawdbot)
+Handles memory persistence when conversation context is trimmed or overflows:
+- Uses LLM to summarize discarded messages into concise key-information entries
+- Writes to daily memory files (lazy creation)
+- Deduplicates trim flushes to avoid repeated writes
+- Runs summarization asynchronously to avoid blocking normal replies
+- Provides daily summary interface for scheduler
 """
 
-from typing import Optional, Callable, Any
+import threading
+from typing import Optional, Callable, Any, List, Dict
 from pathlib import Path
 from datetime import datetime
+from common.log import logger
+
+
+SUMMARIZE_SYSTEM_PROMPT = """你是一个记忆提取助手。你的任务是从对话记录中提取值得记住的信息，生成简洁的记忆摘要。
+
+输出要求：
+1. 以事件/关键信息为维度记录，每条一行，用 "- " 开头
+2. 记录有价值的关键信息，例如用户提出的要求及助手的解决方案，对话中涉及的事实信息，用户的偏好、决策或重要结论
+3. 每条摘要需要简明扼要，只保留关键信息
+4. 直接输出摘要内容，不要加任何前缀说明
+5. 当对话没有任何记录价值例如只是简单问候，可回复"无\""""
+
+SUMMARIZE_USER_PROMPT = """请从以下对话记录中提取关键信息，生成记忆摘要：
+
+{conversation}"""
 
 
 class MemoryFlushManager:
     """
-    Manages memory flush operations before context compaction
+    Manages memory flush operations.
     
-    Similar to clawdbot's memory flush mechanism:
-    - Triggers when context approaches token limit
-    - Runs a silent agent turn to write memories to disk
-    - Uses memory/YYYY-MM-DD.md for daily notes
-    - Uses MEMORY.md (workspace root) for long-term curated memories
+    Flush is triggered by agent_stream in two scenarios:
+    1. Context trim: _trim_messages discards old turns → flush discarded content
+    2. Context overflow: API rejects request → emergency flush before clearing
+    
+    Additionally, create_daily_summary() can be called by scheduler for end-of-day summaries.
     """
     
     def __init__(
         self,
         workspace_dir: Path,
-        llm_model: Optional[Any] = None
+        llm_model: Optional[Any] = None,
     ):
-        """
-        Initialize memory flush manager
-        
-        Args:
-            workspace_dir: Workspace directory
-            llm_model: LLM model for agent execution (optional)
-        """
         self.workspace_dir = workspace_dir
         self.llm_model = llm_model
         
         self.memory_dir = workspace_dir / "memory"
         self.memory_dir.mkdir(parents=True, exist_ok=True)
         
-        # Tracking
-        self.last_flush_token_count: Optional[int] = None
         self.last_flush_timestamp: Optional[datetime] = None
-        self.turn_count: int = 0  # 对话轮数计数器
+        self._trim_flushed_hashes: set = set()  # Content hashes of already-flushed messages
+        self._last_flushed_content_hash: str = ""  # Content hash at last flush, for daily dedup
     
-    def should_flush(
-        self,
-        current_tokens: int = 0,
-        token_threshold: int = 50000,
-        turn_threshold: int = 20
-    ) -> bool:
-        """
-        Determine if memory flush should be triggered
-        
-        独立的 flush 触发机制，不依赖模型 context window:
-        - Token 阈值: 达到 50K tokens 时触发
-        - 轮次阈值: 达到 20 轮对话时触发
-        
-        Args:
-            current_tokens: Current session token count
-            token_threshold: Token threshold to trigger flush (default: 50K)
-            turn_threshold: Turn threshold to trigger flush (default: 20)
-            
-        Returns:
-            True if flush should run
-        """
-        # 检查 token 阈值
-        if current_tokens > 0 and current_tokens >= token_threshold:
-            # 避免重复 flush
-            if self.last_flush_token_count is not None:
-                if current_tokens <= self.last_flush_token_count + 5000:
-                    return False
-            return True
-        
-        # 检查轮次阈值
-        if self.turn_count >= turn_threshold:
-            return True
-        
-        return False
-    
-    def get_today_memory_file(self, user_id: Optional[str] = None) -> Path:
-        """
-        Get today's memory file path: memory/YYYY-MM-DD.md
-        
-        Args:
-            user_id: Optional user ID for user-specific memory
-            
-        Returns:
-            Path to today's memory file
-        """
+    def get_today_memory_file(self, user_id: Optional[str] = None, ensure_exists: bool = False) -> Path:
+        """Get today's memory file path: memory/YYYY-MM-DD.md"""
         today = datetime.now().strftime("%Y-%m-%d")
         
