Release/v0.2.3 (#355)

* feat(web): add PageEmpty component

* feat(web): add PageTabs component

* feat(web): add PageEmpty component

* feat(web): add PageTabs component

* feat(prompt): add history tracking for prompt releases

* feat(web): add prompt menu

* refactor: The PageScrollList component supports two generic parameters

* feat(web): BodyWrapper compoent update PageLoading

* feat(web): add Ontology menu

* feat(web): memory management add scene

* feat(tasks): add celery task configuration for periodic jobs

- Add ignore_result=True to prevent storing results for periodic tasks
- Set max_retries=0 to skip failed periodic tasks without retry attempts
- Configure acks_late=False for immediate acknowledgment in beat tasks
- Add time_limit and soft_time_limit to regenerate_memory_cache task (3600s/3300s)
- Add time_limit and soft_time_limit to workspace_reflection_task (300s/240s)
- Add time_limit and soft_time_limit to run_forgetting_cycle_task (7200s/7000s)
- Improve task reliability and resource management for scheduled jobs

* feat(sandbox): add Node.js code execution support to sandbox

* Release/v0.2.2 (#260)

* [modify] migration script

* [add] migration script

* fix(web): change form message

* fix(web): the memoryContent field is compatible with numbers and strings

* feat(web): code node hidden

* fix(model):
1. create a basic model to check if the name and provider are duplicated.
2. The result shows error models because the provider created API Keys for all matching models.

---------

Co-authored-by: Mark <zhuwenhui5566@163.com>
Co-authored-by: zhaoying <yzhao96@best-inc.com>
Co-authored-by: yingzhao <zhaoyingyz@126.com>
Co-authored-by: Timebomb2018 <18868801967@163.com>

* Feature/ontology class clean (#249)

* [add] Complete ontology engineering feature implementation

* [add] Add ontology feature integration and validation utilities

* [add] Add OWL validator and validation utilities

* [fix] Add missing render_ontology_extraction_prompt function

* [fix]Add dependencies, fix functionality

* [add] migration script

* feat(celery): add dedicated periodic tasks worker and queue (#261)

* fix(web): conflict resolve

* Fix/v022 bug (#263)

* [fix]Fix the issue of inconsistent language in explicit and episodic memory.

* [fix]Fix the issue of inconsistent language in explicit and episodic memory.

* [add]Add scene_id

* [fix]Based on the AI review to fix the code

* Fix/develop memory reflex (#265)

* 遗漏的历史映射

* 遗漏的历史映射

* 反思后台报错处理

* [add] migration script

* fix: chat conversation_id add node_start

* feat(web): show code node

* fix(web): Restructure the CustomSelect component, repair the interface that is called multiple times when the form is updated

* feat(web): RadioGroupCard support block mode

* feat(web): create space add icon

* feat(app and model): token consumption statistics

* Add/develop memory (#264)

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 遗漏的历史映射

* 新增长期记忆功能

* 新增长期记忆功能

* 新增长期记忆功能

* 知识库检索多余字段

* 长期

* feat(app and model): token consumption statistics of the cluster

* memory_BUG_fix

* fix(web): prompt history remove pageLoading

* fix(prompt): remove hard-coded import of prompt file paths (#279)

* Fix/develop memory bug (#274)

* 遗漏的历史映射

* 遗漏的历史映射

* fix_timeline_memories

* fix(web): update retrieve_type key

* Fix/develop memory bug (#276)

* 遗漏的历史映射

* 遗漏的历史映射

* fix_timeline_memories

* fix_timeline_memories

* write_gragp/bug_fix

* write_gragp/bug_fix

* write_gragp/bug_fix

* chore(celery): disable periodic task scheduling

* fix(prompt): remove hard-coded import of prompt file paths

---------

Co-authored-by: lixinyue11 <94037597+lixinyue11@users.noreply.github.com>
Co-authored-by: zhaoying <yzhao96@best-inc.com>
Co-authored-by: yingzhao <zhaoyingyz@126.com>
Co-authored-by: Ke Sun <kesun5@illinois.edu>

* fix(web): remove delete confirm content

* refactor(workflow): relocate template directory into workflow

* feat(memory): add long-term storage task routing and batching

* fix(web): PageScrollList loading update

* fix(web): PageScrollList loading update

* Ontology v1 bug (#291)

* [changes]Add 'id' as the secondary sorting key, and 'scene_id' now returns a UUID object

* [fix]Fix the "end_user" return to be sorted by update time.

* [fix]Set the default values of the memory configuration model based on the spatial model.

* [fix]Remove the entity extraction check combination model, read the configuration list, and add the return of scene_id

* [fix]Fix the "end_user" return to be sorted by update time.

* [fix]

* fix(memory): add Redis session validation

- Add macOS fork() safety configuration in celery_app.py to prevent initialization issues
- Add null/False checks for Redis session queries in term_memory_save to handle missing sessions gracefully
- Add null/False checks in memory_long_term_storage to prevent processing empty Redis results
- Add null/False checks in aggregate_judgment before format_parsing to avoid errors on missing data
- Initialize redis_messages variable in window_dialogue for consistency
- Add debug logging when no existing session found in Redis for better troubleshooting
- Add TODO comments for magic numbers (scope=6, time=5) to be extracted as constants
- Improve error handling when Redis returns False or empty results instead of crashing

* fix(web): PageScrollList style update

* fix(workflow): fix argument passing in code execution nodes

* fix(web): prompt add disabled

* fix(web): space icon required

* feat(app): modify the key of the token

* fix(fix the key of the app's token):

* fix(workflow): switch code input encoding to base64+URL encoding

* [add]The main project adds multi-API Key load balancing.

* [changes]Attribute security access, secure numerical conversion, unified use of local variables

* fix(web): save add session update

* fix(web): language editor support paste

* [changes]Active status filtering logic, API Key selection strategy

* memory_BUG

* memory_BUG_long_term

* [changes]

* memory_BUG_long_term

* memory_BUG_long_term

* Fix/release memory bug (#306)

* memory_BUG_fix

* memory_BUG

* memory_BUG_long_term

* memory_BUG_long_term

* memory_BUG_long_term

* knowledge_retrieval/bug/fix

* knowledge_retrieval/bug/fix

* knowledge_retrieval/bug/fix

* [fix]1.The "read_all_config" interface returns "scene_name";2.Memory configuration for lightweight query ontology scenarios

* fix(web): replace code editor

* [changes]Modify the description of the time for the recent event

* [changes]Modify the code based on the AI review

* feat(web): update memory config ontology api

* fix(web): ui update

* knowledge_retrieval/bug/fix

* knowledge_retrieval/bug/fix

* knowledge_retrieval/bug/fix

* feat(workflow): add token usage statistics for question classifier and parameter extraction

* feat(web): move prompt menu

* Multiple independent transactions - single transaction

* Multiple independent transactions - single transaction

* Multiple independent transactions - single transaction

* Multiple independent transactions - single transaction

* Write Missing None (#321)

* Write Missing None

* Write Missing None

* Write Missing None

* Apply suggestion from @sourcery-ai[bot]

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Write Missing None

---------

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Fix/release memory bug (#324)

* Write Missing None

* Write Missing None

* Write Missing None

* Apply suggestion from @sourcery-ai[bot]

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Write Missing None

* redis update

* redis update

* redis update

* redis update

---------

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Fix/writer memory bug (#326)

* [fix]Fix the bug

* [fix]Fix the bug

* [fix]Correct the direction indication.

* fix(web): markdown table ui update

* Fix/release memory bug (#332)

* Write Missing None

* Write Missing None

* Write Missing None

* Apply suggestion from @sourcery-ai[bot]

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Write Missing None

* redis update

* redis update

* redis update

* redis update

* writer_dup_bug/fix

---------

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Fix/fact summary (#333)

* [fix]Disable the contents related to fact_summary

* [fix]Disable the contents related to fact_summary

* [fix]Modify the code based on the AI review

* Fix/release memory bug (#335)

* Write Missing None

* Write Missing None

* Write Missing None

* Apply suggestion from @sourcery-ai[bot]

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Write Missing None

* redis update

* redis update

* redis update

* redis update

* writer_dup_bug/fix

* writer_graph_bug/fix

* writer_graph_bug/fix

---------

Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

* Revert "feat(web): move prompt menu"

This reverts commit 9e6e8f50f8.

* fix(web): ui update

* fix(web): update text

* fix(web): ui update

* fix(model): change the "vl" model type of dashscope to "chat"

* fix(model): change the "vl" model type of dashscope to "chat"

---------

Co-authored-by: zhaoying <yzhao96@best-inc.com>
Co-authored-by: Eternity <1533512157@qq.com>
Co-authored-by: Mark <zhuwenhui5566@163.com>
Co-authored-by: yingzhao <zhaoyingyz@126.com>
Co-authored-by: Timebomb2018 <18868801967@163.com>
Co-authored-by: 乐力齐 <162269739+lanceyq@users.noreply.github.com>
Co-authored-by: lixinyue11 <94037597+lixinyue11@users.noreply.github.com>
Co-authored-by: lixinyue <2569494688@qq.com>
Co-authored-by: Eternity <61316157+myhMARS@users.noreply.github.com>
Co-authored-by: lanceyq <1982376970@qq.com>
Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>

.gitignore

@@ -35,3 +35,6 @@ nltk_data/
 tika-server*.jar*
 cl100k_base.tiktoken
 libssl*.deb
+
+sandbox/lib/seccomp_python/target
+sandbox/lib/seccomp_nodejs/target

README.md

@@ -334,7 +334,13 @@ step6: Log In to the Frontend Interface.
 ## License
 This project is licensed under the Apache License 2.0. For details, see the LICENSE file.
 
-## Acknowledgements & Community
-- Feedback & Issues: Please submit an Issue in the repository for bug reports or discussions.
-- Contributions Welcome: When submitting a Pull Request, please create a feature branch and follow conventional commit message guidelines.
-- Contact: If you are interested in contributing or collaborating, feel free to reach out at tianyou_hubm@redbearai.com
+## Community & Support
+
+Join our community to ask questions, share your work, and connect with fellow developers.
+
+- **GitHub Issues**: Report bugs, request features, or track known issues via [GitHub Issues](https://github.com/SuanmoSuanyangTechnology/MemoryBear/issues).
+- **GitHub Pull Requests**: Contribute code improvements or fixes through [Pull Requests](https://github.com/SuanmoSuanyangTechnology/MemoryBear/pulls).
+- **GitHub Discussions**: Ask questions, share ideas, and engage with the community in [GitHub Discussions](https://github.com/SuanmoSuanyangTechnology/MemoryBear/discussions).
+- **WeChat**: Scan the QR code below to join our WeChat community group.
+- ![wecom-temp-114020-47fe87a75da439f09f5dc93a01593046](https://github.com/user-attachments/assets/8c81885c-4134-40d5-96e2-7f78cc082dc6)
+- **Contact**: If you are interested in contributing or collaborating, feel free to reach out at tianyou_hubm@redbearai.com

api/app/cache/__init__.py

@@ -0,0 +1,11 @@
+"""
+Cache 缓存模块
+
+提供各种缓存功能的统一入口
+"""
+from .memory import EmotionMemoryCache, ImplicitMemoryCache
+
+__all__ = [
+    "EmotionMemoryCache",
+    "ImplicitMemoryCache",
+]

api/app/cache/memory/__init__.py

@@ -0,0 +1,12 @@
+"""
+Memory 缓存模块
+
+提供记忆系统相关的缓存功能
+"""
+from .emotion_memory import EmotionMemoryCache
+from .implicit_memory import ImplicitMemoryCache
+
+__all__ = [
+    "EmotionMemoryCache",
+    "ImplicitMemoryCache",
+]

api/app/cache/memory/emotion_memory.py

@@ -0,0 +1,134 @@
+"""
+Emotion Suggestions Cache
+
+情绪个性化建议缓存模块
+用于缓存用户的情绪个性化建议数据
+"""
+import json
+import logging
+from typing import Optional, Dict, Any
+from datetime import datetime
+
+from app.aioRedis import aio_redis
+
+logger = logging.getLogger(__name__)
+
+
+class EmotionMemoryCache:
+    """情绪建议缓存类"""
+    
+    # Key 前缀
+    PREFIX = "cache:memory:emotion_memory"
+    
+    @classmethod
+    def _get_key(cls, *parts: str) -> str:
+        """生成 Redis key
+        
+        Args:
+            *parts: key 的各个部分
+            
+        Returns:
+            完整的 Redis key
+        """
+        return ":".join([cls.PREFIX] + list(parts))
+    
+    @classmethod
+    async def set_emotion_suggestions(
+        cls,
+        user_id: str,
+        suggestions_data: Dict[str, Any],
+        expire: int = 86400
+    ) -> bool:
+        """设置用户情绪建议缓存
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            suggestions_data: 建议数据字典，包含：
+                - health_summary: 健康状态摘要
+                - suggestions: 建议列表
+                - generated_at: 生成时间（可选）
+            expire: 过期时间（秒），默认24小时（86400秒）
+            
+        Returns:
+            是否设置成功
+        """
+        try:
+            key = cls._get_key("suggestions", user_id)
+            
+            # 添加生成时间戳
+            if "generated_at" not in suggestions_data:
+                suggestions_data["generated_at"] = datetime.now().isoformat()
+            
+            # 添加缓存标记
+            suggestions_data["cached"] = True
+            
+            value = json.dumps(suggestions_data, ensure_ascii=False)
+            await aio_redis.set(key, value, ex=expire)
+            logger.info(f"设置情绪建议缓存成功: {key}, 过期时间: {expire}秒")
+            return True
+        except Exception as e:
+            logger.error(f"设置情绪建议缓存失败: {e}", exc_info=True)
+            return False
+    
+    @classmethod
+    async def get_emotion_suggestions(cls, user_id: str) -> Optional[Dict[str, Any]]:
+        """获取用户情绪建议缓存
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            
+        Returns:
+            建议数据字典，如果不存在或已过期返回 None
+        """
+        try:
+            key = cls._get_key("suggestions", user_id)
+            value = await aio_redis.get(key)
+            
+            if value:
+                data = json.loads(value)
+                logger.info(f"成功获取情绪建议缓存: {key}")
+                return data
+            
+            logger.info(f"情绪建议缓存不存在或已过期: {key}")
+            return None
+        except Exception as e:
+            logger.error(f"获取情绪建议缓存失败: {e}", exc_info=True)
+            return None
+    
+    @classmethod
+    async def delete_emotion_suggestions(cls, user_id: str) -> bool:
+        """删除用户情绪建议缓存
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            
+        Returns:
+            是否删除成功
+        """
+        try:
+            key = cls._get_key("suggestions", user_id)
+            result = await aio_redis.delete(key)
+            logger.info(f"删除情绪建议缓存: {key}, 结果: {result}")
+            return result > 0
+        except Exception as e:
+            logger.error(f"删除情绪建议缓存失败: {e}", exc_info=True)
+            return False
+    
+    @classmethod
+    async def get_suggestions_ttl(cls, user_id: str) -> int:
+        """获取情绪建议缓存的剩余过期时间
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            
+        Returns:
+            剩余秒数，-1表示永不过期，-2表示key不存在
+        """
+        try:
+            key = cls._get_key("suggestions", user_id)
+            ttl = await aio_redis.ttl(key)
+            logger.debug(f"情绪建议缓存TTL: {key} = {ttl}秒")
+            return ttl
+        except Exception as e:
+            logger.error(f"获取情绪建议缓存TTL失败: {e}")
+            return -2

api/app/cache/memory/implicit_memory.py

@@ -0,0 +1,136 @@
+"""
+Implicit Memory Profile Cache
+
+隐式记忆用户画像缓存模块
+用于缓存用户的完整画像数据（偏好标签、四维画像、兴趣领域、行为习惯）
+"""
+import json
+import logging
+from typing import Optional, Dict, Any
+from datetime import datetime
+
+from app.aioRedis import aio_redis
+
+logger = logging.getLogger(__name__)
+
+
+class ImplicitMemoryCache:
+    """隐式记忆用户画像缓存类"""
+    
+    # Key 前缀
+    PREFIX = "cache:memory:implicit_memory"
+    
+    @classmethod
+    def _get_key(cls, *parts: str) -> str:
+        """生成 Redis key
+        
+        Args:
+            *parts: key 的各个部分
+            
+        Returns:
+            完整的 Redis key
+        """
+        return ":".join([cls.PREFIX] + list(parts))
+    
+    @classmethod
+    async def set_user_profile(
+        cls,
+        user_id: str,
+        profile_data: Dict[str, Any],
+        expire: int = 86400
+    ) -> bool:
+        """设置用户完整画像缓存
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            profile_data: 画像数据字典，包含：
+                - preferences: 偏好标签列表
+                - portrait: 四维画像对象
+                - interest_areas: 兴趣领域分布对象
+                - habits: 行为习惯列表
+                - generated_at: 生成时间（可选）
+            expire: 过期时间（秒），默认24小时（86400秒）
+            
+        Returns:
+            是否设置成功
+        """
+        try:
+            key = cls._get_key("profile", user_id)
+            
+            # 添加生成时间戳
+            if "generated_at" not in profile_data:
+                profile_data["generated_at"] = datetime.now().isoformat()
+            
+            # 添加缓存标记
+            profile_data["cached"] = True
+            
+            value = json.dumps(profile_data, ensure_ascii=False)
+            await aio_redis.set(key, value, ex=expire)
+            logger.info(f"设置用户画像缓存成功: {key}, 过期时间: {expire}秒")
+            return True
+        except Exception as e:
+            logger.error(f"设置用户画像缓存失败: {e}", exc_info=True)
+            return False
+    
+    @classmethod
+    async def get_user_profile(cls, user_id: str) -> Optional[Dict[str, Any]]:
+        """获取用户完整画像缓存
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            
+        Returns:
+            画像数据字典，如果不存在或已过期返回 None
+        """
+        try:
+            key = cls._get_key("profile", user_id)
+            value = await aio_redis.get(key)
+            
+            if value:
+                data = json.loads(value)
+                logger.info(f"成功获取用户画像缓存: {key}")
+                return data
+            
+            logger.info(f"用户画像缓存不存在或已过期: {key}")
+            return None
+        except Exception as e:
+            logger.error(f"获取用户画像缓存失败: {e}", exc_info=True)
+            return None
+    
+    @classmethod
+    async def delete_user_profile(cls, user_id: str) -> bool:
+        """删除用户完整画像缓存
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            
+        Returns:
+            是否删除成功
+        """
+        try:
+            key = cls._get_key("profile", user_id)
+            result = await aio_redis.delete(key)
+            logger.info(f"删除用户画像缓存: {key}, 结果: {result}")
+            return result > 0
+        except Exception as e:
+            logger.error(f"删除用户画像缓存失败: {e}", exc_info=True)
+            return False
+    
+    @classmethod
+    async def get_profile_ttl(cls, user_id: str) -> int:
+        """获取用户画像缓存的剩余过期时间
+        
+        Args:
+            user_id: 用户ID（end_user_id）
+            
+        Returns:
+            剩余秒数，-1表示永不过期，-2表示key不存在
+        """
+        try:
+            key = cls._get_key("profile", user_id)
+            ttl = await aio_redis.ttl(key)
+            logger.debug(f"用户画像缓存TTL: {key} = {ttl}秒")
+            return ttl
+        except Exception as e:
+            logger.error(f"获取用户画像缓存TTL失败: {e}")
+            return -2

api/app/celery_app.py

@@ -1,10 +1,16 @@
 import os
+import platform
 from datetime import timedelta
 from urllib.parse import quote
 
-from app.core.config import settings
 from celery import Celery
 
+from app.core.config import settings
+
+# macOS fork() safety - must be set before any Celery initialization
+if platform.system() == 'Darwin':
+    os.environ.setdefault('OBJC_DISABLE_INITIALIZE_FORK_SAFETY', 'YES')
+
 # 创建 Celery 应用实例
 # broker: 任务队列（使用 Redis DB 0）
 # backend: 结果存储（使用 Redis DB 10）
@@ -14,27 +20,12 @@ celery_app = Celery(
     backend=f"redis://:{quote(settings.REDIS_PASSWORD)}@{settings.REDIS_HOST}:{settings.REDIS_PORT}/{settings.CELERY_BACKEND}",
 )
 
-# 配置使用本地队列，避免与远程 worker 冲突
-celery_app.conf.task_default_queue = 'localhost_test_wyl'
-celery_app.conf.task_default_exchange = 'localhost_test_wyl'
-celery_app.conf.task_default_routing_key = 'localhost_test_wyl'
+# Default queue for unrouted tasks
+celery_app.conf.task_default_queue = 'memory_tasks'
 
 # macOS 兼容性配置
-import platform
-
-if platform.system() == 'Darwin':  # macOS
-    # 设置环境变量解决 fork 问题
+if platform.system() == 'Darwin':
     os.environ.setdefault('OBJC_DISABLE_INITIALIZE_FORK_SAFETY', 'YES')
-    
-    # 使用 solo 池避免多进程问题
-    celery_app.conf.worker_pool = 'solo'
-    
-    # 设置唯一的节点名称
-    import socket
-    import time
-    hostname = socket.gethostname()
-    timestamp = int(time.time())
-    celery_app.conf.worker_name = f"celery@{hostname}-{timestamp}"
 
 # Celery 配置
 celery_app.conf.update(
@@ -52,76 +43,86 @@ celery_app.conf.update(
     task_ignore_result=False,
     
     # 超时设置
-    task_time_limit=30 * 60,  # 30 分钟硬超时
-    task_soft_time_limit=25 * 60,  # 25 分钟软超时
+    task_time_limit=1800,  # 30分钟硬超时
+    task_soft_time_limit=1500,  # 25分钟软超时
     
-    # Worker 设置 - 针对 macOS 优化
-    worker_prefetch_multiplier=1,  # 减少预取任务数，避免内存堆积
-    worker_max_tasks_per_child=10,  # 大幅减少每个 worker 执行的任务数，频繁重启防止内存泄漏
-    worker_max_memory_per_child=200000,  # 200MB 内存限制，超过后重启 worker
+    # Worker 设置 (per-worker settings are in docker-compose command line)
+    worker_prefetch_multiplier=1,  # Don't hoard tasks, fairer distribution
     
     # 结果过期时间
-    result_expires=3600,  # 结果保存 1 小时
+    result_expires=3600,  # 结果保存1小时
     
     # 任务确认设置
-    task_acks_late=True,  # 任务完成后才确认，避免任务丢失
-    worker_disable_rate_limits=True,  # 禁用速率限制
+    task_acks_late=True,
+    task_reject_on_worker_lost=True,
+    worker_disable_rate_limits=True,
     
-    # 任务路由（可选，用于不同队列）
-    # task_routes={
-    #     'app.core.rag.tasks.parse_document': {'queue': 'document_processing'},
-    #     'app.core.memory.agent.read_message': {'queue': 'memory_processing'},
-    #     'app.core.memory.agent.write_message': {'queue': 'memory_processing'},
-    #     'tasks.process_item': {'queue': 'default'},
-    # },
+    # FLower setting 
+    worker_send_task_events=True,
+    task_send_sent_event=True,
+    
+    # task routing
+    task_routes={
+        # Memory tasks → memory_tasks queue (threads worker)
+        'app.core.memory.agent.read_message_priority': {'queue': 'memory_tasks'},
+        'app.core.memory.agent.read_message': {'queue': 'memory_tasks'},
+        'app.core.memory.agent.write_message': {'queue': 'memory_tasks'},
+        
+        # Long-term storage tasks → memory_tasks queue (batched write strategies)
+        'app.core.memory.agent.long_term_storage.window': {'queue': 'memory_tasks'},
+        'app.core.memory.agent.long_term_storage.time': {'queue': 'memory_tasks'},
+        'app.core.memory.agent.long_term_storage.aggregate': {'queue': 'memory_tasks'},
+        
+        # Document tasks → document_tasks queue (prefork worker)
+        'app.core.rag.tasks.parse_document': {'queue': 'document_tasks'},
+        'app.core.rag.tasks.build_graphrag_for_kb': {'queue': 'document_tasks'},
+        
+        # Beat/periodic tasks → periodic_tasks queue (dedicated periodic worker)
+        'app.tasks.workspace_reflection_task': {'queue': 'periodic_tasks'},
+        'app.tasks.regenerate_memory_cache': {'queue': 'periodic_tasks'},
+        'app.tasks.run_forgetting_cycle_task': {'queue': 'periodic_tasks'},
+        'app.controllers.memory_storage_controller.search_all': {'queue': 'periodic_tasks'},
+    },
 )
 
 # 自动发现任务模块
 celery_app.autodiscover_tasks(['app'])
 
 # Celery Beat schedule for periodic tasks
-reflection_schedule = timedelta(seconds=settings.REFLECTION_INTERVAL_SECONDS)
-health_schedule = timedelta(seconds=settings.HEALTH_CHECK_SECONDS)
-memory_increment_schedule = timedelta(hours=settings.MEMORY_INCREMENT_INTERVAL_HOURS)
-memory_cache_regeneration_schedule = timedelta(hours=settings.MEMORY_CACHE_REGENERATION_HOURS)
-workspace_reflection_schedule = timedelta(seconds=30)  # 每30秒运行一次settings.REFLECTION_INTERVAL_TIME
-forgetting_cycle_schedule = timedelta(hours=24)  # 每24小时运行一次遗忘周期
+# memory_increment_schedule = timedelta(hours=settings.MEMORY_INCREMENT_INTERVAL_HOURS)
+# memory_cache_regeneration_schedule = timedelta(hours=settings.MEMORY_CACHE_REGENERATION_HOURS)
+# workspace_reflection_schedule = timedelta(seconds=30)  # 每30秒运行一次settings.REFLECTION_INTERVAL_TIME
+# forgetting_cycle_schedule = timedelta(hours=24)  # 每24小时运行一次遗忘周期
 
 # 构建定时任务配置
-beat_schedule_config = {
-
-    # "check-read-service": {
-    #     "task": "app.core.memory.agent.health.check_read_service",
-    #     "schedule": health_schedule,
-    #     "args": (),
-    # },
-    "run-workspace-reflection": {
-        "task": "app.tasks.workspace_reflection_task",
-        "schedule": workspace_reflection_schedule,
-        "args": (),
-    },
-    "regenerate-memory-cache": {
-        "task": "app.tasks.regenerate_memory_cache",
-        "schedule": memory_cache_regeneration_schedule,
-        "args": (),
-    },
-    "run-forgetting-cycle": {
-        "task": "app.tasks.run_forgetting_cycle_task",
-        "schedule": forgetting_cycle_schedule,
-        "kwargs": {
-            "config_id": None,  # 使用默认配置，可以通过环境变量配置
-        },
-    },
-}
+# beat_schedule_config = {
+#     "run-workspace-reflection": {
+#         "task": "app.tasks.workspace_reflection_task",
+#         "schedule": workspace_reflection_schedule,
+#         "args": (),
+#     },
+#     "regenerate-memory-cache": {
+#         "task": "app.tasks.regenerate_memory_cache",
+#         "schedule": memory_cache_regeneration_schedule,
+#         "args": (),
+#     },
+#     "run-forgetting-cycle": {
+#         "task": "app.tasks.run_forgetting_cycle_task",
+#         "schedule": forgetting_cycle_schedule,
+#         "kwargs": {
+#             "config_id": None,  # 使用默认配置，可以通过环境变量配置
+#         },
+#     },
+# }
 
 # 如果配置了默认工作空间ID，则添加记忆总量统计任务
-if settings.DEFAULT_WORKSPACE_ID:
-    beat_schedule_config["write-total-memory"] = {
-        "task": "app.controllers.memory_storage_controller.search_all",
-        "schedule": memory_increment_schedule,
-        "kwargs": {
-            "workspace_id": settings.DEFAULT_WORKSPACE_ID,
-        },
-    }
+# if settings.DEFAULT_WORKSPACE_ID:
+#     beat_schedule_config["write-total-memory"] = {
+#         "task": "app.controllers.memory_storage_controller.search_all",
+#         "schedule": memory_increment_schedule,
+#         "kwargs": {
+#             "workspace_id": settings.DEFAULT_WORKSPACE_ID,
+#         },
+#     }
 
-celery_app.conf.beat_schedule = beat_schedule_config
+# celery_app.conf.beat_schedule = beat_schedule_config

api/app/celery_worker.py

@@ -3,6 +3,12 @@ Celery Worker 入口点
 用于启动 Celery Worker: celery -A app.celery_worker worker --loglevel=info
 """
 from app.celery_app import celery_app
+from app.core.logging_config import LoggingConfig, get_logger
+
+# Initialize logging system for Celery worker
+LoggingConfig.setup_logging()
+logger = get_logger(__name__)
+logger.info("Celery worker logging initialized")
 
 # 导入任务模块以注册任务
 import app.tasks

api/app/controllers/__init__.py

@@ -14,6 +14,7 @@ from . import (
     emotion_config_controller,
     emotion_controller,
     file_controller,
+    file_storage_controller,
     home_page_controller,
     implicit_memory_controller,
     knowledge_controller,
@@ -44,6 +45,7 @@ from . import (
     home_page_controller,
     memory_perceptual_controller,
     memory_working_controller,
+    ontology_controller,
 )
 
 # 创建管理端 API 路由器
@@ -88,5 +90,7 @@ manager_router.include_router(home_page_controller.router)
 manager_router.include_router(implicit_memory_controller.router)
 manager_router.include_router(memory_perceptual_controller.router)
 manager_router.include_router(memory_working_controller.router)
+manager_router.include_router(file_storage_controller.router)
+manager_router.include_router(ontology_controller.router)
 
 __all__ = ["manager_router"]

api/app/controllers/app_controller.py

@@ -7,7 +7,7 @@ from sqlalchemy.orm import Session
 
 from app.core.error_codes import BizCode
 from app.core.logging_config import get_business_logger
-from app.core.response_utils import success
+from app.core.response_utils import success, fail
 from app.db import get_db
 from app.dependencies import get_current_user, cur_workspace_access_guard
 from app.models import User
@@ -661,6 +661,11 @@ async def draft_run(
             data=result,
             msg="工作流任务执行成功"
         )
+    else:
+        return fail(
+            msg="未知应用类型",
+            code=422
+        )
 
 
 @router.post("/{app_id}/draft/run/compare", summary="多模型对比试运行")
@@ -867,3 +872,44 @@ async def update_workflow_config(
     workspace_id = current_user.current_workspace_id
     cfg = app_service.update_workflow_config(db, app_id=app_id, data=payload, workspace_id=workspace_id)
     return success(data=WorkflowConfigSchema.model_validate(cfg))
+
+
+@router.get("/{app_id}/statistics", summary="应用统计数据")
+@cur_workspace_access_guard()
+def get_app_statistics(
+        app_id: uuid.UUID,
+        start_date: int,
+        end_date: int,
+        db: Session = Depends(get_db),
+        current_user=Depends(get_current_user),
+):
+    """获取应用统计数据
+    
+    Args:
+        app_id: 应用ID
+        start_date: 开始时间戳（毫秒）
+        end_date: 结束时间戳（毫秒）
+    
+    Returns:
+        - daily_conversations: 每日会话数统计
+        - total_conversations: 总会话数
+        - daily_new_users: 每日新增用户数
+        - total_new_users: 总新增用户数
+        - daily_api_calls: 每日API调用次数
+        - total_api_calls: 总API调用次数
+        - daily_tokens: 每日token消耗
+        - total_tokens: 总token消耗
+    """
+    workspace_id = current_user.current_workspace_id
+    
+    from app.services.app_statistics_service import AppStatisticsService
+    stats_service = AppStatisticsService(db)
+    
+    result = stats_service.get_app_statistics(
+        app_id=app_id,
+        workspace_id=workspace_id,
+        start_date=start_date,
+        end_date=end_date
+    )
+    
+    return success(data=result)

api/app/controllers/emotion_config_controller.py

@@ -7,11 +7,13 @@ Routes:
     GET /memory/config/emotion - 获取情绪引擎配置
     POST /memory/config/emotion - 更新情绪引擎配置
 """
+import uuid
 
 from fastapi import APIRouter, Depends, Query, HTTPException, status
 from pydantic import BaseModel, Field
-from typing import Optional
+from typing import Optional, Union
 from sqlalchemy.orm import Session
+from uuid import UUID
 
 from app.core.response_utils import success
 from app.dependencies import get_current_user
@@ -20,6 +22,7 @@ from app.schemas.response_schema import ApiResponse
 from app.services.emotion_config_service import EmotionConfigService
 from app.core.logging_config import get_api_logger
 from app.db import get_db
+from app.utils.config_utils import resolve_config_id
 
 # 获取API专用日志器
 api_logger = get_api_logger()
@@ -32,11 +35,11 @@ router = APIRouter(
 
 class EmotionConfigQuery(BaseModel):
     """情绪配置查询请求模型"""
-    config_id: int = Field(..., description="配置ID")
+    config_id: UUID = Field(..., description="配置ID")
 
 class EmotionConfigUpdate(BaseModel):
     """情绪配置更新请求模型"""
-    config_id: int = Field(..., description="配置ID")
+    config_id: Union[uuid.UUID, int, str]= Field(..., description="配置ID")
     emotion_enabled: bool = Field(..., description="是否启用情绪提取")
     emotion_model_id: Optional[str] = Field(None, description="情绪分析专用模型ID")
     emotion_extract_keywords: bool = Field(..., description="是否提取情绪关键词")
@@ -45,7 +48,7 @@ class EmotionConfigUpdate(BaseModel):
 
 @router.get("/read_config", response_model=ApiResponse)
 def get_emotion_config(
-    config_id: int = Query(..., description="配置ID"),
+    config_id: UUID|int = Query(..., description="配置ID"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user),
 ):
@@ -78,7 +81,7 @@ def get_emotion_config(
             f"用户 {current_user.username} 请求获取情绪配置",
             extra={"config_id": config_id}
         )
-        
+        config_id=resolve_config_id(config_id, db)
         # 初始化服务
         config_service = EmotionConfigService(db)
         
@@ -157,6 +160,7 @@ def update_emotion_config(
             }
         }
     """
+    config.config_id=resolve_config_id(config.config_id, db)
     try:
         api_logger.info(
             f"用户 {current_user.username} 请求更新情绪配置",

api/app/controllers/emotion_controller.py

@@ -24,7 +24,7 @@ from app.schemas.emotion_schema import (
 )
 from app.schemas.response_schema import ApiResponse
 from app.services.emotion_analytics_service import EmotionAnalyticsService
-from fastapi import APIRouter, Depends, HTTPException, status
+from fastapi import APIRouter, Depends, HTTPException, status,Header
 from sqlalchemy.orm import Session
 
 # 获取API专用日志器
@@ -45,6 +45,7 @@ emotion_service = EmotionAnalyticsService()
 @router.post("/tags", response_model=ApiResponse)
 async def get_emotion_tags(
     request: EmotionTagsRequest,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     current_user: User = Depends(get_current_user),
 ):
 
@@ -52,38 +53,38 @@ async def get_emotion_tags(
         api_logger.info(
             f"用户 {current_user.username} 请求获取情绪标签统计",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "emotion_type": request.emotion_type,
                 "start_date": request.start_date,
                 "end_date": request.end_date,
                 "limit": request.limit
             }
         )
-        
+
         # 调用服务层
         data = await emotion_service.get_emotion_tags(
-            end_user_id=request.group_id,
+            end_user_id=request.end_user_id,
             emotion_type=request.emotion_type,
             start_date=request.start_date,
             end_date=request.end_date,
             limit=request.limit
         )
-        
+
         api_logger.info(
             "情绪标签统计获取成功",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "total_count": data.get("total_count", 0),
                 "tags_count": len(data.get("tags", []))
             }
         )
-        
+
         return success(data=data, msg="情绪标签获取成功")
-        
+
     except Exception as e:
         api_logger.error(
             f"获取情绪标签统计失败: {str(e)}",
-            extra={"group_id": request.group_id},
+            extra={"end_user_id": request.end_user_id},
             exc_info=True
         )
         raise HTTPException(
@@ -96,6 +97,7 @@ async def get_emotion_tags(
 @router.post("/wordcloud", response_model=ApiResponse)
 async def get_emotion_wordcloud(
     request: EmotionWordcloudRequest,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     current_user: User = Depends(get_current_user),
 ):
 
@@ -103,33 +105,33 @@ async def get_emotion_wordcloud(
         api_logger.info(
             f"用户 {current_user.username} 请求获取情绪词云数据",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "emotion_type": request.emotion_type,
                 "limit": request.limit
             }
         )
-        
+
         # 调用服务层
         data = await emotion_service.get_emotion_wordcloud(
-            end_user_id=request.group_id,
+            end_user_id=request.end_user_id,
             emotion_type=request.emotion_type,
             limit=request.limit
         )
-        
+
         api_logger.info(
             "情绪词云数据获取成功",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "total_keywords": data.get("total_keywords", 0)
             }
         )
-        
+
         return success(data=data, msg="情绪词云获取成功")
-        
+
     except Exception as e:
         api_logger.error(
             f"获取情绪词云数据失败: {str(e)}",
-            extra={"group_id": request.group_id},
+            extra={"end_user_id": request.end_user_id},
             exc_info=True
         )
         raise HTTPException(
@@ -142,6 +144,7 @@ async def get_emotion_wordcloud(
 @router.post("/health", response_model=ApiResponse)
 async def get_emotion_health(
     request: EmotionHealthRequest,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     current_user: User = Depends(get_current_user),
 ):
 
@@ -152,38 +155,38 @@ async def get_emotion_health(
                 status_code=status.HTTP_400_BAD_REQUEST,
                 detail="时间范围参数无效，必须是 7d、30d 或 90d"
             )
-        
+
         api_logger.info(
             f"用户 {current_user.username} 请求获取情绪健康指数",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "time_range": request.time_range
             }
         )
-        
+
         # 调用服务层
         data = await emotion_service.calculate_emotion_health_index(
-            end_user_id=request.group_id,
+            end_user_id=request.end_user_id,
             time_range=request.time_range
         )
-        
+
         api_logger.info(
             "情绪健康指数获取成功",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "health_score": data.get("health_score", 0),
                 "level": data.get("level", "未知")
             }
         )
-        
+
         return success(data=data, msg="情绪健康指数获取成功")
-        
+
     except HTTPException:
         raise
     except Exception as e:
         api_logger.error(
             f"获取情绪健康指数失败: {str(e)}",
-            extra={"group_id": request.group_id},
+            extra={"end_user_id": request.end_user_id},
             exc_info=True
         )
         raise HTTPException(
@@ -196,16 +199,17 @@ async def get_emotion_health(
 @router.post("/suggestions", response_model=ApiResponse)
 async def get_emotion_suggestions(
     request: EmotionSuggestionsRequest,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user),
 ):
     """获取个性化情绪建议（从缓存读取）
-    
+
     Args:
-        request: 包含 group_id 和可选的 config_id
+        request: 包含 end_user_id 和可选的 config_id
         db: 数据库会话
         current_user: 当前用户
-        
+
     Returns:
         缓存的个性化情绪建议响应
     """
@@ -213,43 +217,43 @@ async def get_emotion_suggestions(
         api_logger.info(
             f"用户 {current_user.username} 请求获取个性化情绪建议（缓存）",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "config_id": request.config_id
             }
         )
-        
+
         # 从缓存获取建议
         data = await emotion_service.get_cached_suggestions(
-            end_user_id=request.group_id,
+            end_user_id=request.end_user_id,
             db=db
         )
-        
+
         if data is None:
             # 缓存不存在或已过期
             api_logger.info(
-                f"用户 {request.group_id} 的建议缓存不存在或已过期",
-                extra={"group_id": request.group_id}
+                f"用户 {request.end_user_id} 的建议缓存不存在或已过期",
+                extra={"end_user_id": request.end_user_id}
             )
             return fail(
-                BizCode.RESOURCE_NOT_FOUND,
-                "建议缓存不存在或已过期，请调用 /generate_suggestions 接口生成新建议",
-                None
+                BizCode.NOT_FOUND,
+                "建议缓存不存在或已过期，请右上角刷新生成新建议",
+                ""
             )
-        
+
         api_logger.info(
             "个性化建议获取成功（缓存）",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "suggestions_count": len(data.get("suggestions", []))
             }
         )
-        
+
         return success(data=data, msg="个性化建议获取成功（缓存）")
-        
+
     except Exception as e:
         api_logger.error(
             f"获取个性化建议失败: {str(e)}",
-            extra={"group_id": request.group_id},
+            extra={"end_user_id": request.end_user_id},
             exc_info=True
         )
         raise HTTPException(
@@ -261,80 +265,56 @@ async def get_emotion_suggestions(
 @router.post("/generate_suggestions", response_model=ApiResponse)
 async def generate_emotion_suggestions(
     request: EmotionGenerateSuggestionsRequest,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user),
 ):
     """生成个性化情绪建议（调用LLM并缓存）
-    
+
     Args:
-        request: 包含 group_id、可选的 config_id 和 force_refresh
+        request: 包含 end_user_id
         db: 数据库会话
         current_user: 当前用户
-        
+
     Returns:
         新生成的个性化情绪建议响应
     """
     try:
-        # 验证 config_id（如果提供）
-        # 获取终端用户关联的配置
-        config_id = request.config_id
-        if config_id is None:
-            # 如果没有提供 config_id，尝试获取用户关联的配置
-            try:
-                from app.services.memory_agent_service import (
-                    get_end_user_connected_config,
-                )
-                connected_config = get_end_user_connected_config(request.group_id, db)
-                config_id = connected_config.get("memory_config_id")
-            except ValueError as e:
-                return fail(BizCode.INVALID_PARAMETER, "无法获取用户关联的配置", str(e))
-        else:
-            # 如果提供了 config_id，验证其有效性
-            from app.services.memory_config_service import MemoryConfigService
-            try:
-                config_service = MemoryConfigService(db)
-                config = config_service.get_config_by_id(config_id)
-                if not config:
-                    return fail(BizCode.INVALID_PARAMETER, "配置ID无效", f"配置 {config_id} 不存在")
-            except Exception as e:
-                return fail(BizCode.INVALID_PARAMETER, "配置ID验证失败", str(e))
-        
         api_logger.info(
             f"用户 {current_user.username} 请求生成个性化情绪建议",
             extra={
-                "group_id": request.group_id,
-                "config_id": config_id
+                "end_user_id": request.end_user_id
             }
         )
-        
+
         # 调用服务层生成建议
         data = await emotion_service.generate_emotion_suggestions(
-            end_user_id=request.group_id,
+            end_user_id=request.end_user_id,
             db=db
         )
-        
+
         # 保存到缓存
         await emotion_service.save_suggestions_cache(
-            end_user_id=request.group_id,
+            end_user_id=request.end_user_id,
             suggestions_data=data,
             db=db,
             expires_hours=24
         )
-        
+
         api_logger.info(
             "个性化建议生成成功",
             extra={
-                "group_id": request.group_id,
+                "end_user_id": request.end_user_id,
                 "suggestions_count": len(data.get("suggestions", []))
             }
         )
-        
+
         return success(data=data, msg="个性化建议生成成功")
-        
+
     except Exception as e:
         api_logger.error(
             f"生成个性化建议失败: {str(e)}",
-            extra={"group_id": request.group_id},
+            extra={"end_user_id": request.end_user_id},
             exc_info=True
         )
         raise HTTPException(