         if user_id:
             user_dir = self.memory_dir / "users" / user_id
-            user_dir.mkdir(parents=True, exist_ok=True)
-            return user_dir / f"{today}.md"
+            if ensure_exists:
+                user_dir.mkdir(parents=True, exist_ok=True)
+            today_file = user_dir / f"{today}.md"
         else:
-            return self.memory_dir / f"{today}.md"
+            today_file = self.memory_dir / f"{today}.md"
+        
+        if ensure_exists and not today_file.exists():
+            today_file.parent.mkdir(parents=True, exist_ok=True)
+            today_file.write_text(f"# Daily Memory: {today}\n\n")
+        
+        return today_file
     
     def get_main_memory_file(self, user_id: Optional[str] = None) -> Path:
-        """
-        Get main memory file path: MEMORY.md (workspace root)
-        
-        Args:
-            user_id: Optional user ID for user-specific memory
-            
-        Returns:
-            Path to main memory file
-        """
+        """Get main memory file path: MEMORY.md (workspace root)"""
         if user_id:
             user_dir = self.memory_dir / "users" / user_id
             user_dir.mkdir(parents=True, exist_ok=True)
             return user_dir / "MEMORY.md"
         else:
-            # Return workspace root MEMORY.md
             return Path(self.workspace_dir) / "MEMORY.md"
     
-    def create_flush_prompt(self) -> str:
-        """
-        Create prompt for memory flush turn
-        
-        Similar to clawdbot's DEFAULT_MEMORY_FLUSH_PROMPT
-        """
-        today = datetime.now().strftime("%Y-%m-%d")
-        return (
-            f"Pre-compaction memory flush. "
-            f"Store durable memories now (use memory/{today}.md for daily notes; "
-            f"create memory/ if needed). "
-            f"\n\n"
-            f"重要提示:\n"
-            f"- MEMORY.md: 记录最核心、最常用的信息（例如重要规则、偏好、决策、要求等）\n"
-            f"  如果 MEMORY.md 过长，可以精简或移除不再重要的内容。避免冗长描述，用关键词和要点形式记录\n"
-            f"- memory/{today}.md: 记录当天发生的事件、关键信息、经验教训、对话过程摘要等，突出重点\n"
-            f"- 如果没有重要内容需要记录，回复 NO_REPLY\n"
-        )
-    
-    def create_flush_system_prompt(self) -> str:
-        """
-        Create system prompt for memory flush turn
-        
-        Similar to clawdbot's DEFAULT_MEMORY_FLUSH_SYSTEM_PROMPT
-        """
-        return (
-            "Pre-compaction memory flush turn. "
-            "The session is near auto-compaction; capture durable memories to disk. "
-            "\n\n"
-            "记忆写入原则:\n"
-            "1. MEMORY.md 精简原则: 只记录核心信息（<2000 tokens）\n"
-            "   - 记录重要规则、偏好、决策、要求等需要长期记住的关键信息，无需记录过多细节\n"
-            "   - 如果 MEMORY.md 过长，可以根据需要精简或删除过时内容\n"
-            "\n"
-            "2. 天级记忆 (memory/YYYY-MM-DD.md):\n"
-            "   - 记录当天的重要事件、关键信息、经验教训、对话过程摘要等，确保核心信息点被完整记录\n"
-            "\n"
-            "3. 判断标准:\n"
-            "   - 这个信息未来会经常用到吗？→ MEMORY.md\n"
-            "   - 这是今天的重要事件或决策吗？→ memory/YYYY-MM-DD.md\n"
-            "   - 这是临时性的、不重要的内容吗？→ 不记录\n"
-            "\n"
-            "You may reply, but usually NO_REPLY is correct."
-        )
-    
-    async def execute_flush(
-        self,
-        agent_executor: Callable,
-        current_tokens: int,
-        user_id: Optional[str] = None,
-        **executor_kwargs
-    ) -> bool:
-        """
-        Execute memory flush by running a silent agent turn
-        
-        Args:
-            agent_executor: Function to execute agent with prompt
-            current_tokens: Current token count
-            user_id: Optional user ID
-            **executor_kwargs: Additional kwargs for agent executor
-            
-        Returns:
-            True if flush completed successfully
-        """
-        try:
-            # Create flush prompts
-            prompt = self.create_flush_prompt()
-            system_prompt = self.create_flush_system_prompt()
-            
-            # Execute agent turn (silent, no user-visible reply expected)
-            await agent_executor(
-                prompt=prompt,
-                system_prompt=system_prompt,
-                silent=True,  # NO_REPLY expected
-                **executor_kwargs
-            )
-            
-            # Track flush
-            self.last_flush_token_count = current_tokens
-            self.last_flush_timestamp = datetime.now()
-            self.turn_count = 0  # 重置轮数计数器
-            
-            return True
-            
-        except Exception as e:
-            print(f"Memory flush failed: {e}")
-            return False
-    
-    def increment_turn(self):
-        """增加对话轮数计数"""
-        self.turn_count += 1
-    
     def get_status(self) -> dict:
-        """Get memory flush status"""
         return {
-            'last_flush_tokens': self.last_flush_token_count,
             'last_flush_time': self.last_flush_timestamp.isoformat() if self.last_flush_timestamp else None,
             'today_file': str(self.get_today_memory_file()),
             'main_file': str(self.get_main_memory_file())
         }
 
+    # ---- Flush execution (called by agent_stream or scheduler) ----
+    
+    def flush_from_messages(
+        self,
+        messages: List[Dict],
+        user_id: Optional[str] = None,
+        reason: str = "trim",
+        max_messages: int = 0,
+    ) -> bool:
+        """
+        Asynchronously summarize and flush messages to daily memory.
+        
+        Deduplication runs synchronously, then LLM summarization + file write
+        run in a background thread so the main reply flow is never blocked.
+        
+        Args:
+            messages: Conversation message list (OpenAI/Claude format)
+            user_id: Optional user ID for user-scoped memory
+            reason: Why flush was triggered ("trim" | "overflow" | "daily_summary")
+            max_messages: Max recent messages to summarize (0 = all)
+        
+        Returns:
+            True if flush was dispatched
+        """
+        try:
+            import hashlib
+            deduped = []
+            for m in messages:
+                text = self._extract_text_from_content(m.get("content", ""))
+                if not text or not text.strip():
+                    continue
+                h = hashlib.md5(text.encode("utf-8")).hexdigest()
+                if h not in self._trim_flushed_hashes:
+                    self._trim_flushed_hashes.add(h)
+                    deduped.append(m)
+            if not deduped:
+                return False
+            
+            import copy
+            snapshot = copy.deepcopy(deduped)
+            thread = threading.Thread(
+                target=self._flush_worker,
+                args=(snapshot, user_id, reason, max_messages),
+                daemon=True,
+            )
+            thread.start()
+            logger.info(f"[MemoryFlush] Async flush dispatched (reason={reason}, msgs={len(snapshot)})")
+            return True
+            
+        except Exception as e:
+            logger.warning(f"[MemoryFlush] Failed to dispatch flush (reason={reason}): {e}")
+            return False
+
+    def _flush_worker(
+        self,
+        messages: List[Dict],
+        user_id: Optional[str],
+        reason: str,
+        max_messages: int,
+    ):
+        """Background worker: summarize with LLM and write to daily file."""
+        try:
+            summary = self._summarize_messages(messages, max_messages)
+            if not summary or not summary.strip() or summary.strip() == "无":
+                logger.info(f"[MemoryFlush] No valuable content to flush (reason={reason})")
+                return
+            
+            daily_file = ensure_daily_memory_file(self.workspace_dir, user_id)
+            
+            if reason == "overflow":
+                header = f"## Context Overflow Recovery ({datetime.now().strftime('%H:%M')})"
+                note = "The following conversation was trimmed due to context overflow:\n"
+            elif reason == "trim":
+                header = f"## Trimmed Context ({datetime.now().strftime('%H:%M')})"
+                note = ""
+            elif reason == "daily_summary":
+                header = f"## Daily Summary ({datetime.now().strftime('%H:%M')})"
+                note = ""
+            else:
+                header = f"## Session Notes ({datetime.now().strftime('%H:%M')})"
+                note = ""
+            
+            flush_entry = f"\n{header}\n\n{note}{summary}\n"
+            
+            with open(daily_file, "a", encoding="utf-8") as f:
+                f.write(flush_entry)
+            
+            self.last_flush_timestamp = datetime.now()
+            
+            logger.info(f"[MemoryFlush] Wrote to {daily_file.name} (reason={reason}, chars={len(summary)})")
+            
+        except Exception as e:
+            logger.warning(f"[MemoryFlush] Async flush failed (reason={reason}): {e}")
+    
+    def create_daily_summary(
+        self,
+        messages: List[Dict],
+        user_id: Optional[str] = None
+    ) -> bool:
+        """
+        Generate end-of-day summary. Called by daily timer.
+        Skips if messages haven't changed since last flush.
+        """
+        import hashlib
+        content = "".join(
+            self._extract_text_from_content(m.get("content", ""))
+            for m in messages
+        )
+        content_hash = hashlib.md5(content.encode("utf-8")).hexdigest()
+        if content_hash == self._last_flushed_content_hash:
+            logger.debug("[MemoryFlush] Daily summary skipped: no new content since last flush")
+            return False