api/app/controllers/file_storage_controller.py

@@ -0,0 +1,499 @@
+"""
+File storage controller module.
+
+This module provides API endpoints for file storage operations using the
+configurable storage backend. It is a new controller that does not modify
+the existing file_controller.py.
+
+Routes:
+    POST /storage/files - Upload a file
+    GET /storage/files/{file_id} - Download a file
+    DELETE /storage/files/{file_id} - Delete a file
+"""
+
+import os
+import uuid
+from typing import Any
+
+from fastapi import APIRouter, Depends, File, HTTPException, UploadFile, status
+from fastapi.responses import FileResponse, RedirectResponse
+from sqlalchemy.orm import Session
+
+from app.core.config import settings
+from app.core.logging_config import get_api_logger
+from app.core.response_utils import success
+from app.core.storage import LocalStorage
+from app.core.storage.url_signer import generate_signed_url, verify_signed_url
+from app.core.storage_exceptions import (
+    StorageDeleteError,
+    StorageUploadError,
+)
+from app.db import get_db
+from app.dependencies import get_current_user
+from app.models.file_metadata_model import FileMetadata
+from app.models.user_model import User
+from app.schemas.response_schema import ApiResponse
+from app.services.file_storage_service import (
+    FileStorageService,
+    generate_file_key,
+    get_file_storage_service,
+)
+
+api_logger = get_api_logger()
+
+router = APIRouter(
+    prefix="/storage",
+    tags=["storage"]
+)
+
+
+@router.post("/files", response_model=ApiResponse)
+async def upload_file(
+    file: UploadFile = File(...),
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+):
+    """
+    Upload a file to the configured storage backend.
+    """
+    tenant_id = current_user.tenant_id
+    workspace_id = current_user.current_workspace_id
+
+    api_logger.info(
+        f"Storage upload request: tenant_id={tenant_id}, workspace_id={workspace_id}, "
+        f"filename={file.filename}, username={current_user.username}"
+    )
+
+    # Read file contents
+    contents = await file.read()
+    file_size = len(contents)
+
+    # Validate file size
+    if file_size == 0:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="The file is empty."
+        )
+
+    if file_size > settings.MAX_FILE_SIZE:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=f"The file size exceeds the {settings.MAX_FILE_SIZE} byte limit"
+        )
+
+    # Extract file extension
+    _, file_extension = os.path.splitext(file.filename)
+    file_ext = file_extension.lower()
+
+    # Generate file_id and file_key
+    file_id = uuid.uuid4()
+    file_key = generate_file_key(
+        tenant_id=tenant_id,
+        workspace_id=workspace_id,
+        file_id=file_id,
+        file_ext=file_ext,
+    )
+
+    # Create file metadata record with pending status
+    file_metadata = FileMetadata(
+        id=file_id,
+        tenant_id=tenant_id,
+        workspace_id=workspace_id,
+        file_key=file_key,
+        file_name=file.filename,
+        file_ext=file_ext,
+        file_size=file_size,
+        content_type=file.content_type,
+        status="pending",
+    )
+    db.add(file_metadata)
+    db.commit()
+    db.refresh(file_metadata)
+
+    # Upload file to storage backend
+    try:
+        await storage_service.upload_file(
+            tenant_id=tenant_id,
+            workspace_id=workspace_id,
+            file_id=file_id,
+            file_ext=file_ext,
+            content=contents,
+            content_type=file.content_type,
+        )
+        # Update status to completed
+        file_metadata.status = "completed"
+        db.commit()
+        api_logger.info(f"File uploaded to storage: file_key={file_key}")
+    except StorageUploadError as e:
+        # Update status to failed
+        file_metadata.status = "failed"
+        db.commit()
+        api_logger.error(f"Storage upload failed: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=f"File storage failed: {str(e)}"
+        )
+
+    api_logger.info(f"File upload successful: {file.filename} (file_id: {file_id})")
+
+    return success(
+        data={"file_id": str(file_id), "file_key": file_key},
+        msg="File upload successful"
+    )
+
+
+@router.get("/files/{file_id}", response_model=Any)
+async def download_file(
+    file_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+) -> Any:
+    """
+    Download a file from the configured storage backend.
+    """
+    api_logger.info(f"Storage download request: file_id={file_id}")
+
+    # Query file metadata from database
+    file_metadata = db.query(FileMetadata).filter(FileMetadata.id == file_id).first()
+    if not file_metadata:
+        api_logger.warning(f"File not found in database: file_id={file_id}")
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="The file does not exist"
+        )
+
+    if file_metadata.status != "completed":
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=f"File upload not completed, status: {file_metadata.status}"
+        )
+
+    file_key = file_metadata.file_key
+    storage = storage_service.storage
+
+    if isinstance(storage, LocalStorage):
+        full_path = storage._get_full_path(file_key)
+
+        if not full_path.exists():
+            api_logger.warning(f"File not found on disk: file_key={file_key}")
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="File not found (possibly deleted)"
+            )
+
+        api_logger.info(f"Serving local file: file_key={file_key}")
+        return FileResponse(
+            path=str(full_path),
+            filename=file_metadata.file_name,
+            media_type=file_metadata.content_type or "application/octet-stream"
+        )
+    else:
+        try:
+            presigned_url = await storage_service.get_file_url(file_key, expires=3600)
+            api_logger.info(f"Redirecting to presigned URL: file_key={file_key}")
+            return RedirectResponse(url=presigned_url, status_code=status.HTTP_302_FOUND)
+        except FileNotFoundError:
+            api_logger.warning(f"File not found in remote storage: file_key={file_key}")
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="File not found in storage"
+            )
+        except Exception as e:
+            api_logger.error(f"Failed to get presigned URL: {e}")
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail=f"Failed to retrieve file: {str(e)}"
+            )
+
+
+@router.delete("/files/{file_id}", response_model=ApiResponse)
+async def delete_file(
+    file_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+):
+    """
+    Delete a file from the configured storage backend.
+    """
+    api_logger.info(
+        f"Storage delete request: file_id={file_id}, username={current_user.username}"
+    )
+
+    # Query file metadata from database
+    file_metadata = db.query(FileMetadata).filter(FileMetadata.id == file_id).first()
+    if not file_metadata:
+        api_logger.warning(f"File not found in database: file_id={file_id}")
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="The file does not exist"
+        )
+
+    file_key = file_metadata.file_key
+
+    # Delete file from storage
+    try:
+        deleted = await storage_service.delete_file(file_key)
+        if deleted:
+            api_logger.info(f"File deleted from storage: file_key={file_key}")
+        else:
+            api_logger.info(f"File did not exist in storage: file_key={file_key}")
+    except StorageDeleteError as e:
+        api_logger.error(f"Storage delete failed: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=f"Failed to delete file from storage: {str(e)}"
+        )
+
+    # Delete database record
+    try:
+        db.delete(file_metadata)
+        db.commit()
+        api_logger.info(f"File record deleted from database: file_id={file_id}")
+    except Exception as e:
+        api_logger.error(f"Database delete failed: {e}")
+        db.rollback()
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=f"Failed to delete file record: {str(e)}"
+        )
+
+    return success(msg="File deleted successfully")
+
+
+@router.get("/files/{file_id}/url", response_model=ApiResponse)
+async def get_file_url(
+    file_id: uuid.UUID,
+    expires: int = None,
+    permanent: bool = False,
+    db: Session = Depends(get_db),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+):
+    """
+    Get an access URL for a file (no authentication required).
+
+    Args:
+        file_id: The UUID of the file.
+        expires: URL validity period in seconds (default from FILE_URL_EXPIRES env).
+        permanent: If True, return a permanent URL without expiration.
+        db: Database session.
+        storage_service: The file storage service.
+
+    Returns:
+        ApiResponse with the access URL.
+    """
+    if expires is None:
+        expires = settings.FILE_URL_EXPIRES
+
+    api_logger.info(f"Get file URL request: file_id={file_id}, expires={expires}, permanent={permanent}")
+
+    # Query file metadata from database
+    file_metadata = db.query(FileMetadata).filter(FileMetadata.id == file_id).first()
+    if not file_metadata:
+        api_logger.warning(f"File not found in database: file_id={file_id}")
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="The file does not exist"
+        )
+
+    if file_metadata.status != "completed":
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=f"File upload not completed, status: {file_metadata.status}"
+        )
+
+    file_key = file_metadata.file_key
+    storage = storage_service.storage
+
+    try:
+        if permanent:
+            # Generate permanent URL (no expiration check)
+            server_url = settings.FILE_LOCAL_SERVER_URL
+            url = f"{server_url}/storage/permanent/{file_id}"
+            return success(
+                data={
+                    "url": url,
+                    "expires_in": None,
+                    "permanent": True,
+                    "file_name": file_metadata.file_name,
+                },
+                msg="Permanent file URL generated successfully"
+            )
+
+        if isinstance(storage, LocalStorage):
+            # For local storage, generate signed URL with expiration
+            url = generate_signed_url(str(file_id), expires)
+        else:
+            # For remote storage (OSS/S3), get presigned URL
+            url = await storage_service.get_file_url(file_key, expires=expires)
+
+        api_logger.info(f"Generated file URL: file_id={file_id}")
+        return success(
+            data={
+                "url": url,
+                "expires_in": expires,
+                "permanent": False,
+                "file_name": file_metadata.file_name,
+            },
+            msg="File URL generated successfully"
+        )
+    except Exception as e:
+        api_logger.error(f"Failed to generate file URL: {e}")
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail=f"Failed to generate file URL: {str(e)}"
+        )
+
+
+@router.get("/public/{file_id}", response_model=Any)
+async def public_download_file(
+    file_id: uuid.UUID,
+    expires: int = 0,
+    signature: str = "",
+    db: Session = Depends(get_db),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+) -> Any:
+    """
+    Public file download endpoint with signature verification.
+
+    This endpoint allows downloading files without authentication,
+    but requires a valid signature and non-expired timestamp.
+
+    Args:
+        file_id: The UUID of the file.
+        expires: Expiration timestamp.
+        signature: HMAC signature for verification.
+        db: Database session.
+        storage_service: The file storage service.
+
+    Returns:
+        FileResponse for the requested file.
+    """
+    api_logger.info(f"Public download request: file_id={file_id}")
+
+    # Verify signature
+    is_valid, error_msg = verify_signed_url(str(file_id), expires, signature)
+    if not is_valid:
+        api_logger.warning(f"Invalid signed URL: file_id={file_id}, error={error_msg}")
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN,
+            detail=error_msg
+        )
+
+    # Query file metadata from database
+    file_metadata = db.query(FileMetadata).filter(FileMetadata.id == file_id).first()
+    if not file_metadata:
+        api_logger.warning(f"File not found in database: file_id={file_id}")
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="The file does not exist"
+        )
+
+    if file_metadata.status != "completed":
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=f"File upload not completed, status: {file_metadata.status}"
+        )
+
+    file_key = file_metadata.file_key
+    storage = storage_service.storage
+
+    if isinstance(storage, LocalStorage):
+        full_path = storage._get_full_path(file_key)
+
+        if not full_path.exists():
+            api_logger.warning(f"File not found on disk: file_key={file_key}")
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="File not found"
+            )
+
+        api_logger.info(f"Serving public file: file_key={file_key}")
+        return FileResponse(
+            path=str(full_path),
+            filename=file_metadata.file_name,
+            media_type=file_metadata.content_type or "application/octet-stream"
+        )
+    else:
+        # For remote storage, redirect to presigned URL
+        try:
+            presigned_url = await storage_service.get_file_url(file_key, expires=3600)
+            return RedirectResponse(url=presigned_url, status_code=status.HTTP_302_FOUND)
+        except Exception as e:
+            api_logger.error(f"Failed to get presigned URL: {e}")
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail=f"Failed to retrieve file: {str(e)}"
+            )
+
+
+@router.get("/permanent/{file_id}", response_model=Any)
+async def permanent_download_file(
+    file_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    storage_service: FileStorageService = Depends(get_file_storage_service),
+) -> Any:
+    """
+    Permanent file download endpoint (no expiration, no signature required).
+
+    This endpoint allows downloading files without authentication or expiration.
+    Use with caution as URLs are permanently accessible.
+
+    Args:
+        file_id: The UUID of the file.
+        db: Database session.
+        storage_service: The file storage service.
+
+    Returns:
+        FileResponse for the requested file.
+    """
+    api_logger.info(f"Permanent download request: file_id={file_id}")
+
+    # Query file metadata from database
+    file_metadata = db.query(FileMetadata).filter(FileMetadata.id == file_id).first()
+    if not file_metadata:
+        api_logger.warning(f"File not found in database: file_id={file_id}")
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail="The file does not exist"
+        )
+
+    if file_metadata.status != "completed":
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=f"File upload not completed, status: {file_metadata.status}"
+        )
+
+    file_key = file_metadata.file_key
+    storage = storage_service.storage
+
+    if isinstance(storage, LocalStorage):
+        full_path = storage._get_full_path(file_key)
+
+        if not full_path.exists():
+            api_logger.warning(f"File not found on disk: file_key={file_key}")
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="File not found"
+            )
+
+        api_logger.info(f"Serving permanent file: file_key={file_key}")
+        return FileResponse(
+            path=str(full_path),
+            filename=file_metadata.file_name,
+            media_type=file_metadata.content_type or "application/octet-stream"
+        )
+    else:
+        # For remote storage, redirect to presigned URL with long expiration
+        try:
+            # Use a very long expiration (7 days max for most cloud providers)
+            presigned_url = await storage_service.get_file_url(file_key, expires=604800)
+            return RedirectResponse(url=presigned_url, status_code=status.HTTP_302_FOUND)
+        except Exception as e:
+            api_logger.error(f"Failed to get presigned URL: {e}")
+            raise HTTPException(
+                status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+                detail=f"Failed to retrieve file: {str(e)}"
+            )

api/app/controllers/implicit_memory_controller.py

@@ -122,10 +122,10 @@ def validate_confidence_threshold(threshold: float) -> None:
         raise ValueError("confidence_threshold must be between 0.0 and 1.0")
 
 
-@router.get("/preferences/{user_id}", response_model=ApiResponse)
+@router.get("/preferences/{end_user_id}", response_model=ApiResponse)
 @cur_workspace_access_guard()
 async def get_preference_tags(
-    user_id: str,
+    end_user_id: str,
     confidence_threshold: float = Query(0.5, ge=0.0, le=1.0, description="Minimum confidence threshold"),
     tag_category: Optional[str] = Query(None, description="Filter by tag category"),
     start_date: Optional[datetime] = Query(None, description="Filter start date"),
@@ -137,7 +137,7 @@ async def get_preference_tags(
     Get user preference tags from cache.
     
     Args:
-        user_id: Target user ID
+        end_user_id: Target end user ID
         confidence_threshold: Minimum confidence score (0.0-1.0)
         tag_category: Optional category filter
         start_date: Optional start date filter
@@ -146,24 +146,24 @@ async def get_preference_tags(
     Returns:
         List of preference tags from cache
     """
-    api_logger.info(f"Preference tags requested for user: {user_id} (from cache)")
+    api_logger.info(f"Preference tags requested for user: {end_user_id} (from cache)")
     
     try:
         # Validate inputs
-        validate_user_id(user_id)
+        validate_user_id(end_user_id)
         
         # Create service with user-specific config
-        service = ImplicitMemoryService(db=db, end_user_id=user_id)
+        service = ImplicitMemoryService(db=db, end_user_id=end_user_id)
         
         # Get cached profile
-        cached_profile = await service.get_cached_profile(end_user_id=user_id, db=db)
+        cached_profile = await service.get_cached_profile(end_user_id=end_user_id, db=db)
         
         if cached_profile is None:
-            api_logger.info(f"用户 {user_id} 的画像缓存不存在或已过期")
+            api_logger.info(f"用户 {end_user_id} 的画像缓存不存在或已过期")
             return fail(
-                BizCode.RESOURCE_NOT_FOUND,
-                "画像缓存不存在或已过期，请调用 /generate_profile 接口生成新画像",
-                None
+                BizCode.NOT_FOUND,
+                "画像缓存不存在或已过期，请右上角刷新生成新画像",
+                ""
             )
         
         # Extract preferences from cache
@@ -192,17 +192,17 @@ async def get_preference_tags(
             
             filtered_preferences.append(pref)
         
-        api_logger.info(f"Retrieved {len(filtered_preferences)} preference tags for user: {user_id} (from cache)")
+        api_logger.info(f"Retrieved {len(filtered_preferences)} preference tags for user: {end_user_id} (from cache)")
         return success(data=filtered_preferences, msg="偏好标签获取成功（缓存）")
         
     except Exception as e:
-        return handle_implicit_memory_error(e, "偏好标签获取", user_id)
+        return handle_implicit_memory_error(e, "偏好标签获取", end_user_id)
 
 
-@router.get("/portrait/{user_id}", response_model=ApiResponse)
+@router.get("/portrait/{end_user_id}", response_model=ApiResponse)
 @cur_workspace_access_guard()
 async def get_dimension_portrait(
-    user_id: str,
+    end_user_id: str,
     include_history: bool = Query(False, description="Include historical trends"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user)
@@ -211,46 +211,46 @@ async def get_dimension_portrait(
     Get user's four-dimension personality portrait from cache.
     