+        self._last_flushed_content_hash = content_hash
+        return self.flush_from_messages(
+            messages=messages,
+            user_id=user_id,
+            reason="daily_summary",
+            max_messages=0,
+        )
+    
+    # ---- Internal helpers ----
+    
+    def _summarize_messages(self, messages: List[Dict], max_messages: int = 0) -> str:
+        """
+        Summarize conversation messages using LLM, with rule-based fallback.
+        """
+        conversation_text = self._format_conversation_for_summary(messages, max_messages)
+        if not conversation_text.strip():
+            return ""
+        
+        # Try LLM summarization first
+        if self.llm_model:
+            try:
+                summary = self._call_llm_for_summary(conversation_text)
+                if summary and summary.strip() and summary.strip() != "无":
+                    return summary.strip()
+            except Exception as e:
+                logger.warning(f"[MemoryFlush] LLM summarization failed, using fallback: {e}")
+        
+        return self._extract_summary_fallback(messages, max_messages)
+
+    def _format_conversation_for_summary(self, messages: List[Dict], max_messages: int = 0) -> str:
+        """Format messages into readable conversation text for LLM summarization."""
+        msgs = messages if max_messages == 0 else messages[-max_messages * 2:]
+        lines = []
+        for msg in msgs:
+            role = msg.get("role", "")
+            text = self._extract_text_from_content(msg.get("content", ""))
+            if not text or not text.strip():
+                continue
+            text = text.strip()
+            if role == "user":
+                lines.append(f"用户: {text[:500]}")
+            elif role == "assistant":
+                lines.append(f"助手: {text[:500]}")
+        return "\n".join(lines)
+
+    def _call_llm_for_summary(self, conversation_text: str) -> str:
+        """Call LLM to generate a concise summary of the conversation."""
+        from agent.protocol.models import LLMRequest
+        
+        request = LLMRequest(
+            messages=[{"role": "user", "content": SUMMARIZE_USER_PROMPT.format(conversation=conversation_text)}],
+            temperature=0,
+            max_tokens=500,
+            stream=False,
+            system=SUMMARIZE_SYSTEM_PROMPT,
+        )
+        
+        response = self.llm_model.call(request)
+        
+        if isinstance(response, dict):
+            if response.get("error"):
+                raise RuntimeError(response.get("message", "LLM call failed"))
+            # OpenAI format
+            choices = response.get("choices", [])
+            if choices:
+                return choices[0].get("message", {}).get("content", "")
+        
+        # Handle response object with attribute access (e.g. OpenAI SDK response)
+        if hasattr(response, "choices") and response.choices:
+            return response.choices[0].message.content or ""
+        
+        return ""
+
+    @staticmethod
+    def _extract_summary_fallback(messages: List[Dict], max_messages: int = 0) -> str:
+        """Rule-based fallback when LLM is unavailable."""
+        msgs = messages if max_messages == 0 else messages[-max_messages * 2:]
+        
+        items = []
+        for msg in msgs:
+            role = msg.get("role", "")
+            text = MemoryFlushManager._extract_text_from_content(msg.get("content", ""))
+            if not text or not text.strip():
+                continue
+            text = text.strip()
+            
+            if role == "user":
+                if len(text) <= 5:
+                    continue
+                items.append(f"- 用户请求: {text[:200]}")
+            elif role == "assistant":
+                first_line = text.split("\n")[0].strip()
+                if len(first_line) > 10:
+                    items.append(f"- 处理结果: {first_line[:200]}")
+        
+        return "\n".join(items[:15])
+    
+    @staticmethod
+    def _extract_text_from_content(content) -> str:
+        """Extract plain text from message content (string or content blocks)."""
+        if isinstance(content, str):
+            return content
+        if isinstance(content, list):
+            parts = []
+            for block in content:
+                if isinstance(block, dict) and block.get("type") == "text":
+                    parts.append(block.get("text", ""))
+                elif isinstance(block, str):
+                    parts.append(block)
+            return "\n".join(parts)
+        return ""
+
 
 def create_memory_files_if_needed(workspace_dir: Path, user_id: Optional[str] = None):
     """
-    Create default memory files if they don't exist
+    Create essential memory files if they don't exist.
+    Only creates MEMORY.md; daily files are created lazily on first write.
     
     Args:
         workspace_dir: Workspace directory
@@ -228,7 +327,7 @@ def create_memory_files_if_needed(workspace_dir: Path, user_id: Optional[str] =
     memory_dir = workspace_dir / "memory"
     memory_dir.mkdir(parents=True, exist_ok=True)
     
-    # Create main MEMORY.md in workspace root
+    # Create main MEMORY.md in workspace root (always needed for bootstrap)
     if user_id:
         user_dir = memory_dir / "users" / user_id
         user_dir.mkdir(parents=True, exist_ok=True)
@@ -237,14 +336,28 @@ def create_memory_files_if_needed(workspace_dir: Path, user_id: Optional[str] =
         main_memory = Path(workspace_dir) / "MEMORY.md"
     
     if not main_memory.exists():
-        # Create empty file or with minimal structure (no obvious "Memory" header)
-        # Following clawdbot's approach: memories should blend naturally into context
         main_memory.write_text("")
+
+
+def ensure_daily_memory_file(workspace_dir: Path, user_id: Optional[str] = None) -> Path:
+    """
+    Ensure today's daily memory file exists, creating it only when actually needed.
+    Called lazily before first write to daily memory.
+    
+    Args:
+        workspace_dir: Workspace directory
+        user_id: Optional user ID for user-specific files
+        
+    Returns:
+        Path to today's memory file
+    """
+    memory_dir = workspace_dir / "memory"
+    memory_dir.mkdir(parents=True, exist_ok=True)
     
-    # Create today's memory file
     today = datetime.now().strftime("%Y-%m-%d")
     if user_id:
         user_dir = memory_dir / "users" / user_id
+        user_dir.mkdir(parents=True, exist_ok=True)
         today_memory = user_dir / f"{today}.md"
     else:
         today_memory = memory_dir / f"{today}.md"
@@ -252,5 +365,6 @@ def create_memory_files_if_needed(workspace_dir: Path, user_id: Optional[str] =
     if not today_memory.exists():
         today_memory.write_text(
             f"# Daily Memory: {today}\n\n"
-            f"Day-to-day notes and running context.\n\n"
         )
+    
+    return today_memory

agent/prompt/builder.py

@@ -280,9 +280,14 @@ def _build_memory_section(memory_manager: Any, tools: Optional[List[Any]], langu
     if not has_memory_tools:
         return []
     
+    from datetime import datetime
+    today_file = datetime.now().strftime("%Y-%m-%d") + ".md"
+    
     lines = [
         "## 记忆系统",
         "",
+        "### 检索记忆",
+        "",
         "在回答关于以前的工作、决定、日期、人物、偏好或待办事项的任何问题之前：",
         "",
         "1. 不确定记忆文件位置 → 先用 `memory_search` 通过关键词和语义检索相关内容",
@@ -290,13 +295,24 @@ def _build_memory_section(memory_manager: Any, tools: Optional[List[Any]], langu
         "3. search 无结果 → 尝试用 `memory_get` 读取MEMORY.md及最近两天记忆文件",
         "",
         "**记忆文件结构**:",
-        "- `MEMORY.md`: 长期记忆（核心信息、偏好、决策等）",
-        "- `memory/YYYY-MM-DD.md`: 每日记忆，记录当天的事件和对话信息",
+        f"- `MEMORY.md`: 长期记忆（核心信息、偏好、决策等）",
+        f"- `memory/YYYY-MM-DD.md`: 每日记忆，今天是 `memory/{today_file}`",
         "",
-        "**写入记忆**:",
+        "### 写入记忆",
+        "",
+        "**主动存储**：遇到以下情况时，应主动将信息写入记忆文件（无需告知用户）：",
+        "",
+        "- 用户明确要求你记住某些信息",
+        "- 用户分享了重要的个人偏好、习惯、决策",
+        "- 对话中产生了重要的结论、方案、约定",
+        "- 完成了复杂任务，值得记录关键步骤和结果",
+        "- 发现了用户经常遇到的问题或解决方案",
+        "",
+        "**存储规则**:",
+        f"- 长期有效的核心信息 → `MEMORY.md`（文件保持精简，< 2000 tokens）",
+        f"- 当天的事件、进展、笔记 → `memory/{today_file}`",
         "- 追加内容 → `edit` 工具，oldText 留空",
         "- 修改内容 → `edit` 工具，oldText 填写要替换的文本",
-        "- 新建文件 → `write` 工具",
         "- **禁止写入敏感信息**：API密钥、令牌等敏感信息严禁写入记忆文件",
         "",
         "**使用原则**: 自然使用记忆，就像你本来就知道；不用刻意提起，除非用户问起。",
@@ -369,7 +385,7 @@ def _build_workspace_section(workspace_dir: str, language: str, is_first_convers
         "",
         "**交流规范**:",
         "",
-        "- 在对话中，不要直接输出工作空间中的技术细节，特别是不要输出 AGENT.md、USER.md、MEMORY.md 等文件名称",
+        "- 在对话中，无需直接输出工作空间中的技术细节，例如 AGENT.md、USER.md、MEMORY.md 等文件名称",
         "- 例如用自然表达例如「我已记住」而不是「已更新 MEMORY.md」",
         "",
     ]

agent/protocol/agent.py

@@ -544,11 +544,15 @@ class Agent:
                     logger.info("[Agent] Cleared Agent message history after executor recovery")
             raise
 