     Args:
-        user_id: Target user ID
+        end_user_id: Target end user ID
         include_history: Whether to include historical trend data (ignored for cached data)
         
     Returns:
         Four-dimension personality portrait from cache
     """
-    api_logger.info(f"Dimension portrait requested for user: {user_id} (from cache)")
+    api_logger.info(f"Dimension portrait requested for user: {end_user_id} (from cache)")
     
     try:
         # Validate inputs
-        validate_user_id(user_id)
+        validate_user_id(end_user_id)
         
         # Create service with user-specific config
-        service = ImplicitMemoryService(db=db, end_user_id=user_id)
+        service = ImplicitMemoryService(db=db, end_user_id=end_user_id)
         
         # Get cached profile
-        cached_profile = await service.get_cached_profile(end_user_id=user_id, db=db)
+        cached_profile = await service.get_cached_profile(end_user_id=end_user_id, db=db)
         
         if cached_profile is None:
-            api_logger.info(f"用户 {user_id} 的画像缓存不存在或已过期")
+            api_logger.info(f"用户 {end_user_id} 的画像缓存不存在或已过期")
             return fail(
-                BizCode.RESOURCE_NOT_FOUND,
-                "画像缓存不存在或已过期，请调用 /generate_profile 接口生成新画像",
-                None
+                BizCode.NOT_FOUND,
+                "画像缓存不存在或已过期，请右上角刷新生成新画像",
+                ""
             )
         
         # Extract portrait from cache
         portrait = cached_profile.get("portrait", {})
         
-        api_logger.info(f"Dimension portrait retrieved for user: {user_id} (from cache)")
+        api_logger.info(f"Dimension portrait retrieved for user: {end_user_id} (from cache)")
         return success(data=portrait, msg="四维画像获取成功（缓存）")
         
     except Exception as e:
-        return handle_implicit_memory_error(e, "四维画像获取", user_id)
+        return handle_implicit_memory_error(e, "四维画像获取", end_user_id)
 
 
-@router.get("/interest-areas/{user_id}", response_model=ApiResponse)
+@router.get("/interest-areas/{end_user_id}", response_model=ApiResponse)
 @cur_workspace_access_guard()
 async def get_interest_area_distribution(
-    user_id: str,
+    end_user_id: str,
     include_trends: bool = Query(False, description="Include trend analysis"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user)
@@ -259,46 +259,46 @@ async def get_interest_area_distribution(
     Get user's interest area distribution from cache.
     
     Args:
-        user_id: Target user ID
+        end_user_id: Target end user ID
         include_trends: Whether to include trend analysis data (ignored for cached data)
         
     Returns:
         Interest area distribution from cache
     """
-    api_logger.info(f"Interest area distribution requested for user: {user_id} (from cache)")
+    api_logger.info(f"Interest area distribution requested for user: {end_user_id} (from cache)")
     
     try:
         # Validate inputs
-        validate_user_id(user_id)
+        validate_user_id(end_user_id)
         
         # Create service with user-specific config
-        service = ImplicitMemoryService(db=db, end_user_id=user_id)
+        service = ImplicitMemoryService(db=db, end_user_id=end_user_id)
         
         # Get cached profile
-        cached_profile = await service.get_cached_profile(end_user_id=user_id, db=db)
+        cached_profile = await service.get_cached_profile(end_user_id=end_user_id, db=db)
         
         if cached_profile is None:
-            api_logger.info(f"用户 {user_id} 的画像缓存不存在或已过期")
+            api_logger.info(f"用户 {end_user_id} 的画像缓存不存在或已过期")
             return fail(
-                BizCode.RESOURCE_NOT_FOUND,
-                "画像缓存不存在或已过期，请调用 /generate_profile 接口生成新画像",
-                None
+                BizCode.NOT_FOUND,
+                "画像缓存不存在或已过期，请右上角刷新生成新画像",
+                ""
             )
         
         # Extract interest areas from cache
         interest_areas = cached_profile.get("interest_areas", {})
         
-        api_logger.info(f"Interest area distribution retrieved for user: {user_id} (from cache)")
+        api_logger.info(f"Interest area distribution retrieved for user: {end_user_id} (from cache)")
         return success(data=interest_areas, msg="兴趣领域分布获取成功（缓存）")
         
     except Exception as e:
-        return handle_implicit_memory_error(e, "兴趣领域分布获取", user_id)
+        return handle_implicit_memory_error(e, "兴趣领域分布获取", end_user_id)
 
 
-@router.get("/habits/{user_id}", response_model=ApiResponse)
+@router.get("/habits/{end_user_id}", response_model=ApiResponse)
 @cur_workspace_access_guard()
 async def get_behavior_habits(
-    user_id: str,
+    end_user_id: str,
     confidence_level: Optional[str] = Query(None, regex="^(high|medium|low)$", description="Filter by confidence level"),
     frequency_pattern: Optional[str] = Query(None, regex="^(daily|weekly|monthly|seasonal|occasional|event_triggered)$", description="Filter by frequency pattern"),
     time_period: Optional[str] = Query(None, regex="^(current|past)$", description="Filter by time period"),
@@ -309,7 +309,7 @@ async def get_behavior_habits(
     Get user's behavioral habits from cache.
     
     Args:
-        user_id: Target user ID
+        end_user_id: Target end user ID
         confidence_level: Filter by confidence level (high, medium, low)
         frequency_pattern: Filter by frequency pattern (daily, weekly, monthly, seasonal, occasional, event_triggered)
         time_period: Filter by time period (current, past)
@@ -317,24 +317,24 @@ async def get_behavior_habits(
     Returns:
         List of behavioral habits from cache
     """
-    api_logger.info(f"Behavior habits requested for user: {user_id} (from cache)")
+    api_logger.info(f"Behavior habits requested for user: {end_user_id} (from cache)")
     
     try:
         # Validate inputs
-        validate_user_id(user_id)
+        validate_user_id(end_user_id)
         
         # Create service with user-specific config
-        service = ImplicitMemoryService(db=db, end_user_id=user_id)
+        service = ImplicitMemoryService(db=db, end_user_id=end_user_id)
         
         # Get cached profile
-        cached_profile = await service.get_cached_profile(end_user_id=user_id, db=db)
+        cached_profile = await service.get_cached_profile(end_user_id=end_user_id, db=db)
         
         if cached_profile is None:
-            api_logger.info(f"用户 {user_id} 的画像缓存不存在或已过期")
+            api_logger.info(f"用户 {end_user_id} 的画像缓存不存在或已过期")
             return fail(
-                BizCode.RESOURCE_NOT_FOUND,
-                "画像缓存不存在或已过期，请调用 /generate_profile 接口生成新画像",
-                None
+                BizCode.NOT_FOUND,
+                "画像缓存不存在或已过期，请右上角刷新生成新画像",
+                ""
             )
         
         # Extract habits from cache
@@ -368,11 +368,11 @@ async def get_behavior_habits(
             
             filtered_habits.append(habit)
         
-        api_logger.info(f"Retrieved {len(filtered_habits)} behavior habits for user: {user_id} (from cache)")
+        api_logger.info(f"Retrieved {len(filtered_habits)} behavior habits for user: {end_user_id} (from cache)")
         return success(data=filtered_habits, msg="行为习惯获取成功（缓存）")
         
     except Exception as e:
-        return handle_implicit_memory_error(e, "行为习惯获取", user_id)
+        return handle_implicit_memory_error(e, "行为习惯获取", end_user_id)
 
 
 

api/app/controllers/memory_agent_controller.py

@@ -9,14 +9,16 @@ from app.db import get_db
 from app.dependencies import cur_workspace_access_guard, get_current_user
 from app.models import ModelApiKey
 from app.models.user_model import User
-from app.repositories import knowledge_repository
+from app.core.memory.agent.utils.session_tools import SessionService
+from app.core.memory.agent.utils.redis_tool import store
+from app.repositories import knowledge_repository, WorkspaceRepository
 from app.schemas.memory_agent_schema import UserInput, Write_UserInput
 from app.schemas.response_schema import ApiResponse
 from app.services import task_service, workspace_service
 from app.services.memory_agent_service import MemoryAgentService
 from app.services.model_service import ModelConfigService
 from dotenv import load_dotenv
-from fastapi import APIRouter, Depends, File, Form, Query, UploadFile
+from fastapi import APIRouter, Depends, File, Form, Query, UploadFile,Header
 from sqlalchemy.orm import Session
 from starlette.responses import StreamingResponse
 
@@ -123,7 +125,7 @@ async def write_server(
     Write service endpoint - processes write operations synchronously
     
     Args:
-        user_input: Write request containing message and group_id
+        user_input: Write request containing message and end_user_id
     
     Returns:
         Response with write operation status
@@ -158,16 +160,18 @@ async def write_server(
             api_logger.warning("workspace_id 为空，无法使用 rag 存储，将使用 neo4j 存储")
             storage_type = 'neo4j'
     
-    api_logger.info(f"Write service requested for group {user_input.group_id}, storage_type: {storage_type}, user_rag_memory_id: {user_rag_memory_id}")
+    api_logger.info(f"Write service requested for group {user_input.end_user_id}, storage_type: {storage_type}, user_rag_memory_id: {user_rag_memory_id}")
     try:
+        messages_list = memory_agent_service.get_messages_list(user_input)
         result = await memory_agent_service.write_memory(
-            user_input.group_id, 
-            user_input.message, 
+            user_input.end_user_id,
+            messages_list,
             config_id,
             db,
             storage_type, 
             user_rag_memory_id
         )
+
         return success(data=result, msg="写入成功")
     except BaseException as e:
         # Handle ExceptionGroup from TaskGroup (Python 3.11+) or BaseExceptionGroup
@@ -191,7 +195,7 @@ async def write_server_async(
     Async write service endpoint - enqueues write processing to Celery
     
     Args:
-        user_input: Write request containing message and group_id
+        user_input: Write request containing message and end_user_id
     
     Returns:
         Task ID for tracking async operation
@@ -219,9 +223,12 @@ async def write_server_async(
         if knowledge: user_rag_memory_id = str(knowledge.id)
     api_logger.info(f"Async write: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
     try:
+        # 获取标准化的消息列表
+        messages_list = memory_agent_service.get_messages_list(user_input)
+
         task = celery_app.send_task(
             "app.core.memory.agent.write_message",
-            args=[user_input.group_id, user_input.message, config_id, storage_type, user_rag_memory_id]
+            args=[user_input.end_user_id, messages_list, config_id, storage_type, user_rag_memory_id]
         )
         api_logger.info(f"Write task queued: {task.id}")
         
@@ -247,16 +254,14 @@ async def read_server(
     - "2": Direct answer based on context
     
     Args:
-        user_input: Read request with message, history, search_switch, and group_id
+        user_input: Read request with message, history, search_switch, and end_user_id
     
     Returns:
         Response with query answer
     """
     config_id = user_input.config_id
     workspace_id = current_user.current_workspace_id
-    api_logger.info(f"Read service: workspace_id={workspace_id}, config_id={config_id}")
 
-    # 获取 storage_type，如果为 None 则使用默认值
     storage_type = workspace_service.get_workspace_storage_type(
         db=db,
         workspace_id=workspace_id,
@@ -271,12 +276,13 @@ async def read_server(
             name="USER_RAG_MERORY",
             workspace_id=workspace_id
         )
-        if knowledge: user_rag_memory_id = str(knowledge.id)
+        if knowledge:
+            user_rag_memory_id = str(knowledge.id)
     
-    api_logger.info(f"Read service: group={user_input.group_id}, storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}, workspace_id={workspace_id}")
+    api_logger.info(f"Read service: group={user_input.end_user_id}, storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}, workspace_id={workspace_id}")
     try:
         result = await memory_agent_service.read_memory(
-            user_input.group_id,
+            user_input.end_user_id,
             user_input.message,
             user_input.history,
             user_input.search_switch,
@@ -285,6 +291,22 @@ async def read_server(
             storage_type,
             user_rag_memory_id
         )
+        if str(user_input.search_switch) == "2":
+            retrieve_info = result['answer']
+            history = await SessionService(store).get_history(user_input.end_user_id, user_input.end_user_id, user_input.end_user_id)
+            query = user_input.message
+
+            # 调用 memory_agent_service 的方法生成最终答案
+            result['answer'] = await memory_agent_service.generate_summary_from_retrieve(
+                end_user_id=user_input.end_user_id,
+                retrieve_info=retrieve_info,
+                history=history,
+                query=query,
+                config_id=config_id,
+                db=db
+            )
+            if "信息不足，无法回答" in result['answer']:
+                result['answer']=retrieve_info
         return success(data=result, msg="回复对话消息成功")
     except BaseException as e:
         # Handle ExceptionGroup from TaskGroup (Python 3.11+) or BaseExceptionGroup
@@ -382,7 +404,7 @@ async def read_server_async(
     try:
         task = celery_app.send_task(
             "app.core.memory.agent.read_message",
-            args=[user_input.group_id, user_input.message, user_input.history, user_input.search_switch,
+            args=[user_input.end_user_id, user_input.message, user_input.history, user_input.search_switch,
                   config_id, storage_type, user_rag_memory_id]
         )
         api_logger.info(f"Read task queued: {task.id}")
@@ -426,7 +448,7 @@ async def get_read_task_result(
                 return success(
                     data={
                         "result": task_result.get("result"),
-                        "group_id": task_result.get("group_id"),
+                        "end_user_id": task_result.get("end_user_id"),
                         "elapsed_time": task_result.get("elapsed_time"),
                         "task_id": task_id
                     },
@@ -503,7 +525,7 @@ async def get_write_task_result(
                 return success(
                     data={
                         "result": task_result.get("result"),
-                        "group_id": task_result.get("group_id"),
+                        "end_user_id": task_result.get("end_user_id"),
                         "elapsed_time": task_result.get("elapsed_time"),
                         "task_id": task_id
                     },
@@ -557,15 +579,30 @@ async def status_type(
     Determine the type of user message (read or write)
     
     Args:
-        user_input: Request containing user message and group_id
+        user_input: Request containing user message and end_user_id
     
     Returns:
         Type classification result
     """
-    api_logger.info(f"Status type check requested for group {user_input.group_id}")
+    api_logger.info(f"Status type check requested for group {user_input.end_user_id}")
     try:
+        # 获取标准化的消息列表
+        messages_list = memory_agent_service.get_messages_list(user_input)
+
+        # 将消息列表转换为字符串用于分类
+        # 只取最后一条用户消息进行分类
+        last_user_message = ""
+        for msg in reversed(messages_list):
+            if msg.get('role') == 'user':
+                last_user_message = msg.get('content', '')
+                break
+
+        if not last_user_message:
+            # 如果没有用户消息，使用所有消息的内容
+            last_user_message = " ".join([msg.get('content', '') for msg in messages_list])
+
         result = await memory_agent_service.classify_message_type(
-            user_input.message,
+            last_user_message,
             user_input.config_id,
             db
         )
@@ -588,7 +625,7 @@ async def get_knowledge_type_stats_api(
     会对缺失类型补 0，返回字典形式。
     可选按状态过滤。
     - 知识库类型根据当前用户的 current_workspace_id 过滤
-    - memory 是 Neo4j 中 Chunk 的数量，根据 end_user_id (group_id) 过滤
+    - memory 是 Neo4j 中 Chunk 的数量，根据 end_user_id (end_user_id) 过滤
     - 如果用户没有当前工作空间或未提供 end_user_id，对应的统计返回 0
     """
     api_logger.info(f"Knowledge type stats requested for workspace_id: {current_user.current_workspace_id}, end_user_id: {end_user_id}")
@@ -616,8 +653,10 @@ async def get_knowledge_type_stats_api(
 @router.get("/analytics/hot_memory_tags/by_user", response_model=ApiResponse)
 async def get_hot_memory_tags_by_user_api(
     end_user_id: Optional[str] = Query(None, description="用户ID（可选）"),
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     limit: int = Query(20, description="返回标签数量限制"),
-    current_user: User = Depends(get_current_user)
+    current_user: User = Depends(get_current_user),
+    db: Session=Depends(get_db),
 ):
     """
     获取指定用户的热门记忆标签
@@ -628,10 +667,22 @@ async def get_hot_memory_tags_by_user_api(
         ...
     ]
     """
+
+    workspace_id=current_user.current_workspace_id
+    workspace_repo = WorkspaceRepository(db)
+    workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
+
+    if workspace_models:
+        model_id = workspace_models.get("llm", None)
+    else:
+        model_id = None
+
     api_logger.info(f"Hot memory tags by user requested: end_user_id={end_user_id}")
     try:
         result = await memory_agent_service.get_hot_memory_tags_by_user(
             end_user_id=end_user_id,
+            language_type=language_type,
+            model_id=model_id,
             limit=limit
         )
         return success(data=result, msg="获取热门记忆标签成功")

api/app/controllers/memory_dashboard_controller.py

@@ -5,7 +5,6 @@ from app.core.response_utils import success
 from app.db import get_db
 from app.dependencies import get_current_user
 from app.models.user_model import User
-from app.schemas.memory_agent_schema import End_User_Information
 from app.schemas.response_schema import ApiResponse
 
 from app.services import memory_dashboard_service, memory_storage_service, workspace_service
@@ -40,54 +39,7 @@ def get_workspace_total_end_users(
     api_logger.info(f"成功获取最新用户总数: total_num={total_end_users.get('total_num', 0)}")
     return success(data=total_end_users, msg="用户数量获取成功")
 
-@router.post("/update/end_users", response_model=ApiResponse)
-async def update_workspace_end_users(
-    user_input: End_User_Information,
-    db: Session = Depends(get_db),
-    current_user: User = Depends(get_current_user),
-):
-    """
-    更新工作空间的宿主信息
-    """
-    username = user_input.end_user_name  # 要更新的用户名
-    end_user_input_id = user_input.id  # 宿主ID
-    workspace_id = current_user.current_workspace_id
-    
-    api_logger.info(f"用户 {current_user.username} 请求更新工作空间 {workspace_id} 的宿主信息")
-    api_logger.info(f"更新参数: username={username}, end_user_id={end_user_input_id}")
 
-    try:
-        # 导入更新函数
-        from app.repositories.end_user_repository import update_end_user_other_name
-        import uuid
-        
-        # 转换 end_user_id 为 UUID 类型
-        end_user_uuid = uuid.UUID(end_user_input_id)
-        
-        # 直接更新数据库中的 other_name 字段
-        updated_count = update_end_user_other_name(
-            db=db,
-            end_user_id=end_user_uuid,
-            other_name=username
-        )
-        
-        api_logger.info(f"成功更新宿主 {end_user_input_id} 的 other_name 为: {username}")
-
-        return success(
-            data={
-                "updated_count": updated_count,
-                "end_user_id": end_user_input_id,
-                "updated_other_name": username
-            },
-            msg=f"成功更新 {updated_count} 个宿主的信息"
-        )
-        
-    except Exception as e:
-        api_logger.error(f"更新宿主信息失败: {str(e)}")
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail=f"更新宿主信息失败: {str(e)}"
-        )
 
 
 
@@ -97,63 +49,134 @@ async def get_workspace_end_users(
     current_user: User = Depends(get_current_user),
 ):
     """
-    获取工作空间的宿主列表
+    获取工作空间的宿主列表（高性能优化版本 v2）
     
-    返回格式与原 memory_list 接口中的 end_users 字段相同，
-    并包含每个用户的记忆配置信息（memory_config_id 和 memory_config_name）
+    优化策略：
+    1. 批量查询 end_users（一次查询而非循环）
+    2. 并发查询所有用户的记忆数量（Neo4j）
+    3. RAG 模式使用批量查询（一次 SQL）
+    4. 只返回必要字段减少数据传输
+    5. 添加短期缓存减少重复查询
+    6. 并发执行配置查询和记忆数量查询
+    
+    返回格式：
+    {
+        "end_user": {"id": "uuid", "other_name": "名称"},
+        "memory_num": {"total": 数量},
+        "memory_config": {"memory_config_id": "id", "memory_config_name": "名称"}
+    }
     """
+    import asyncio
+    import json
+    from app.aioRedis import aio_redis_get, aio_redis_set
+    
     workspace_id = current_user.current_workspace_id
+    
+    # 尝试从缓存获取（30秒缓存）
+    cache_key = f"end_users:workspace:{workspace_id}"
+    try:
+        cached_data = await aio_redis_get(cache_key)
+        if cached_data:
+            api_logger.info(f"从缓存获取宿主列表: workspace_id={workspace_id}")
+            return success(data=json.loads(cached_data), msg="宿主列表获取成功")
+    except Exception as e:
+        api_logger.warning(f"Redis 缓存读取失败: {str(e)}")
+    
     # 获取当前空间类型
     current_workspace_type = memory_dashboard_service.get_current_workspace_type(db, workspace_id, current_user)
     api_logger.info(f"用户 {current_user.username} 请求获取工作空间 {workspace_id} 的宿主列表")
+    
+    # 获取 end_users（已优化为批量查询）
     end_users = memory_dashboard_service.get_workspace_end_users(
         db=db,
         workspace_id=workspace_id,
         current_user=current_user
     )
-    
-    # 批量获取所有用户的记忆配置信息（优化：一次查询而非 N 次）
-    end_user_ids = [str(user.id) for user in end_users]
-    memory_configs_map = {}
-    if end_user_ids:
+    if not end_users:
+        api_logger.info("工作空间下没有宿主")
+        # 缓存空结果，避免重复查询
         try:
-            memory_configs_map = get_end_users_connected_configs_batch(end_user_ids, db)
+            await aio_redis_set(cache_key, json.dumps([]), expire=30)
+        except Exception as e:
+            api_logger.warning(f"Redis 缓存写入失败: {str(e)}")
+        return success(data=[], msg="宿主列表获取成功")
+    
+    end_user_ids = [str(user.id) for user in end_users]
+    
+    # 并发执行两个独立的查询任务
+    async def get_memory_configs():
+        """获取记忆配置（在线程池中执行同步查询）"""
+        try:
+            return await asyncio.to_thread(
+                get_end_users_connected_configs_batch,
+                end_user_ids, db
+            )
         except Exception as e:
             api_logger.error(f"批量获取记忆配置失败: {str(e)}")
-            # 失败时使用空字典，不影响其他数据返回
+            return {}
     