-        # Append only the NEW messages from this execution (thread-safe)
-        # This allows concurrent requests to both contribute to history
+        # Sync executor's messages back to agent (thread-safe).
+        # If the executor trimmed context, its message list is shorter than
+        # original_length, so we must replace rather than append.
         with self.messages_lock:
-            new_messages = executor.messages[original_length:]
-            self.messages.extend(new_messages)
+            self.messages = list(executor.messages)
+            # Track messages added in this run (user query + all assistant/tool messages)
+            # original_length may exceed executor.messages length after trimming
+            trim_adjusted_start = min(original_length, len(executor.messages))
+            self._last_run_new_messages = list(executor.messages[trim_adjusted_start:])
         
         # Store executor reference for agent_bridge to access files_to_send
         self.stream_executor = executor

agent/protocol/agent_stream.py

@@ -201,26 +201,6 @@ class AgentStreamExecutor:
                 logger.info(f"[Agent] 第 {turn} 轮")
                 self._emit_event("turn_start", {"turn": turn})
 
-                # Check if memory flush is needed (before calling LLM)
-                # 使用独立的 flush 阈值（50K tokens 或 20 轮）
-                if self.agent.memory_manager and hasattr(self.agent, 'last_usage'):
-                    usage = self.agent.last_usage
-                    if usage and 'input_tokens' in usage:
-                        current_tokens = usage.get('input_tokens', 0)
-
-                        if self.agent.memory_manager.should_flush_memory(
-                                current_tokens=current_tokens
-                        ):
-                            self._emit_event("memory_flush_start", {
-                                "current_tokens": current_tokens,
-                                "turn_count": self.agent.memory_manager.flush_manager.turn_count
-                            })
-
-                            # TODO: Execute memory flush in background
-                            # This would require async support
-                            logger.info(
-                                f"Memory flush recommended: tokens={current_tokens}, turns={self.agent.memory_manager.flush_manager.turn_count}")
-
                 # Call LLM (enable retry_on_empty for better reliability)
                 assistant_msg, tool_calls = self._call_llm_stream(retry_on_empty=True)
                 final_response = assistant_msg
@@ -473,10 +453,6 @@ class AgentStreamExecutor:
             logger.info(f"[Agent] 🏁 完成 ({turn}轮)")
             self._emit_event("agent_end", {"final_response": final_response})
 
-            # 每轮对话结束后增加计数（用户消息+AI回复=1轮）
-            if self.agent.memory_manager:
-                self.agent.memory_manager.increment_turn()
-
         return final_response
 
     def _call_llm_stream(self, retry_on_empty=True, retry_count=0, max_retries=3,
@@ -501,7 +477,8 @@ class AgentStreamExecutor:
 
         # Prepare messages
         messages = self._prepare_messages()
-        logger.info(f"Sending {len(messages)} messages to LLM")
+        turns = self._identify_complete_turns()
+        logger.info(f"Sending {len(messages)} messages ({len(turns)} turns) to LLM")
 
         # Prepare tool definitions (OpenAI/Claude format)
         tools_schema = None
@@ -655,6 +632,14 @@ class AgentStreamExecutor:
                 error_type = "context overflow" if is_context_overflow else "message format error"
                 logger.error(f"💥 {error_type} detected: {e}")
 
+                # Flush memory before trimming to preserve context that will be lost
+                if is_context_overflow and self.agent.memory_manager:
+                    user_id = getattr(self.agent, '_current_user_id', None)
+                    self.agent.memory_manager.flush_memory(
+                        messages=self.messages, user_id=user_id,
+                        reason="overflow", max_messages=0
+                    )
+
                 # Strategy: try aggressive trimming first, only clear as last resort
                 if is_context_overflow and not _overflow_retry:
                     trimmed = self._aggressive_trim_for_overflow()
@@ -1204,14 +1189,28 @@ class AgentStreamExecutor:
         if not turns:
             return
         
-        # Step 2: 轮次限制 - 保留最近 N 轮
+        # Step 2: 轮次限制 - 超出时裁到 max_turns/2，批量 flush 被裁的轮次
         if len(turns) > self.max_context_turns:
-            removed_turns = len(turns) - self.max_context_turns
-            turns = turns[-self.max_context_turns:]  # 保留最近的轮次
+            keep_count = max(1, self.max_context_turns // 2)
+            removed_count = len(turns) - keep_count
+            
+            # Flush discarded turns to daily memory
+            if self.agent.memory_manager:
+                discarded_messages = []
+                for turn in turns[:removed_count]:
+                    discarded_messages.extend(turn["messages"])
+                if discarded_messages:
+                    user_id = getattr(self.agent, '_current_user_id', None)
+                    self.agent.memory_manager.flush_memory(
+                        messages=discarded_messages, user_id=user_id,
+                        reason="trim", max_messages=0
+                    )
+            
+            turns = turns[-keep_count:]
             
             logger.info(
-                f"💾 上下文轮次超限: {len(turns) + removed_turns} > {self.max_context_turns}，"
-                f"移除最早的 {removed_turns} 轮完整对话"
+                f"💾 上下文轮次超限: {keep_count + removed_count} > {self.max_context_turns}，"
+                f"裁剪至 {keep_count} 轮（移除 {removed_count} 轮）"
             )
 
         # Step 3: Token 限制 - 保留完整轮次
@@ -1248,56 +1247,41 @@ class AgentStreamExecutor:
                 logger.info(f"   重建消息列表: {old_count} -> {len(self.messages)} 条消息")
             return
 
-        # Token limit exceeded - keep complete turns from newest
+        # Token limit exceeded - keep the latest half of turns (same strategy as turn limit)
+        keep_count = max(1, len(turns) // 2)
+        removed_count = len(turns) - keep_count
+        kept_turns = turns[-keep_count:]
+        kept_tokens = sum(self._estimate_turn_tokens(t) for t in kept_turns)
+
         logger.info(
             f"🔄 上下文tokens超限: ~{current_tokens + system_tokens} > {max_tokens}，"
-            f"将按完整轮次移除最早的对话"
+            f"裁剪至 {keep_count} 轮（移除 {removed_count} 轮）"
         )
 
-        # 从最新轮次开始，反向累加（保持完整轮次）
-        kept_turns = []
-        accumulated_tokens = 0
-        min_turns = 3  # 尽量保留至少 3 轮，但不强制（避免超出 token 限制）
+        # Flush discarded turns to daily memory
+        if self.agent.memory_manager:
+            discarded_messages = []
+            for turn in turns[:removed_count]:
+                discarded_messages.extend(turn["messages"])
+            if discarded_messages:
+                user_id = getattr(self.agent, '_current_user_id', None)
+                self.agent.memory_manager.flush_memory(
+                    messages=discarded_messages, user_id=user_id,
+                    reason="trim", max_messages=0
+                )
         
-        for i, turn in enumerate(reversed(turns)):
-            turn_tokens = self._estimate_turn_tokens(turn)
-            turns_from_end = i + 1
-            
-            # 检查是否超出限制
-            if accumulated_tokens + turn_tokens <= available_tokens:
-                kept_turns.insert(0, turn)
-                accumulated_tokens += turn_tokens
-            else:
-                # 超出限制
-                # 如果还没有保留足够的轮次，且这是最后的机会，尝试保留
-                if len(kept_turns) < min_turns and turns_from_end <= min_turns:
-                    # 检查是否严重超出（超出 20% 以上则放弃）
-                    overflow_ratio = (accumulated_tokens + turn_tokens - available_tokens) / available_tokens
-                    if overflow_ratio < 0.2:  # 允许最多超出 20%
-                        kept_turns.insert(0, turn)
-                        accumulated_tokens += turn_tokens
-                        logger.debug(f"   为保留最少轮次，允许超出 {overflow_ratio*100:.1f}%")
-                        continue
-                # 停止保留更早的轮次
-                break
-        
-        # 重建消息列表
         new_messages = []
         for turn in kept_turns:
             new_messages.extend(turn['messages'])
         
         old_count = len(self.messages)
-        old_turn_count = len(turns)
         self.messages = new_messages
-        new_count = len(self.messages)
-        new_turn_count = len(kept_turns)
         
-        if old_count > new_count:
-            logger.info(
-                f"   移除了 {old_turn_count - new_turn_count} 轮对话 "
-                f"({old_count} -> {new_count} 条消息，"
-                f"~{current_tokens + system_tokens} -> ~{accumulated_tokens + system_tokens} tokens)"
-            )
+        logger.info(
+            f"   移除了 {removed_count} 轮对话 "
+            f"({old_count} -> {len(self.messages)} 条消息，"
+            f"~{current_tokens + system_tokens} -> ~{kept_tokens + system_tokens} tokens)"
+        )
 
     def _clear_session_db(self):
         """

bridge/agent_bridge.py

@@ -375,10 +375,6 @@ class AgentBridge:
             # Store session_id on agent so executor can clear DB on fatal errors
             agent._current_session_id = session_id
 