+    async def get_memory_nums():
+        """获取记忆数量"""
+        if current_workspace_type == "rag":
+            # RAG 模式：批量查询
+            try:
+                chunk_map = await asyncio.to_thread(
+                    memory_dashboard_service.get_users_total_chunk_batch,
+                    end_user_ids, db, current_user
+                )
+                return {uid: {"total": count} for uid, count in chunk_map.items()}
+            except Exception as e:
+                api_logger.error(f"批量获取 RAG chunk 数量失败: {str(e)}")
+                return {uid: {"total": 0} for uid in end_user_ids}
+        
+        elif current_workspace_type == "neo4j":
+            # Neo4j 模式：并发查询（带并发限制）
+            # 使用信号量限制并发数，避免大量用户时压垮 Neo4j
+            MAX_CONCURRENT_QUERIES = 10
+            semaphore = asyncio.Semaphore(MAX_CONCURRENT_QUERIES)
+            
+            async def get_neo4j_memory_num(end_user_id: str):
+                async with semaphore:
+                    try:
+                        return await memory_storage_service.search_all(end_user_id)
+                    except Exception as e:
+                        api_logger.error(f"获取用户 {end_user_id} Neo4j 记忆数量失败: {str(e)}")
+                        return {"total": 0}
+            
+            memory_nums_list = await asyncio.gather(*[get_neo4j_memory_num(uid) for uid in end_user_ids])
+            return {end_user_ids[i]: memory_nums_list[i] for i in range(len(end_user_ids))}
+        
+        return {uid: {"total": 0} for uid in end_user_ids}
+    
+    # 并发执行配置查询和记忆数量查询
+    memory_configs_map, memory_nums_map = await asyncio.gather(
+        get_memory_configs(),
+        get_memory_nums()
+    )
+    
+    # 构建结果（优化：使用列表推导式）
     result = []
     for end_user in end_users:
-        memory_num = {}
-        if current_workspace_type == "neo4j":
-            # EndUser 是 Pydantic 模型，直接访问属性而不是使用 .get()
-            memory_num = await memory_storage_service.search_all(str(end_user.id))
-        elif current_workspace_type == "rag":
-            memory_num = {
-                "total":memory_dashboard_service.get_current_user_total_chunk(str(end_user.id), db, current_user)
-            }
-        
-        # 从批量查询结果中获取配置信息
         user_id = str(end_user.id)
-        memory_config_info = memory_configs_map.get(user_id, {
-            "memory_config_id": None,
-            "memory_config_name": None
-        })
-        
-        # 只保留需要的字段，移除 error 字段（如果有）
-        memory_config = {
-            "memory_config_id": memory_config_info.get("memory_config_id"),
-            "memory_config_name": memory_config_info.get("memory_config_name")
-        }
-        
-        result.append(
-            {
-                'end_user': end_user,
-                'memory_num': memory_num,
-                'memory_config': memory_config
+        config_info = memory_configs_map.get(user_id, {})
+        result.append({
+            'end_user': {
+                'id': user_id,
+                'other_name': end_user.other_name
+            },
+            'memory_num': memory_nums_map.get(user_id, {"total": 0}),
+            'memory_config': {
+                "memory_config_id": config_info.get("memory_config_id"),
+                "memory_config_name": config_info.get("memory_config_name")
             }
-        )
-        
+        })
+    
+    # 写入缓存（30秒过期）
+    try:
+        await aio_redis_set(cache_key, json.dumps(result), expire=30)
+    except Exception as e:
+        api_logger.warning(f"Redis 缓存写入失败: {str(e)}")
+    
     api_logger.info(f"成功获取 {len(end_users)} 个宿主记录")
     return success(data=result, msg="宿主列表获取成功")
 

api/app/controllers/memory_forget_controller.py

@@ -11,6 +11,7 @@
 """
 
 from typing import Optional
+from uuid import UUID
 
 from fastapi import APIRouter, Depends
 from sqlalchemy.orm import Session
@@ -33,7 +34,7 @@ from app.schemas.memory_storage_schema import (
 )
 from app.schemas.response_schema import ApiResponse
 from app.services.memory_forget_service import MemoryForgetService
-
+from app.utils.config_utils import resolve_config_id
 
 # 获取API专用日志器
 api_logger = get_api_logger()
@@ -83,7 +84,8 @@ async def trigger_forgetting_cycle(
         
         connected_config = get_end_user_connected_config(end_user_id, db)
         config_id = connected_config.get("memory_config_id")
-        
+        config_id = resolve_config_id((config_id), db)
+
         if config_id is None:
             api_logger.warning(f"终端用户 {end_user_id} 未关联记忆配置")
             return fail(BizCode.INVALID_PARAMETER, f"终端用户 {end_user_id} 未关联记忆配置", "memory_config_id is None")
@@ -106,7 +108,7 @@ async def trigger_forgetting_cycle(
         # 调用服务层执行遗忘周期
         report = await forget_service.trigger_forgetting_cycle(
             db=db,
-            group_id=end_user_id,  # 服务层方法的参数名是 group_id
+            end_user_id=end_user_id,  # 服务层方法的参数名是 end_user_id
             max_merge_batch_size=payload.max_merge_batch_size,
             min_days_since_access=payload.min_days_since_access,
             config_id=config_id
@@ -128,7 +130,7 @@ async def trigger_forgetting_cycle(
 
 @router.get("/read_config", response_model=ApiResponse)
 async def read_forgetting_config(
-    config_id: int,
+    config_id: UUID|int,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db)
 ):
@@ -157,6 +159,7 @@ async def read_forgetting_config(
     )
     
     try:
+        config_id=resolve_config_id(config_id, db)
         # 调用服务层读取配置
         config = forget_service.read_forgetting_config(db=db, config_id=config_id)
         
@@ -194,6 +197,8 @@ async def update_forgetting_config(
         ApiResponse: 包含更新结果的响应
     """
     workspace_id = current_user.current_workspace_id
+    payload.config_id=resolve_config_id((payload.config_id), db)
+
     
     # 检查用户是否已选择工作空间
     if workspace_id is None:
@@ -236,7 +241,7 @@ async def update_forgetting_config(
 
 @router.get("/stats", response_model=ApiResponse)
 async def get_forgetting_stats(
-    group_id: Optional[str] = None,
+    end_user_id: Optional[str] = None,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db)
 ):
@@ -246,7 +251,7 @@ async def get_forgetting_stats(
     返回知识层节点统计、激活值分布等信息。
     
     Args:
-        group_id: 组ID（即 end_user_id，可选）
+        end_user_id: 组ID（即 end_user_id，可选）
         current_user: 当前用户
         db: 数据库会话
     
@@ -254,26 +259,25 @@ async def get_forgetting_stats(
         ApiResponse: 包含统计信息的响应
     """
     workspace_id = current_user.current_workspace_id
-    
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试获取遗忘引擎统计但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
-    
-    # 如果提供了 group_id，通过它获取 config_id
+    # 如果提供了 end_user_id，通过它获取 config_id
     config_id = None
-    if group_id:
+    if end_user_id:
         try:
             from app.services.memory_agent_service import get_end_user_connected_config
             
-            connected_config = get_end_user_connected_config(group_id, db)
+            connected_config = get_end_user_connected_config(end_user_id, db)
             config_id = connected_config.get("memory_config_id")
+            config_id = resolve_config_id(config_id, db)
             
             if config_id is None:
-                api_logger.warning(f"终端用户 {group_id} 未关联记忆配置")
-                return fail(BizCode.INVALID_PARAMETER, f"终端用户 {group_id} 未关联记忆配置", "memory_config_id is None")
+                api_logger.warning(f"终端用户 {end_user_id} 未关联记忆配置")
+                return fail(BizCode.INVALID_PARAMETER, f"终端用户 {end_user_id} 未关联记忆配置", "memory_config_id is None")
             
-            api_logger.debug(f"通过 group_id={group_id} 获取到 config_id={config_id}")
+            api_logger.debug(f"通过 end_user_id={end_user_id} 获取到 config_id={config_id}")
         except ValueError as e:
             api_logger.warning(f"获取终端用户配置失败: {str(e)}")
             return fail(BizCode.INVALID_PARAMETER, str(e), "ValueError")
@@ -283,14 +287,14 @@ async def get_forgetting_stats(
     
     api_logger.info(
         f"用户 {current_user.username} 在工作空间 {workspace_id} 请求获取遗忘引擎统计: "
-        f"group_id={group_id}, config_id={config_id}"
+        f"end_user_id={end_user_id}, config_id={config_id}"
     )
     
     try:
         # 调用服务层获取统计信息
         stats = await forget_service.get_forgetting_stats(
             db=db,
-            group_id=group_id,
+            end_user_id=end_user_id,
             config_id=config_id
         )
         
@@ -324,7 +328,7 @@ async def get_forgetting_curve(
         ApiResponse: 包含遗忘曲线数据的响应
     """
     workspace_id = current_user.current_workspace_id
-    
+    request.config_id = resolve_config_id((request.config_id), db)
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试获取遗忘曲线但未选择工作空间")

api/app/controllers/memory_perceptual_controller.py

@@ -27,27 +27,27 @@ router = APIRouter(
 )
 
 
-@router.get("/{group_id}/count", response_model=ApiResponse)
+@router.get("/{end_user_id}/count", response_model=ApiResponse)
 def get_memory_count(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
     """Retrieve perceptual memory statistics for a user group.
 
     Args:
-        group_id: ID of the user group (usually end_user_id in this context)
+        end_user_id: ID of the user group (usually end_user_id in this context)
         current_user: Current authenticated user
         db: Database session
 
     Returns:
         ApiResponse: Response containing memory count statistics
     """
-    api_logger.info(f"Fetching perceptual memory statistics: user={current_user.username}, group_id={group_id}")
+    api_logger.info(f"Fetching perceptual memory statistics: user={current_user.username}, end_user_id={end_user_id}")
 
     try:
         service = MemoryPerceptualService(db)
-        count_stats = service.get_memory_count(group_id)
+        count_stats = service.get_memory_count(end_user_id)
 
         api_logger.info(f"Memory statistics fetched successfully: total={count_stats.get('total', 0)}")
 
@@ -57,37 +57,37 @@ def get_memory_count(
         )
 
     except Exception as e:
-        api_logger.error(f"Failed to fetch memory statistics: group_id={group_id}, error={str(e)}")
+        api_logger.error(f"Failed to fetch memory statistics: end_user_id={end_user_id}, error={str(e)}")
         return fail(
             code=BizCode.INTERNAL_ERROR,
             msg="Failed to fetch memory statistics",
         )
 
 
-@router.get("/{group_id}/last_visual", response_model=ApiResponse)
+@router.get("/{end_user_id}/last_visual", response_model=ApiResponse)
 def get_last_visual_memory(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
     """Retrieve the most recent VISION-type memory for a user.
 
     Args:
-        group_id: ID of the user group
+        end_user_id: ID of the user group
         current_user: Current authenticated user
         db: Database session
 
     Returns:
         ApiResponse: Metadata of the latest visual memory
     """
-    api_logger.info(f"Fetching latest visual memory: user={current_user.username}, group_id={group_id}")
+    api_logger.info(f"Fetching latest visual memory: user={current_user.username}, end_user_id={end_user_id}")
 
     try:
         service = MemoryPerceptualService(db)
-        visual_memory = service.get_latest_visual_memory(group_id)
+        visual_memory = service.get_latest_visual_memory(end_user_id)
 
         if visual_memory is None:
-            api_logger.info(f"No visual memory found: group_id={group_id}")
+            api_logger.info(f"No visual memory found: end_user_id={end_user_id}")
             return success(
                 data=None,
                 msg="No visual memory available"
@@ -101,37 +101,37 @@ def get_last_visual_memory(
         )
 
     except Exception as e:
-        api_logger.error(f"Failed to fetch latest visual memory: group_id={group_id}, error={str(e)}")
+        api_logger.error(f"Failed to fetch latest visual memory: end_user_id={end_user_id}, error={str(e)}")
         return fail(
             code=BizCode.INTERNAL_ERROR,
             msg="Failed to fetch latest visual memory",
         )
 
 
-@router.get("/{group_id}/last_listen", response_model=ApiResponse)
+@router.get("/{end_user_id}/last_listen", response_model=ApiResponse)
 def get_last_memory_listen(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
     """Retrieve the most recent AUDIO-type memory for a user.
 
     Args:
-        group_id: ID of the user group
+        end_user_id: ID of the user group
         current_user: Current authenticated user
         db: Database session
 
     Returns:
         ApiResponse: Metadata of the latest audio memory
     """
-    api_logger.info(f"Fetching latest audio memory: user={current_user.username}, group_id={group_id}")
+    api_logger.info(f"Fetching latest audio memory: user={current_user.username}, end_user_id={end_user_id}")
 
     try:
         service = MemoryPerceptualService(db)
-        audio_memory = service.get_latest_audio_memory(group_id)
+        audio_memory = service.get_latest_audio_memory(end_user_id)
 
         if audio_memory is None:
-            api_logger.info(f"No audio memory found: group_id={group_id}")
+            api_logger.info(f"No audio memory found: end_user_id={end_user_id}")
             return success(
                 data=None,
                 msg="No audio memory available"
@@ -145,38 +145,38 @@ def get_last_memory_listen(
         )
 
     except Exception as e:
-        api_logger.error(f"Failed to fetch latest audio memory: group_id={group_id}, error={str(e)}")
+        api_logger.error(f"Failed to fetch latest audio memory: end_user_id={end_user_id}, error={str(e)}")
         return fail(
             code=BizCode.INTERNAL_ERROR,
             msg="Failed to fetch latest audio memory",
         )
 
 
-@router.get("/{group_id}/last_text", response_model=ApiResponse)
+@router.get("/{end_user_id}/last_text", response_model=ApiResponse)
 def get_last_text_memory(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
     """Retrieve the most recent TEXT-type memory for a user.
 
     Args:
-        group_id: ID of the user group
+        end_user_id: ID of the user group
         current_user: Current authenticated user
         db: Database session
 
     Returns:
         ApiResponse: Metadata of the latest text memory
     """
-    api_logger.info(f"Fetching latest text memory: user={current_user.username}, group_id={group_id}")
+    api_logger.info(f"Fetching latest text memory: user={current_user.username}, end_user_id={end_user_id}")
 
     try:
         # 调用服务层获取最近的文本记忆
         service = MemoryPerceptualService(db)
-        text_memory = service.get_latest_text_memory(group_id)
+        text_memory = service.get_latest_text_memory(end_user_id)
 
         if text_memory is None:
-            api_logger.info(f"No text memory found: group_id={group_id}")
+            api_logger.info(f"No text memory found: end_user_id={end_user_id}")
             return success(
                 data=None,
                 msg="No text memory available"
@@ -190,16 +190,16 @@ def get_last_text_memory(
         )
 
     except Exception as e:
-        api_logger.error(f"Failed to fetch latest text memory: group_id={group_id}, error={str(e)}")
+        api_logger.error(f"Failed to fetch latest text memory: end_user_id={end_user_id}, error={str(e)}")
         return fail(
             code=BizCode.INTERNAL_ERROR,
             msg="Failed to fetch latest text memory",
         )
 
 
-@router.get("/{group_id}/timeline", response_model=ApiResponse)
+@router.get("/{end_user_id}/timeline", response_model=ApiResponse)
 def get_memory_time_line(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         perceptual_type: Optional[PerceptualType] = Query(None, description="感知类型过滤"),
         page: int = Query(1, ge=1, description="页码"),
         page_size: int = Query(10, ge=1, le=100, description="每页大小"),
@@ -209,7 +209,7 @@ def get_memory_time_line(
     """Retrieve a timeline of perceptual memories for a user group.
 
     Args:
-        group_id: ID of the user group
+        end_user_id: ID of the user group
         perceptual_type: Optional filter for perceptual type
         page: Page number for pagination
         page_size: Number of items per page
@@ -221,7 +221,7 @@ def get_memory_time_line(
     """
     api_logger.info(
         f"Fetching perceptual memory timeline: user={current_user.username}, "
-        f"group_id={group_id}, type={perceptual_type}, page={page}"
+        f"end_user_id={end_user_id}, type={perceptual_type}, page={page}"
     )
 
     try:
@@ -232,7 +232,7 @@ def get_memory_time_line(
         )
 
         service = MemoryPerceptualService(db)
-        timeline_data = service.get_time_line(group_id, query)
+        timeline_data = service.get_time_line(end_user_id, query)
 
         api_logger.info(
             f"Perceptual memory timeline retrieved successfully: total={timeline_data.total}, "
@@ -246,7 +246,7 @@ def get_memory_time_line(
 
     except Exception as e:
         api_logger.error(
-            f"Failed to fetch perceptual memory timeline: group_id={group_id}, "
+            f"Failed to fetch perceptual memory timeline: end_user_id={end_user_id}, "
             f"error={str(e)}"
         )
         return fail(

api/app/controllers/memory_reflection_controller.py

@@ -1,16 +1,18 @@
 import asyncio
 import time
+import uuid
+from uuid import UUID
 
 from app.core.logging_config import get_api_logger
 from app.core.memory.storage_services.reflection_engine.self_reflexion import (
     ReflectionConfig,
-    ReflectionEngine,
+    ReflectionEngine, ReflectionRange, ReflectionBaseline,
 )
 from app.core.response_utils import success
 from app.db import get_db
 from app.dependencies import get_current_user
 from app.models.user_model import User
-from app.repositories.data_config_repository import DataConfigRepository
+from app.repositories.memory_config_repository import MemoryConfigRepository
 from app.repositories.neo4j.neo4j_connector import Neo4jConnector
 from app.schemas.memory_reflection_schemas import Memory_Reflection
 from app.services.memory_reflection_service import (
@@ -19,10 +21,12 @@ from app.services.memory_reflection_service import (
 )
 from app.services.model_service import ModelConfigService
 from dotenv import load_dotenv
-from fastapi import APIRouter, Depends, HTTPException, status
+from fastapi import APIRouter, Depends, HTTPException, status,Header
 from sqlalchemy import text
 from sqlalchemy.orm import Session
 
+from app.utils.config_utils import resolve_config_id
+
 load_dotenv()
 api_logger = get_api_logger()
 
@@ -39,64 +43,40 @@ async def save_reflection_config(
     db: Session = Depends(get_db),
 ) -> dict:
     """Save reflection configuration to data_comfig table"""
-    
-
-    
     try:
         config_id = request.config_id
+        config_id = resolve_config_id(config_id, db)
         if not config_id:
             raise HTTPException(
                 status_code=status.HTTP_400_BAD_REQUEST,
                 detail="缺少必需参数: config_id"
             )
-
         api_logger.info(f"用户 {current_user.username} 保存反思配置，config_id: {config_id}")
 
-        update_params = {
-            "enable_self_reflexion": request.reflection_enabled,
-            "iteration_period": request.reflection_period_in_hours,
-            "reflexion_range": request.reflexion_range,
-            "baseline": request.baseline,
-            "reflection_model_id": request.reflection_model_id,
-            "memory_verify": request.memory_verify,
-            "quality_assessment": request.quality_assessment,
-        }
+        memory_config = MemoryConfigRepository.update_reflection_config(
+            db,
+            config_id=config_id,
+            enable_self_reflexion=request.reflection_enabled,
+            iteration_period=request.reflection_period_in_hours,
+            reflexion_range=request.reflexion_range,
+            baseline=request.baseline,
+            reflection_model_id=request.reflection_model_id,
+            memory_verify=request.memory_verify,
+            quality_assessment=request.quality_assessment
+        )
 