-            # Record message count before execution so we can diff new messages
-            with agent.messages_lock:
-                pre_run_len = len(agent.messages)
-
             try:
                 # Use agent's run_stream method with event handler
                 response = agent.run_stream(
@@ -397,19 +393,19 @@ class AgentBridge:
             # Persist new messages generated during this run
             if session_id:
                 channel_type = (context.get("channel_type") or "") if context else ""
-                with agent.messages_lock:
-                    new_messages = agent.messages[pre_run_len:]
+                new_messages = getattr(agent, '_last_run_new_messages', [])
                 if new_messages:
                     self._persist_messages(session_id, list(new_messages), channel_type)
-                elif pre_run_len > 0 and len(agent.messages) == 0:
-                    # Agent cleared its messages (recovery from format error / overflow)
-                    # Also clear the DB to prevent reloading dirty data
-                    try:
-                        from agent.memory import get_conversation_store
-                        get_conversation_store().clear_session(session_id)
-                        logger.info(f"[AgentBridge] Cleared DB for recovered session: {session_id}")
-                    except Exception as e:
-                        logger.warning(f"[AgentBridge] Failed to clear DB after recovery: {e}")
+                else:
+                    with agent.messages_lock:
+                        msg_count = len(agent.messages)
+                    if msg_count == 0:
+                        try:
+                            from agent.memory import get_conversation_store
+                            get_conversation_store().clear_session(session_id)
+                            logger.info(f"[AgentBridge] Cleared DB for recovered session: {session_id}")
+                        except Exception as e:
+                            logger.warning(f"[AgentBridge] Failed to clear DB after recovery: {e}")
             
             # Check if there are files to send (from read tool)
             if hasattr(agent, 'stream_executor') and hasattr(agent.stream_executor, 'files_to_send'):

bridge/agent_initializer.py

@@ -115,14 +115,19 @@ class AgentInitializer:
             runtime_info=runtime_info  # Pass runtime_info for dynamic time updates
         )
         
-        # Attach memory manager
+        # Attach memory manager and share LLM model for summarization
         if memory_manager:
             agent.memory_manager = memory_manager
+            if hasattr(agent, 'model') and agent.model:
+                memory_manager.flush_manager.llm_model = agent.model
 
         # Restore persisted conversation history for this session
         if session_id:
             self._restore_conversation_history(agent, session_id)
 
+        # Start daily memory flush timer (once, on first agent init regardless of session)
+        self._start_daily_flush_timer()
+
         return agent
 
     def _restore_conversation_history(self, agent, session_id: str) -> None:
@@ -514,3 +519,59 @@ class AgentInitializer:
                 logger.info(f"[AgentInitializer] Migrated {len(keys_to_migrate)} API keys to .env: {list(keys_to_migrate.keys())}")
             except Exception as e:
                 logger.warning(f"[AgentInitializer] Failed to migrate API keys: {e}")
+
+    def _start_daily_flush_timer(self):
+        """Start a background thread that flushes all agents' memory daily at 23:55."""
+        if getattr(self.agent_bridge, '_daily_flush_started', False):
+            return
+        self.agent_bridge._daily_flush_started = True
+
+        import threading
+
+        def _daily_flush_loop():
+            while True:
+                try:
+                    now = datetime.datetime.now()
+                    target = now.replace(hour=23, minute=55, second=0, microsecond=0)
+                    if target <= now:
+                        target += datetime.timedelta(days=1)
+                    wait_seconds = (target - now).total_seconds()
+                    logger.info(f"[DailyFlush] Next flush at {target.strftime('%Y-%m-%d %H:%M')} (in {wait_seconds/3600:.1f}h)")
+                    time.sleep(wait_seconds)
+
+                    self._flush_all_agents()
+                except Exception as e:
+                    logger.warning(f"[DailyFlush] Error in daily flush loop: {e}")
+                    time.sleep(3600)
+
+        t = threading.Thread(target=_daily_flush_loop, daemon=True)
+        t.start()
+
+    def _flush_all_agents(self):
+        """Flush memory for all active agent sessions."""
+        agents = []
+        if self.agent_bridge.default_agent:
+            agents.append(("default", self.agent_bridge.default_agent))
+        for sid, agent in self.agent_bridge.agents.items():
+            agents.append((sid, agent))
+
+        if not agents:
+            return
+
+        flushed = 0
+        for label, agent in agents:
+            try:
+                if not agent.memory_manager:
+                    continue
+                with agent.messages_lock:
+                    messages = list(agent.messages)
+                if not messages:
+                    continue
+                result = agent.memory_manager.flush_manager.create_daily_summary(messages)
+                if result:
+                    flushed += 1
+            except Exception as e:
+                logger.warning(f"[DailyFlush] Failed for session {label}: {e}")
+
+        if flushed:
+            logger.info(f"[DailyFlush] Flushed {flushed}/{len(agents)} agent session(s)")