-
-
-        query, params = DataConfigRepository.build_update_reflection(config_id, **update_params)
-
-        result = db.execute(text(query), params)
-        if result.rowcount == 0:
-            raise HTTPException(
-                status_code=status.HTTP_404_NOT_FOUND,
-                detail=f"未找到config_id为 {config_id} 的配置"
-            )
-        
         db.commit()
-        
-        # 查询更新后的配置
-        select_query, select_params = DataConfigRepository.build_select_reflection(config_id)
-        result = db.execute(text(select_query), select_params).fetchone()
-        
-        if not result:
-            raise HTTPException(
-                status_code=status.HTTP_404_NOT_FOUND,
-                detail=f"更新后未找到config_id为 {config_id} 的配置"
-            )
-        
-        api_logger.info(f"成功保存反思配置到数据库，config_id: {config_id}")
+        db.refresh(memory_config)
 
         reflection_result={
-                "config_id": result.config_id,
-                "enable_self_reflexion": result.enable_self_reflexion,
-                "iteration_period": result.iteration_period,
-                "reflexion_range": result.reflexion_range,
-                "baseline": result.baseline,
-                "reflection_model_id": result.reflection_model_id,
-                "memory_verify": result.memory_verify,
-                "quality_assessment": result.quality_assessment,
-                "user_id": result.user_id}
+                "config_id": memory_config.config_id,
+                "enable_self_reflexion": memory_config.enable_self_reflexion,
+                "iteration_period": memory_config.iteration_period,
+                "reflexion_range": memory_config.reflexion_range,
+                "baseline": memory_config.baseline,
+                "reflection_model_id": memory_config.reflection_model_id,
+                "memory_verify": memory_config.memory_verify,
+                "quality_assessment": memory_config.quality_assessment}
 
         return success(data=reflection_result, msg="反思配置成功")
         
@@ -116,13 +96,12 @@ async def save_reflection_config(
         )
 
 
-@router.post("/reflection")
+@router.get("/reflection")
 async def start_workspace_reflection(
-    config_id: int,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
 ) -> dict:
-    """Activate the reflection function for all matching applications in the workspace"""
+    """启动工作空间中所有匹配应用的反思功能"""
     workspace_id = current_user.current_workspace_id
     reflection_service = MemoryReflectionService(db)
 
@@ -131,33 +110,55 @@ async def start_workspace_reflection(
 
         service = WorkspaceAppService(db)
         result = service.get_workspace_apps_detailed(workspace_id)
-        
         reflection_results = []
-        
         for data in result['apps_detailed_info']:
-            if data['data_configs'] == []: 
+            # 跳过没有配置的应用
+            if not data['memory_configs']:
+                api_logger.debug(f"应用 {data['id']} 没有memory_configs，跳过")
                 continue
-                
+
             releases = data['releases']
-            data_configs = data['data_configs']
+            memory_configs = data['memory_configs']
             end_users = data['end_users']
-            
-            for base, config, user in zip(releases, data_configs, end_users):
-                if int(base['config']) == int(config['config_id']) and base['app_id'] == user['app_id']:
-                    # 调用反思服务
-                    api_logger.info(f"为用户 {user['id']} 启动反思，config_id: {config['config_id']}")
-                    
-                    reflection_result = await reflection_service.start_reflection_from_data(
-                        config_data=config,
-                        end_user_id=user['id']
-                    )
-                    
-                    reflection_results.append({
-                        "app_id": base['app_id'],
-                        "config_id": config['config_id'],
-                        "end_user_id": user['id'],
-                        "reflection_result": reflection_result
-                    })
+
+            # 为每个配置和用户组合执行反思
+            for config in memory_configs:
+                config_id_str = str(config['config_id'])
+
+                # 找到匹配此配置的所有release
+                matching_releases = [r for r in releases if str(r['config']) == config_id_str]
+
+                if not matching_releases:
+                    api_logger.debug(f"配置 {config_id_str} 没有匹配的release")
+                    continue
+
+                # 为每个用户执行反思
+                for user in end_users:
+                    api_logger.info(f"为用户 {user['id']} 启动反思，config_id: {config_id_str}")
+
+                    try:
+                        reflection_result = await reflection_service.start_text_reflection(
+                            config_data=config,
+                            end_user_id=user['id']
+                        )
+
+                        reflection_results.append({
+                            "app_id": data['id'],
+                            "config_id": config_id_str,
+                            "end_user_id": user['id'],
+                            "reflection_result": reflection_result
+                        })
+                    except Exception as e:
+                        api_logger.error(f"用户 {user['id']} 反思失败: {str(e)}")
+                        reflection_results.append({
+                            "app_id": data['id'],
+                            "config_id": config_id_str,
+                            "end_user_id": user['id'],
+                            "reflection_result": {
+                                "status": "错误",
+                                "message": f"反思失败: {str(e)}"
+                            }
+                        })
 
         return success(data=reflection_results, msg="反思配置成功")
 
@@ -171,35 +172,27 @@ async def start_workspace_reflection(
 
 @router.get("/reflection/configs")
 async def start_reflection_configs(
-        config_id: int,
+        config_id: uuid.UUID|int,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db),
 ) -> dict:
-    """通过config_id查询data_config表中的反思配置信息"""
+    """通过config_id查询memory_config表中的反思配置信息"""
+    config_id = resolve_config_id(config_id, db)
     try:
+        config_id=resolve_config_id(config_id,db)
         api_logger.info(f"用户 {current_user.username} 查询反思配置，config_id: {config_id}")
-        
-        # 使用DataConfigRepository查询反思配置
-        select_query, select_params = DataConfigRepository.build_select_reflection(config_id)
-        result = db.execute(text(select_query), select_params).fetchone()
-        
-        if not result:
-            raise HTTPException(
-                status_code=status.HTTP_404_NOT_FOUND,
-                detail=f"未找到config_id为 {config_id} 的配置"
-            )
-        
+        result = MemoryConfigRepository.query_reflection_config_by_id(db, config_id)
+        memory_config_id = resolve_config_id(result.config_id, db)
         # 构建返回数据
         reflection_config = {
-            "config_id": result.config_id,
+            "config_id": memory_config_id,
             "reflection_enabled": result.enable_self_reflexion,
             "reflection_period_in_hours": result.iteration_period,
             "reflexion_range": result.reflexion_range,
             "baseline": result.baseline,
             "reflection_model_id": result.reflection_model_id,
             "memory_verify": result.memory_verify,
-            "quality_assessment": result.quality_assessment,
-            "user_id": result.user_id
+            "quality_assessment": result.quality_assessment
         }
         api_logger.info(f"成功查询反思配置，config_id: {config_id}")
         return success(data=reflection_config, msg="反思配置查询成功")
@@ -217,19 +210,17 @@ async def start_reflection_configs(
 
 @router.get("/reflection/run")
 async def reflection_run(
-    config_id: int,
-    language_type: str = "zh",
+    config_id: UUID|int,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
 ) -> dict:
     """Activate the reflection function for all matching applications in the workspace"""
 
     api_logger.info(f"用户 {current_user.username} 查询反思配置，config_id: {config_id}")
-
-    # 使用DataConfigRepository查询反思配置
-    select_query, select_params = DataConfigRepository.build_select_reflection(config_id)
-    result = db.execute(text(select_query), select_params).fetchone()
-
+    config_id = resolve_config_id(config_id, db)
+    # 使用MemoryConfigRepository查询反思配置
+    result = MemoryConfigRepository.query_reflection_config_by_id(db, config_id)
     if not result:
         raise HTTPException(
             status_code=status.HTTP_404_NOT_FOUND,
@@ -242,7 +233,7 @@ async def reflection_run(
     model_id = result.reflection_model_id
     if model_id:
         try:
-            ModelConfigService.get_model_by_id(db=db, model_id=model_id)
+            ModelConfigService.get_model_by_id(db=db, model_id=uuid.UUID(model_id))
             api_logger.info(f"模型ID验证成功: {model_id}")
         except Exception as e:
             api_logger.warning(f"模型ID '{model_id}' 不存在，将使用默认模型: {str(e)}")
@@ -252,8 +243,8 @@ async def reflection_run(
     config = ReflectionConfig(
         enabled=result.enable_self_reflexion,
         iteration_period=result.iteration_period,
-        reflexion_range=result.reflexion_range,
-        baseline=result.baseline,
+        reflexion_range=ReflectionRange(result.reflexion_range),
+        baseline=ReflectionBaseline(result.baseline),
         output_example='',
         memory_verify=result.memory_verify,
         quality_assessment=result.quality_assessment,

api/app/controllers/memory_short_term_controller.py

@@ -1,4 +1,4 @@
-from fastapi import APIRouter, Depends, HTTPException, status
+from fastapi import APIRouter, Depends, HTTPException, status,Header
 from app.core.logging_config import get_api_logger
 from app.core.response_utils import success
 from app.db import get_db
@@ -20,6 +20,7 @@ router = APIRouter(
 @router.get("/short_term")
 async def short_term_configs(
         end_user_id: str,
+        language_type:str = Header(default="zh", alias="X-Language-Type"),
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db),
 ):

api/app/controllers/memory_storage_controller.py

@@ -1,10 +1,9 @@
 import os
-import uuid
 from typing import Optional
+from uuid import UUID
 
 from app.core.error_codes import BizCode
 from app.core.logging_config import get_api_logger
-from app.core.memory.utils.self_reflexion_utils import self_reflexion
 from app.core.response_utils import fail, success
 from app.db import get_db
 from app.dependencies import get_current_user
@@ -30,13 +29,14 @@ from app.services.memory_storage_service import (
     search_dialogue,
     search_edges,
     search_entity,
-    search_entity_graph,
     search_statement,
 )
 from fastapi import APIRouter, Depends
 from fastapi.responses import StreamingResponse
 from sqlalchemy.orm import Session
 
+from app.utils.config_utils import resolve_config_id
+
 # Get API logger
 api_logger = get_api_logger()
 
@@ -143,7 +143,6 @@ def create_config(
     db: Session = Depends(get_db),
     ) -> dict:
     workspace_id = current_user.current_workspace_id
-    
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试创建配置但未选择工作空间")
@@ -163,12 +162,12 @@ def create_config(
 
 @router.delete("/delete_config", response_model=ApiResponse)  # 删除数据库中的内容（按配置名称）
 def delete_config(
-    config_id: str,
+    config_id: UUID|int,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
     ) -> dict:
     workspace_id = current_user.current_workspace_id
-    
+    config_id=resolve_config_id(config_id, db)
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试删除配置但未选择工作空间")
@@ -190,12 +189,17 @@ def update_config(
     db: Session = Depends(get_db),
     ) -> dict:
     workspace_id = current_user.current_workspace_id
-    
+    payload.config_id = resolve_config_id(payload.config_id, db)
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试更新配置但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
     
+    # 校验至少有一个字段需要更新
+    if payload.config_name is None and payload.config_desc is None and payload.scene_id is None:
+        api_logger.warning(f"用户 {current_user.username} 尝试更新配置但未提供任何更新字段")
+        return fail(BizCode.INVALID_PARAMETER, "请至少提供一个需要更新的字段", "config_name, config_desc, scene_id 均为空")
+    
     api_logger.info(f"用户 {current_user.username} 在工作空间 {workspace_id} 请求更新配置: {payload.config_id}")
     try:
         svc = DataConfigService(db)
@@ -213,7 +217,7 @@ def update_config_extracted(
     db: Session = Depends(get_db),
     ) -> dict:
     workspace_id = current_user.current_workspace_id
-    
+    payload.config_id = resolve_config_id(payload.config_id, db)
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试更新提取配置但未选择工作空间")
@@ -235,12 +239,12 @@ def update_config_extracted(
 
 @router.get("/read_config_extracted", response_model=ApiResponse) # 通过查询参数读取某条配置（固定路径） 没有意义的话就删除
 def read_config_extracted(
-    config_id: str,
+    config_id: UUID | int,
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
     ) -> dict:
     workspace_id = current_user.current_workspace_id
-    
+    config_id = resolve_config_id(config_id, db)
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试读取提取配置但未选择工作空间")
@@ -288,6 +292,7 @@ async def pilot_run(
         f"Pilot run requested: config_id={payload.config_id}, "
         f"dialogue_text_length={len(payload.dialogue_text)}"
     )
+    payload.config_id = resolve_config_id(payload.config_id, db)
     svc = DataConfigService(db)
     return StreamingResponse(
         svc.pilot_run_stream(payload),
@@ -414,21 +419,7 @@ async def search_entity_edges(
         api_logger.error(f"Search edges failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "边查询失败", str(e))
 
-@router.get("/search/entity_graph", response_model=ApiResponse)
-async def search_for_entity_graph(
-    end_user_id: Optional[str] = None,
-    current_user: User = Depends(get_current_user),
-    ) -> dict:
-    """
-    搜索所有实体之间的关系网络
-    """
-    api_logger.info(f"Search entity graph requested for end_user_id: {end_user_id}")
-    try:
-        result = await search_entity_graph(end_user_id)
-        return success(data=result, msg="查询成功")
-    except Exception as e:
-        api_logger.error(f"Search entity graph failed: {str(e)}")
-        return fail(BizCode.INTERNAL_ERROR, "实体图查询失败", str(e))
+
 
 
 @router.get("/analytics/hot_memory_tags", response_model=ApiResponse)
@@ -437,15 +428,95 @@ async def get_hot_memory_tags_api(
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user),
     ) -> dict:
-    api_logger.info(f"Hot memory tags requested for current_user: {current_user.id}")
+    """
+    获取热门记忆标签（带Redis缓存）
+    
+    缓存策略：
+    - 缓存键：workspace_id + limit
+    - 过期时间：5分钟（300秒）
+    - 缓存命中：~50ms
+    - 缓存未命中：~600-800ms（取决于LLM速度）
+    """
+    workspace_id = current_user.current_workspace_id
+    
+    # 构建缓存键
+    cache_key = f"hot_memory_tags:{workspace_id}:{limit}"
+    
+    api_logger.info(f"Hot memory tags requested for workspace: {workspace_id}, limit: {limit}")
+    
     try:
+        # 尝试从Redis缓存获取
+        from app.aioRedis import aio_redis_get, aio_redis_set
+        import json
+        
+        cached_result = await aio_redis_get(cache_key)
+        if cached_result:
+            api_logger.info(f"Cache hit for key: {cache_key}")
+            try:
+                data = json.loads(cached_result)
+                return success(data=data, msg="查询成功（缓存）")
+            except json.JSONDecodeError:
+                api_logger.warning(f"Failed to parse cached data, will refresh")
+        
+        # 缓存未命中，执行查询
+        api_logger.info(f"Cache miss for key: {cache_key}, executing query")
         result = await analytics_hot_memory_tags(db, current_user, limit)
+        
+        # 写入缓存（过期时间：5分钟）
+        # 注意：result是列表，需要转换为JSON字符串
+        try:
+            cache_data = json.dumps(result, ensure_ascii=False)
+            await aio_redis_set(cache_key, cache_data, expire=300)
+            api_logger.info(f"Cached result for key: {cache_key}")
+        except Exception as cache_error:
+            # 缓存写入失败不影响主流程
+            api_logger.warning(f"Failed to cache result: {str(cache_error)}")
+        
         return success(data=result, msg="查询成功")
+        
     except Exception as e:
         api_logger.error(f"Hot memory tags failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "热门标签查询失败", str(e))
 
 
+@router.delete("/analytics/hot_memory_tags/cache", response_model=ApiResponse)
+async def clear_hot_memory_tags_cache(
+    current_user: User = Depends(get_current_user),
+    ) -> dict:
+    """
+    清除热门标签缓存
+    
+    用于：
+    - 手动刷新数据
+    - 调试和测试
+    - 数据更新后立即生效
+    """
+    workspace_id = current_user.current_workspace_id
+    
+    api_logger.info(f"Clear hot memory tags cache requested for workspace: {workspace_id}")
+    
+    try:
+        from app.aioRedis import aio_redis_delete
+        
+        # 清除所有limit的缓存（常见的limit值）
+        cleared_count = 0
+        for limit in [5, 10, 15, 20, 30, 50]:
+            cache_key = f"hot_memory_tags:{workspace_id}:{limit}"
+            result = await aio_redis_delete(cache_key)
+            if result:
+                cleared_count += 1
+                api_logger.info(f"Cleared cache for key: {cache_key}")
+        
+        return success(
+            data={"cleared_count": cleared_count}, 
+            msg=f"成功清除 {cleared_count} 个缓存"
+        )
+        
+    except Exception as e:
+        api_logger.error(f"Clear cache failed: {str(e)}")
+        return fail(BizCode.INTERNAL_ERROR, "清除缓存失败", str(e))
+
+
 @router.get("/analytics/recent_activity_stats", response_model=ApiResponse)
 async def get_recent_activity_stats_api(
     current_user: User = Depends(get_current_user),
@@ -458,18 +529,3 @@ async def get_recent_activity_stats_api(
         api_logger.error(f"Recent activity stats failed: {str(e)}")
         return fail(BizCode.INTERNAL_ERROR, "最近活动统计失败", str(e))
 
-
-
-
-@router.get("/self_reflexion")
-async def self_reflexion_endpoint(host_id: uuid.UUID) -> str:
-    """
-    自我反思接口，自动对检索出的信息进行自我反思并返回自我反思结果。
-
-    Args:
-        None
-    Returns:
-        自我反思结果。
-    """
-    return await self_reflexion(host_id)
-

api/app/controllers/memory_working_controller.py

@@ -20,18 +20,18 @@ router = APIRouter(
 )
 
 
-@router.get("/{group_id}/count", response_model=ApiResponse)
+@router.get("/{end_user_id}/count", response_model=ApiResponse)
 def get_memory_count(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
     pass
 
 
-@router.get("/{group_id}/conversations", response_model=ApiResponse)
+@router.get("/{end_user_id}/conversations", response_model=ApiResponse)
 def get_conversations(
-        group_id: uuid.UUID,
+        end_user_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
         db: Session = Depends(get_db)
 ):
@@ -39,7 +39,7 @@ def get_conversations(
     Retrieve all conversations for the current user in a specific group.
 
     Args:
-        group_id (UUID): The group identifier.
+        end_user_id (UUID): The group identifier.
         current_user (User, optional): The authenticated user.
         db (Session, optional): SQLAlchemy session.
 
@@ -53,7 +53,7 @@ def get_conversations(
     """
     conversation_service = ConversationService(db)
     conversations = conversation_service.get_user_conversations(
-        group_id
+        end_user_id
     )
     return success(data=[
         {
@@ -63,7 +63,7 @@ def get_conversations(
     ], msg="get conversations success")
 
 
-@router.get("/{group_id}/messages", response_model=ApiResponse)
+@router.get("/{end_user_id}/messages", response_model=ApiResponse)
 def get_messages(
         conversation_id: uuid.UUID,
         current_user: User = Depends(get_current_user),
@@ -100,7 +100,7 @@ def get_messages(
     return success(data=messages, msg="get conversation history success")
 
 
-@router.get("/{group_id}/detail", response_model=ApiResponse)
+@router.get("/{end_user_id}/detail", response_model=ApiResponse)
 async def get_conversation_detail(
         conversation_id: uuid.UUID,
         current_user: User = Depends(get_current_user),

api/app/controllers/model_controller.py

@@ -3,15 +3,17 @@ from sqlalchemy.orm import Session
 from typing import Optional
 import uuid
 
-
+from app.core.error_codes import BizCode
+from app.core.exceptions import BusinessException
 from app.db import get_db
 from app.dependencies import get_current_user
-from app.models.models_model import ModelProvider, ModelType
+from app.models.models_model import ModelProvider, ModelType, LoadBalanceStrategy
 from app.models.user_model import User
+from app.repositories.model_repository import ModelConfigRepository
 from app.schemas import model_schema
 from app.core.response_utils import success
 from app.schemas.response_schema import ApiResponse, PageData
-from app.services.model_service import ModelConfigService, ModelApiKeyService
+from app.services.model_service import ModelConfigService, ModelApiKeyService, ModelBaseService
 from app.core.logging_config import get_api_logger
 
 # 获取API专用日志器
@@ -24,24 +26,83 @@ router = APIRouter(
 
 @router.get("/type", response_model=ApiResponse)
 def get_model_types():
-    
     return success(msg="获取模型类型成功", data=list(ModelType))
 
 
 @router.get("/provider", response_model=ApiResponse)
 def get_model_providers():
-    return success(msg="获取模型提供商成功", data=list(ModelProvider))
+    providers = [p for p in ModelProvider if p != ModelProvider.COMPOSITE]
+    return success(msg="获取模型提供商成功", data=providers)
+
+@router.get("/strategy", response_model=ApiResponse)
+def get_model_strategies():
+    return success(msg="获取模型策略成功", data=list(LoadBalanceStrategy))
 
 
 @router.get("", response_model=ApiResponse)
 def get_model_list(
-    type: Optional[str] = Query(None, description="模型类型筛选（支持多个，如 ?type=LLM 或 ?type=LLM,EMBEDDING）"),
-    provider: Optional[model_schema.ModelProvider] = Query(None, description="提供商筛选(基于API Key)"),
+        type: Optional[list[str]] = Query(None, description="模型类型筛选（支持多个，如 ?type=LLM 或 ?type=LLM,EMBEDDING）"),
+        provider: Optional[model_schema.ModelProvider] = Query(None, description="提供商筛选(基于API Key)"),
+        is_active: Optional[bool] = Query(None, description="激活状态筛选"),
+        is_public: Optional[bool] = Query(None, description="公开状态筛选"),
+        search: Optional[str] = Query(None, description="搜索关键词"),
+        page: int = Query(1, ge=1, description="页码"),
+        pagesize: int = Query(10, ge=1, le=100, description="每页数量"),
+        db: Session = Depends(get_db),
+        current_user: User = Depends(get_current_user)
+):
+    """
+    获取模型配置列表
+
+    支持多个 type 参数：
+    - 单个：?type=LLM
+    - 多个（逗号分隔）：?type=LLM,EMBEDDING
+    - 多个（重复参数）：?type=LLM&type=EMBEDDING
+    """
+    api_logger.info(
+        f"获取模型配置列表请求: type={type}, provider={provider}, page={page}, pagesize={pagesize}, tenant_id={current_user.tenant_id}")
+
+    try:
+        # 解析 type 参数（支持逗号分隔）
+        type_list = []
+        if type is not None:
+            flat_type = []
+            for item in type:
+                split_items = [t.strip() for t in item.split(',') if t.strip()]
+                flat_type.extend(split_items)
+
+            unique_flat_type = list(dict.fromkeys(flat_type))
+            type_list = [ModelType(t.lower()) for t in unique_flat_type]
+
+        api_logger.error(f"获取模型type_list: {type_list}")
+        query = model_schema.ModelConfigQuery(
+            type=type_list,
+            provider=provider,
+            is_active=is_active,
+            is_public=is_public,
+            search=search,
+            page=page,
+            pagesize=pagesize
+        )
+
+        api_logger.debug(f"开始获取模型配置列表: {query.dict()}")
+        result_orm = ModelConfigService.get_model_list(db=db, query=query, tenant_id=current_user.tenant_id)
+        result = PageData.model_validate(result_orm)
+        api_logger.info(f"模型配置列表获取成功: 总数={result.page.total}, 当前页={len(result.items)}")
+        return success(data=result, msg="模型配置列表获取成功")
+    except Exception as e:
+        api_logger.error(f"获取模型配置列表失败: {str(e)}")
+        raise
+
+
+@router.get("/new", response_model=ApiResponse)
+def get_model_list_new(
+    type: Optional[list[str]] = Query(None, description="模型类型筛选（支持多个，如 ?type=LLM 或 ?type=LLM,EMBEDDING）"),
+    provider: Optional[model_schema.ModelProvider] = Query(None, description="提供商筛选(基于ModelConfig)"),
     is_active: Optional[bool] = Query(None, description="激活状态筛选"),
     is_public: Optional[bool] = Query(None, description="公开状态筛选"),
     search: Optional[str] = Query(None, description="搜索关键词"),
-    page: int = Query(1, ge=1, description="页码"),
-    pagesize: int = Query(10, ge=1, le=100, description="每页数量"),
+    is_composite: Optional[bool] = Query(None, description="组合模型筛选"),
     db: Session = Depends(get_db),
     current_user: User = Depends(get_current_user)
 ):
@@ -53,36 +114,127 @@ def get_model_list(
     - 多个（逗号分隔）：?type=LLM,EMBEDDING
     - 多个（重复参数）：?type=LLM&type=EMBEDDING
     """
-    api_logger.info(f"获取模型配置列表请求: type={type}, provider={provider}, page={page}, pagesize={pagesize}, tenant_id={current_user.tenant_id}")
+    api_logger.info(f"获取模型配置列表请求: type={type}, provider={provider}, tenant_id={current_user.tenant_id}")
     
     try:
         # 解析 type 参数（支持逗号分隔）
-        type_list = None
-        if type:
-            type_values = [t.strip() for t in type.split(',')]
-            type_list = [model_schema.ModelType(t.lower()) for t in type_values if t]
+        type_list = []
+        if type is not None:
+            flat_type = []
+            for item in type:
+                split_items = [t.strip() for t in item.split(',') if t.strip()]
+                flat_type.extend(split_items)
+
+            unique_flat_type = list(dict.fromkeys(flat_type))
+            type_list = [ModelType(t.lower()) for t in unique_flat_type]
         
-        api_logger.error(f"获取模型type_list: {type_list}")
-        query = model_schema.ModelConfigQuery(
+        api_logger.info(f"获取模型type_list: {type_list}")
+        query = model_schema.ModelConfigQueryNew(
             type=type_list,
             provider=provider,
             is_active=is_active,
             is_public=is_public,
-            search=search,
-            page=page,
-            pagesize=pagesize
+            is_composite=is_composite,
+            search=search
         )
         
-        api_logger.debug(f"开始获取模型配置列表: {query.dict()}")
-        result_orm = ModelConfigService.get_model_list(db=db, query=query, tenant_id=current_user.tenant_id)
-        result = PageData.model_validate(result_orm)
-        api_logger.info(f"模型配置列表获取成功: 总数={result.page.total}, 当前页={len(result.items)}")
+        api_logger.debug(f"开始获取模型配置列表: {query.model_dump()}")
+        result = ModelConfigService.get_model_list_new(db=db, query=query, tenant_id=current_user.tenant_id)
+        api_logger.info(f"模型配置列表获取成功: 分组数={len(result)}, 总模型数={sum(len(item['models']) for item in result)}")
         return success(data=result, msg="模型配置列表获取成功")
     except Exception as e:
         api_logger.error(f"获取模型配置列表失败: {str(e)}")
         raise
 
 
+@router.get("/model_plaza", response_model=ApiResponse)
+def get_model_plaza_list(
+    type: Optional[ModelType] = Query(None, description="模型类型"),
+    provider: Optional[ModelProvider] = Query(None, description="供应商"),
+    is_official: Optional[bool] = Query(None, description="是否官方模型"),
+    is_deprecated: Optional[bool] = Query(None, description="是否弃用"),
+    search: Optional[str] = Query(None, description="搜索关键词"),
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """模型广场查询接口（按供应商分组）"""
+    
+    query = model_schema.ModelBaseQuery(
+        type=type,
+        provider=provider,
+        is_official=is_official,
+        is_deprecated=is_deprecated,
+        search=search
+    )
+    result = ModelBaseService.get_model_base_list(db=db, query=query, tenant_id=current_user.tenant_id)
+    return success(data=result, msg="模型广场列表获取成功")
+
+
+@router.get("/model_plaza/{model_base_id}", response_model=ApiResponse)
+def get_model_base_by_id(
+    model_base_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """获取基础模型详情"""
+    
+    result = ModelBaseService.get_model_base_by_id(db=db, model_base_id=model_base_id)
+    return success(data=model_schema.ModelBase.model_validate(result), msg="基础模型获取成功")
+
+
+@router.post("/model_plaza", response_model=ApiResponse)
+def create_model_base(
+    data: model_schema.ModelBaseCreate,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """创建基础模型"""
+    
+    result = ModelBaseService.create_model_base(db=db, data=data)
+    return success(data=model_schema.ModelBase.model_validate(result), msg="基础模型创建成功")
+
+
+@router.put("/model_plaza/{model_base_id}", response_model=ApiResponse)
+def update_model_base(
+    model_base_id: uuid.UUID,
+    data: model_schema.ModelBaseUpdate,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """更新基础模型"""
+    
+    # 不允许更改type类型
+    if data.type is not None or data.provider is not None:
+        raise BusinessException("不允许更改模型类型和供应商", BizCode.INVALID_PARAMETER)
+    
+    result = ModelBaseService.update_model_base(db=db, model_base_id=model_base_id, data=data)
+    return success(data=model_schema.ModelBase.model_validate(result), msg="基础模型更新成功")
+
+
+@router.delete("/model_plaza/{model_base_id}", response_model=ApiResponse)
+def delete_model_base(
+    model_base_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """删除基础模型"""
+    
+    ModelBaseService.delete_model_base(db=db, model_base_id=model_base_id)
+    return success(msg="基础模型删除成功")
+
+
+@router.post("/model_plaza/{model_base_id}/add", response_model=ApiResponse)
+def add_model_from_plaza(
+    model_base_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """从模型广场添加模型到模型列表"""
+    
+    result = ModelBaseService.add_model_from_plaza(db=db, model_base_id=model_base_id, tenant_id=current_user.tenant_id)
+    return success(data=model_schema.ModelConfig.model_validate(result), msg="模型添加成功")
+
+
 @router.get("/{model_id}", response_model=ApiResponse)
 def get_model_by_id(
     model_id: uuid.UUID,
@@ -138,6 +290,73 @@ async def create_model(
         raise
 
 
+@router.post("/composite", response_model=ApiResponse)
+async def create_composite_model(
+    model_data: model_schema.CompositeModelCreate,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """
+    创建组合模型
+    
+    - 绑定一个或多个现有的 API Key
+    - 所有 API Key 必须来自非组合模型
+    - 所有 API Key 关联的模型类型必须与组合模型类型一致
+    """
+    api_logger.info(f"创建组合模型请求: {model_data.name}, 用户: {current_user.username}, tenant_id={current_user.tenant_id}")
+    
+    try:
+        result_orm = await ModelConfigService.create_composite_model(db=db, model_data=model_data, tenant_id=current_user.tenant_id)
+        api_logger.info(f"组合模型创建成功: {result_orm.name} (ID: {result_orm.id})")
+        
+        result = model_schema.ModelConfig.model_validate(result_orm)
+        return success(data=result, msg="组合模型创建成功")
+    except Exception as e:
+        api_logger.error(f"创建组合模型失败: {model_data.name} - {str(e)}")
+        raise
+
+
+@router.put("/composite/{model_id}", response_model=ApiResponse)
+async def update_composite_model(
+    model_id: uuid.UUID,
+    model_data: model_schema.CompositeModelCreate,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """更新组合模型"""
+    api_logger.info(f"更新组合模型请求: model_id={model_id}, 用户: {current_user.username}")
+    
+    try:
+        if model_data.type is not None:
+            raise BusinessException("不允许更改模型类型和供应商", BizCode.INVALID_PARAMETER)
+        result_orm = await ModelConfigService.update_composite_model(db=db, model_id=model_id, model_data=model_data, tenant_id=current_user.tenant_id)
+        api_logger.info(f"组合模型更新成功: {result_orm.name} (ID: {model_id})")
+        
+        result = model_schema.ModelConfig.model_validate(result_orm)
+        return success(data=result, msg="组合模型更新成功")
+    except Exception as e:
+        api_logger.error(f"更新组合模型失败: model_id={model_id} - {str(e)}")
+        raise
+
+
+@router.delete("/composite/{model_id}", response_model=ApiResponse)
+def delete_composite_model(
+    model_id: uuid.UUID,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """删除组合模型"""
+    api_logger.info(f"删除组合模型请求: model_id={model_id}, 用户: {current_user.username}")
+    
+    try:
+        ModelConfigService.delete_model(db=db, model_id=model_id, tenant_id=current_user.tenant_id)
+        api_logger.info(f"组合模型删除成功: model_id={model_id}")
+        return success(msg="组合模型删除成功")
+    except Exception as e:
+        api_logger.error(f"删除组合模型失败: model_id={model_id} - {str(e)}")
+        raise
+
+
 @router.put("/{model_id}", response_model=ApiResponse)
 def update_model(
     model_id: uuid.UUID,
@@ -214,6 +433,53 @@ def get_model_api_keys(
         raise
 
 
+@router.post("/provider/apikeys", response_model=ApiResponse)
+async def create_model_api_key_by_provider(
+        api_key_data: model_schema.ModelApiKeyCreateByProvider,
+        db: Session = Depends(get_db),
+        current_user: User = Depends(get_current_user)
+):
+    """
+    根据供应商为所有匹配的模型创建API Key
+    """
+    api_logger.info(f"创建API Key请求: provider={api_key_data.provider}, 用户: {current_user.username}")
+
+    try:
+        # 根据tenant_id和provider筛选model_config_id列表
+        model_config_ids = api_key_data.model_config_ids
+        if not model_config_ids:
+            model_config_ids = ModelConfigRepository.get_model_config_ids_by_provider(
+                db=db,
+                tenant_id=current_user.tenant_id,
+                provider=api_key_data.provider
+            )
+        
+        if not model_config_ids:
+            raise BusinessException(f"未找到供应商 {api_key_data.provider} 的模型配置", BizCode.MODEL_NOT_FOUND)
+        
+        # 构造schema并调用service
+        create_data = model_schema.ModelApiKeyCreateByProvider(
+            provider=api_key_data.provider,
+            api_key=api_key_data.api_key,
+            api_base=api_key_data.api_base,
+            description=api_key_data.description,
+            config=api_key_data.config,
+            is_active=api_key_data.is_active,
+            priority=api_key_data.priority,
+            model_config_ids=model_config_ids
+        )
+        created_keys, failed_models = await ModelApiKeyService.create_api_key_by_provider(db=db, data=create_data)
+        
+        api_logger.info(f"API Key创建成功: 关联{len(created_keys)}个模型")
+        # result_list = [model_schema.ModelApiKey.model_validate(key) for key in created_keys]
+        result = "API Key已存在" if len(created_keys) == 0 and len(failed_models) == 0 else \
+            f"成功为 {len(created_keys)} 个模型创建API Key, 失败模型列表{failed_models}"
+        return success(data=result, msg=f"成功为 {len(created_keys)} 个模型创建API Key")
+    except Exception as e:
+        api_logger.error(f"创建API Key失败: {str(e)}")
+        raise
+
+
 @router.post("/{model_id}/apikeys", response_model=ApiResponse, status_code=status.HTTP_201_CREATED)
 async def create_model_api_key(
     model_id: uuid.UUID,
@@ -228,11 +494,12 @@ async def create_model_api_key(
     
     try:
         # 设置模型配置ID
-        api_key_data.model_config_id = model_id
+        api_key_data.model_config_ids = [model_id]
         
         api_logger.debug(f"开始创建模型API Key: {api_key_data.model_name}")
-        result = await ModelApiKeyService.create_api_key(db=db, api_key_data=api_key_data)
-        api_logger.info(f"模型API Key创建成功: {result.model_name} (ID: {result.id})")
+        result_orm = await ModelApiKeyService.create_api_key(db=db, api_key_data=api_key_data)
+        api_logger.info(f"模型API Key创建成功: {result_orm.model_name} (ID: {result_orm.id})")
+        result = model_schema.ModelApiKey.model_validate(result_orm)
         return success(data=result, msg="模型API Key创建成功")
     except Exception as e:
         api_logger.error(f"创建模型API Key失败: {api_key_data.model_name} - {str(e)}")
@@ -334,5 +601,3 @@ async def validate_model_config(
     return success(data=model_schema.ModelValidateResponse(**result), msg="验证完成")
 
 
-
-

api/app/controllers/ontology_secondary_routes.py

@@ -0,0 +1,611 @@
+# -*- coding: utf-8 -*-
+"""本体场景和类型路由（续）
+
+由于主Controller文件较大，将剩余路由放在此文件中。
+"""
+
+from uuid import UUID
+from typing import Optional
+
+from fastapi import Depends
+from sqlalchemy.orm import Session
+
+from app.core.error_codes import BizCode
+from app.core.logging_config import get_api_logger
+from app.core.response_utils import fail, success
+from app.db import get_db
+from app.dependencies import get_current_user
+from app.models.user_model import User
+from app.schemas.ontology_schemas import (
+    SceneResponse,
+    SceneListResponse,
+    PaginationInfo,
+    ClassCreateRequest,
+    ClassUpdateRequest,
+    ClassResponse,
+    ClassListResponse,
+    ClassBatchCreateResponse,
+)
+from app.schemas.response_schema import ApiResponse
+from app.services.ontology_service import OntologyService
+from app.core.memory.llm_tools.openai_client import OpenAIClient
+from app.core.models.base import RedBearModelConfig
+
+
+api_logger = get_api_logger()
+
+
+def _get_dummy_ontology_service(db: Session) -> OntologyService:
+    """获取OntologyService实例（不需要LLM）
+    
+    场景和类型管理不需要LLM，创建一个dummy配置。
+    """
+    dummy_config = RedBearModelConfig(
+        model_name="dummy",
+        provider="openai",
+        api_key="dummy",
+        base_url="https://api.openai.com/v1"
+    )
+    llm_client = OpenAIClient(model_config=dummy_config)
+    return OntologyService(llm_client=llm_client, db=db)
+
+
+# 这些函数将被导入到主Controller中
+
+async def scenes_handler(
+    workspace_id: Optional[str] = None,
+    scene_name: Optional[str] = None,
+    page: Optional[int] = None,
+    page_size: Optional[int] = None,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """获取场景列表（支持模糊搜索和全量查询，全量查询支持分页）
+    
+    当提供 scene_name 参数时，进行模糊搜索（不分页）；
+    当不提供 scene_name 参数时，返回所有场景（支持分页）。
+    
+    Args:
+        workspace_id: 工作空间ID（可选，默认当前用户工作空间）
+        scene_name: 场景名称关键词（可选，支持模糊匹配）
+        page: 页码（可选，从1开始，仅在全量查询时有效）
+        page_size: 每页数量（可选，仅在全量查询时有效）
+        db: 数据库会话
+        current_user: 当前用户
+    """
+    operation = "search" if scene_name else "list"
+    api_logger.info(
+        f"Scene {operation} requested by user {current_user.id}, "
+        f"workspace_id={workspace_id}, keyword={scene_name}, page={page}, page_size={page_size}"
+    )
+    
+    try:
+        # 确定工作空间ID
+        if workspace_id:
+            try:
+                ws_uuid = UUID(workspace_id)
+            except ValueError:
+                api_logger.warning(f"Invalid workspace_id format: {workspace_id}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "无效的工作空间ID格式")
+        else:
+            ws_uuid = current_user.current_workspace_id
+            if not ws_uuid:
+                api_logger.warning(f"User {current_user.id} has no current workspace")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "当前用户没有工作空间")
+        
+        # 创建Service
+        service = _get_dummy_ontology_service(db)
+        
+        # 根据是否提供 scene_name 决定查询方式
+        if scene_name and scene_name.strip():
+            # 验证分页参数（模糊搜索也支持分页）
+            if page is not None and page < 1:
+                api_logger.warning(f"Invalid page number: {page}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "页码必须大于0")
+            
+            if page_size is not None and page_size < 1:
+                api_logger.warning(f"Invalid page_size: {page_size}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "每页数量必须大于0")
+            
+            # 如果只提供了page或page_size中的一个，返回错误
+            if (page is not None and page_size is None) or (page is None and page_size is not None):
+                api_logger.warning(f"Incomplete pagination params: page={page}, page_size={page_size}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "分页参数page和pagesize必须同时提供")
+            
+            # 模糊搜索场景（支持分页）
+            scenes = service.search_scenes_by_name(scene_name.strip(), ws_uuid)
+            total = len(scenes)
+            
+            # 如果提供了分页参数，进行分页处理
+            if page is not None and page_size is not None:
+                start_idx = (page - 1) * page_size
+                end_idx = start_idx + page_size
+                scenes = scenes[start_idx:end_idx]
+            
+            # 构建响应
+            items = []
+            for scene in scenes:
+                # 获取前3个class_name作为entity_type
+                entity_type = [cls.class_name for cls in scene.classes[:3]] if scene.classes else None
+                # 动态计算 type_num
+                type_num = len(scene.classes) if scene.classes else 0
+                
+                items.append(SceneResponse(
+                    scene_id=scene.scene_id,
+                    scene_name=scene.scene_name,
+                    scene_description=scene.scene_description,
+                    type_num=type_num,
+                    entity_type=entity_type,
+                    workspace_id=scene.workspace_id,
+                    created_at=scene.created_at,
+                    updated_at=scene.updated_at,
+                    classes_count=type_num
+                ))
+            
+            # 构建响应（包含分页信息）
+            if page is not None and page_size is not None:
+                # 计算是否有下一页
+                hasnext = (page * page_size) < total
+                
+                pagination_info = PaginationInfo(
+                    page=page,
+                    pagesize=page_size,
+                    total=total,
+                    hasnext=hasnext
+                )
+                response = SceneListResponse(items=items, page=pagination_info)
+            else:
+                response = SceneListResponse(items=items)
+            
+            api_logger.info(
+                f"Scene search completed: found {len(items)} scenes matching '{scene_name}' "
+                f"in workspace {ws_uuid}, total={total}"
+            )
+        else:
+            # 获取所有场景（支持分页）
+            # 验证分页参数
+            if page is not None and page < 1:
+                api_logger.warning(f"Invalid page number: {page}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "页码必须大于0")
+            
+            if page_size is not None and page_size < 1:
+                api_logger.warning(f"Invalid page_size: {page_size}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "每页数量必须大于0")
+            
+            # 如果只提供了page或page_size中的一个，返回错误
+            if (page is not None and page_size is None) or (page is None and page_size is not None):
+                api_logger.warning(f"Incomplete pagination params: page={page}, page_size={page_size}")
+                return fail(BizCode.BAD_REQUEST, "请求参数无效", "分页参数page和pagesize必须同时提供")
+            
+            scenes, total = service.list_scenes(ws_uuid, page, page_size)
+            
+            # 构建响应
+            items = []
+            for scene in scenes:
+                # 获取前3个class_name作为entity_type
+                entity_type = [cls.class_name for cls in scene.classes[:3]] if scene.classes else None
+                # 动态计算 type_num
+                type_num = len(scene.classes) if scene.classes else 0
+                
+                items.append(SceneResponse(
+                    scene_id=scene.scene_id,
+                    scene_name=scene.scene_name,
+                    scene_description=scene.scene_description,
+                    type_num=type_num,
+                    entity_type=entity_type,
+                    workspace_id=scene.workspace_id,
+                    created_at=scene.created_at,
+                    updated_at=scene.updated_at,
+                    classes_count=type_num
+                ))
+            
+            # 构建响应（包含分页信息）
+            if page is not None and page_size is not None:
+                # 计算是否有下一页
+                hasnext = (page * page_size) < total
+                
+                pagination_info = PaginationInfo(
+                    page=page,
+                    pagesize=page_size,
+                    total=total,
+                    hasnext=hasnext
+                )
+                response = SceneListResponse(items=items, page=pagination_info)
+            else:
+                response = SceneListResponse(items=items)
+            
+            api_logger.info(f"Scene list retrieved successfully, count={len(items)}, total={total}")
+        
+        return success(data=response.model_dump(mode='json'), msg="查询成功")
+        
+    except ValueError as e:
+        api_logger.warning(f"Validation error in scene {operation}: {str(e)}")
+        return fail(BizCode.BAD_REQUEST, "请求参数无效", str(e))
+        
+    except RuntimeError as e:
+        api_logger.error(f"Runtime error in scene {operation}: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "查询失败", str(e))
+        
+    except Exception as e:
+        api_logger.error(f"Unexpected error in scene {operation}: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "查询失败", str(e))
+
+
+# ==================== 本体类型管理接口 ====================
+
+async def create_class_handler(
+    request: ClassCreateRequest,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """创建本体类型（统一使用列表形式，支持单个或批量）"""
+    
+    # 根据列表长度判断是单个还是批量
+    count = len(request.classes)
+    mode = "single" if count == 1 else "batch"
+    
+    api_logger.info(
+        f"Class creation ({mode}) requested by user {current_user.id}, "
+        f"scene_id={request.scene_id}, count={count}"
+    )
+    
+    try:
+        # 获取当前工作空间ID
+        workspace_id = current_user.current_workspace_id
+        if not workspace_id:
+            api_logger.warning(f"User {current_user.id} has no current workspace")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "当前用户没有工作空间")
+        
+        # 创建Service
+        service = _get_dummy_ontology_service(db)
+        
+        # 准备类型数据
+        classes_data = [
+            {
+                "class_name": item.class_name,
+                "class_description": item.class_description
+            }
+            for item in request.classes
+        ]
+        
+        if count == 1:
+            # 单个创建
+            class_data = classes_data[0]
+            ontology_class = service.create_class(
+                scene_id=request.scene_id,
+                class_name=class_data["class_name"],
+                class_description=class_data["class_description"],
+                workspace_id=workspace_id
+            )
+            
+            # 构建单个响应
+            response = ClassResponse(
+                class_id=ontology_class.class_id,
+                class_name=ontology_class.class_name,
+                class_description=ontology_class.class_description,
+                scene_id=ontology_class.scene_id,
+                created_at=ontology_class.created_at,
+                updated_at=ontology_class.updated_at
+            )
+            
+            api_logger.info(f"Class created successfully: {ontology_class.class_id}")
+            
+            return success(data=response.model_dump(mode='json'), msg="类型创建成功")
+            
+        else:
+            # 批量创建
+            created_classes, errors = service.create_classes_batch(
+                scene_id=request.scene_id,
+                classes=classes_data,
+                workspace_id=workspace_id
+            )
+            
+            # 构建批量响应
+            items = []
+            for ontology_class in created_classes:
+                items.append(ClassResponse(
+                    class_id=ontology_class.class_id,
+                    class_name=ontology_class.class_name,
+                    class_description=ontology_class.class_description,
+                    scene_id=ontology_class.scene_id,
+                    created_at=ontology_class.created_at,
+                    updated_at=ontology_class.updated_at
+                ))
+            
+            response = ClassBatchCreateResponse(
+                total=len(classes_data),
+                success_count=len(created_classes),
+                failed_count=len(errors),
+                items=items,
+                errors=errors if errors else None
+            )
+            
+            api_logger.info(
+                f"Batch class creation completed: "
+                f"success={len(created_classes)}, failed={len(errors)}"
+            )
+            
+            return success(data=response.model_dump(mode='json'), msg="批量创建完成")
+        
+    except ValueError as e:
+        api_logger.warning(f"Validation error in class creation: {str(e)}")
+        return fail(BizCode.BAD_REQUEST, "请求参数无效", str(e))
+        
+    except RuntimeError as e:
+        api_logger.error(f"Runtime error in class creation: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "类型创建失败", str(e))
+        
+    except Exception as e:
+        api_logger.error(f"Unexpected error in class creation: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "类型创建失败", str(e))
+
+
+async def update_class_handler(
+    class_id: str,
+    request: ClassUpdateRequest,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """更新本体类型"""
+    api_logger.info(
+        f"Class update requested by user {current_user.id}, "
+        f"class_id={class_id}"
+    )
+    
+    try:
+        # 验证UUID格式
+        try:
+            class_uuid = UUID(class_id)
+        except ValueError:
+            api_logger.warning(f"Invalid class_id format: {class_id}")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "无效的类型ID格式")
+        
+        # 获取当前工作空间ID
+        workspace_id = current_user.current_workspace_id
+        if not workspace_id:
+            api_logger.warning(f"User {current_user.id} has no current workspace")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "当前用户没有工作空间")
+        
+        # 创建Service
+        service = _get_dummy_ontology_service(db)
+        
+        # 更新类型
+        ontology_class = service.update_class(
+            class_id=class_uuid,
+            class_name=request.class_name,
+            class_description=request.class_description,
+            workspace_id=workspace_id
+        )
+        
+        # 构建响应
+        response = ClassResponse(
+            class_id=ontology_class.class_id,
+            class_name=ontology_class.class_name,
+            class_description=ontology_class.class_description,
+            scene_id=ontology_class.scene_id,
+            created_at=ontology_class.created_at,
+            updated_at=ontology_class.updated_at
+        )
+        
+        api_logger.info(f"Class updated successfully: {class_id}")
+        
+        return success(data=response.model_dump(mode='json'), msg="类型更新成功")
+        
+    except ValueError as e:
+        api_logger.warning(f"Validation error in class update: {str(e)}")
+        return fail(BizCode.BAD_REQUEST, "请求参数无效", str(e))
+        
+    except RuntimeError as e:
+        api_logger.error(f"Runtime error in class update: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "类型更新失败", str(e))
+        
+    except Exception as e:
+        api_logger.error(f"Unexpected error in class update: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "类型更新失败", str(e))
+
+
+async def delete_class_handler(
+    class_id: str,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """删除本体类型"""
+    api_logger.info(
+        f"Class deletion requested by user {current_user.id}, "
+        f"class_id={class_id}"
+    )
+    
+    try:
+        # 验证UUID格式
+        try:
+            class_uuid = UUID(class_id)
+        except ValueError:
+            api_logger.warning(f"Invalid class_id format: {class_id}")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "无效的类型ID格式")
+        
+        # 获取当前工作空间ID
+        workspace_id = current_user.current_workspace_id
+        if not workspace_id:
+            api_logger.warning(f"User {current_user.id} has no current workspace")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "当前用户没有工作空间")
+        
+        # 创建Service
+        service = _get_dummy_ontology_service(db)
+        
+        # 删除类型
+        success_flag = service.delete_class(
+            class_id=class_uuid,
+            workspace_id=workspace_id
+        )
+        
+        api_logger.info(f"Class deleted successfully: {class_id}")
+        
+        return success(data={"deleted": success_flag}, msg="类型删除成功")
+        
+    except ValueError as e:
+        api_logger.warning(f"Validation error in class deletion: {str(e)}")
+        return fail(BizCode.BAD_REQUEST, "请求参数无效", str(e))
+        
+    except RuntimeError as e:
+        api_logger.error(f"Runtime error in class deletion: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "类型删除失败", str(e))
+        
+    except Exception as e:
+        api_logger.error(f"Unexpected error in class deletion: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "类型删除失败", str(e))
+
+
+async def get_class_handler(
+    class_id: str,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """获取单个本体类型"""
+    api_logger.info(
+        f"Get class requested by user {current_user.id}, "
+        f"class_id={class_id}"
+    )
+    
+    try:
+        # 验证UUID格式
+        try:
+            class_uuid = UUID(class_id)
+        except ValueError:
+            api_logger.warning(f"Invalid class_id format: {class_id}")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "无效的类型ID格式")
+        
+        # 获取当前工作空间ID
+        workspace_id = current_user.current_workspace_id
+        if not workspace_id:
+            api_logger.warning(f"User {current_user.id} has no current workspace")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "当前用户没有工作空间")
+        
+        # 创建Service
+        service = _get_dummy_ontology_service(db)
+        
+        # 获取类型（会抛出ValueError如果不存在）
+        ontology_class = service.get_class_by_id(class_uuid, workspace_id)
+        
+        # 构建响应
+        response = ClassResponse(
+            class_id=ontology_class.class_id,
+            class_name=ontology_class.class_name,
+            class_description=ontology_class.class_description,
+            scene_id=ontology_class.scene_id,
+            created_at=ontology_class.created_at,
+            updated_at=ontology_class.updated_at
+        )
+        
+        api_logger.info(f"Class retrieved successfully: {class_id}")
+        
+        return success(data=response.model_dump(mode='json'), msg="查询成功")
+        
+    except ValueError as e:
+        # 类型不存在或无权限访问
+        api_logger.warning(f"Validation error in get class: {str(e)}")
+        return fail(BizCode.NOT_FOUND, "请求参数无效", str(e))
+        
+    except RuntimeError as e:
+        api_logger.error(f"Runtime error in get class: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "查询失败", str(e))
+        
+    except Exception as e:
+        api_logger.error(f"Unexpected error in get class: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "查询失败", str(e))
+
+
+async def classes_handler(
+    scene_id: str,
+    class_name: Optional[str] = None,
+    db: Session = Depends(get_db),
+    current_user: User = Depends(get_current_user)
+):
+    """获取类型列表（支持模糊搜索和全量查询）
+    
+    当提供 class_name 参数时，进行模糊搜索；
+    当不提供 class_name 参数时，返回场景下的所有类型。
+    
+    Args:
+        scene_id: 场景ID（必填）
+        class_name: 类型名称关键词（可选，支持模糊匹配）
+        db: 数据库会话
+        current_user: 当前用户
+    """
+    operation = "search" if class_name else "list"
+    api_logger.info(
+        f"Class {operation} requested by user {current_user.id}, "
+        f"keyword={class_name}, scene_id={scene_id}"
+    )
+    
+    try:
+        # 验证UUID格式
+        try:
+            scene_uuid = UUID(scene_id)
+        except ValueError:
+            api_logger.warning(f"Invalid scene_id format: {scene_id}")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "无效的场景ID格式")
+        
+        # 获取当前工作空间ID
+        workspace_id = current_user.current_workspace_id
+        if not workspace_id:
+            api_logger.warning(f"User {current_user.id} has no current workspace")
+            return fail(BizCode.BAD_REQUEST, "请求参数无效", "当前用户没有工作空间")
+        
+        # 创建Service
+        service = _get_dummy_ontology_service(db)
+        
+        # 获取场景信息
+        scene = service.get_scene_by_id(scene_uuid, workspace_id)
+        if not scene:
+            api_logger.warning(f"Scene not found: {scene_id}")
+            return fail(BizCode.NOT_FOUND, "场景不存在", f"未找到ID为 {scene_id} 的场景")
+        
+        # 根据是否提供 class_name 决定查询方式
+        if class_name and class_name.strip():
+            # 模糊搜索类型
+            classes = service.search_classes_by_name(class_name.strip(), scene_uuid, workspace_id)
+        else:
+            # 获取所有类型
+            classes = service.list_classes_by_scene(scene_uuid, workspace_id)
+        
+        # 构建响应
+        items = []
+        for ontology_class in classes:
+            items.append(ClassResponse(
+                class_id=ontology_class.class_id,
+                class_name=ontology_class.class_name,
+                class_description=ontology_class.class_description,
+                scene_id=ontology_class.scene_id,
+                created_at=ontology_class.created_at,
+                updated_at=ontology_class.updated_at
+            ))
+        
+        response = ClassListResponse(
+            total=len(items),
+            scene_id=scene_uuid,
+            scene_name=scene.scene_name,
+            scene_description=scene.scene_description,
+            items=items
+        )
+        
+        if class_name:
+            api_logger.info(
+                f"Class search completed: found {len(items)} classes matching '{class_name}' "
+                f"in scene {scene_id}"
+            )
+        else:
+            api_logger.info(f"Class list retrieved successfully, count={len(items)}")
+        
+        return success(data=response.model_dump(mode='json'), msg="查询成功")
+        
+    except ValueError as e:
+        api_logger.warning(f"Validation error in class {operation}: {str(e)}")
+        return fail(BizCode.BAD_REQUEST, "请求参数无效", str(e))
+        
+    except RuntimeError as e:
+        api_logger.error(f"Runtime error in class {operation}: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "查询失败", str(e))
+        
+    except Exception as e:
+        api_logger.error(f"Unexpected error in class {operation}: {str(e)}", exc_info=True)
+        return fail(BizCode.INTERNAL_ERROR, "查询失败", str(e))

api/app/controllers/prompt_optimizer_controller.py

@@ -1,5 +1,5 @@
-import uuid
 import json
+import uuid
 
 from fastapi import APIRouter, Depends, Path
 from sqlalchemy.orm import Session
@@ -8,9 +8,13 @@ from starlette.responses import StreamingResponse
 from app.core.logging_config import get_api_logger
 from app.core.response_utils import success
 from app.dependencies import get_current_user, get_db
-from app.models.prompt_optimizer_model import RoleType
-from app.schemas.prompt_optimizer_schema import PromptOptMessage, PromptOptModelSet, CreateSessionResponse, \
-    OptimizePromptResponse, SessionHistoryResponse, SessionMessage
+from app.schemas.prompt_optimizer_schema import (
+    PromptOptMessage,
+    CreateSessionResponse,
+    SessionHistoryResponse,
+    SessionMessage,
+    PromptSaveRequest
+)
 from app.schemas.response_schema import ApiResponse
 from app.services.prompt_optimizer_service import PromptOptimizerService
 
@@ -135,3 +139,109 @@ async def get_prompt_opt(
             "X-Accel-Buffering": "no"
         }
     )
+
+
+@router.post(
+    "/releases",
+    summary="Get prompt optimization",
+    response_model=ApiResponse
+)
+def save_prompt(
+        data: PromptSaveRequest,
+        db: Session = Depends(get_db),
+        current_user=Depends(get_current_user),
+):
+    """
+       Save a prompt release for the current tenant.
+
+       Args:
+           data (PromptSaveRequest): Request body containing session_id, title, and prompt.
+           db (Session): SQLAlchemy database session, injected via dependency.
+           current_user: Currently authenticated user object, injected via dependency.
+
+       Returns:
+           ApiResponse: Standard API response containing the saved prompt release info:
+               - id: UUID of the prompt release
+               - session_id: associated session
+               - title: prompt title
+               - prompt: prompt content
+               - created_at: timestamp of creation
+
+       Raises:
+           Any database or service exceptions are propagated to the global exception handler.
+       """
+    service = PromptOptimizerService(db)
+    prompt_info = service.save_prompt(
+        tenant_id=current_user.tenant_id,
+        session_id=data.session_id,
+        title=data.title,
+        prompt=data.prompt
+    )
+    return success(data=prompt_info)
+
+
+@router.delete(
+    "/releases/{prompt_id}",
+    summary="Delete prompt (soft delete)",
+    response_model=ApiResponse
+)
+def delete_prompt(
+        prompt_id: uuid.UUID = Path(..., description="Prompt ID"),
+        db: Session = Depends(get_db),
+        current_user=Depends(get_current_user),
+):
+    """
+    Soft delete a prompt release.
+
+    Args:
+        prompt_id
+        db (Session): Database session
+        current_user: Current logged-in user
+
+    Returns:
+        ApiResponse: Success message confirming deletion
+    """
+    service = PromptOptimizerService(db)
+    service.delete_prompt(
+        tenant_id=current_user.tenant_id,
+        prompt_id=prompt_id
+    )
+    return success(msg="Prompt deleted successfully")
+
+
+@router.get(
+    "/releases/list",
+    summary="Get paginated list of released prompts with optional filter",
+    response_model=ApiResponse
+)
+def get_release_list(
+        page: int = 1,
+        page_size: int = 20,
+        keyword: str | None = None,
+        db: Session = Depends(get_db),
+        current_user=Depends(get_current_user),
+):
+    """
+    Retrieve paginated list of released prompts for the current tenant.
+    Optionally filter by keyword in title.
+
+    Args:
+        page (int): Page number (starting from 1)
+        page_size (int): Number of items per page (max 100)
+        keyword (str | None): Optional keyword to filter prompt titles
+        db (Session): Database session
+        current_user: Current logged-in user
+
+    Returns:
+        ApiResponse: Contains paginated list of prompt releases with metadata
+    """
+    service = PromptOptimizerService(db)
+    result = service.get_release_list(
+        tenant_id=current_user.tenant_id,
+        page=max(1, page),
+        page_size=min(max(1, page_size), 100),
+        filter_keyword=keyword
+    )
+    return success(data=result)
+
+

api/app/controllers/public_share_controller.py

@@ -8,9 +8,10 @@ from sqlalchemy.orm import Session
 
 from app.core.logging_config import get_business_logger
 from app.core.response_utils import success
-from app.db import get_db
+from app.db import get_db, get_db_read
 from app.dependencies import get_share_user_id, ShareTokenData
 from app.repositories import knowledge_repository
+from app.repositories.workflow_repository import WorkflowConfigRepository
 from app.schemas import release_share_schema, conversation_schema
 from app.schemas.response_schema import PageData, PageMeta
 from app.services import workspace_service
@@ -19,7 +20,8 @@ from app.services.conversation_service import ConversationService
 from app.services.release_share_service import ReleaseShareService
 from app.services.shared_chat_service import SharedChatService
 from app.services.app_chat_service import AppChatService, get_app_chat_service
-from app.utils.app_config_utils import dict_to_multi_agent_config, workflow_config_4_app_release, agent_config_4_app_release, multi_agent_config_4_app_release
+from app.utils.app_config_utils import dict_to_multi_agent_config, workflow_config_4_app_release, \
+    agent_config_4_app_release, multi_agent_config_4_app_release
 
 router = APIRouter(prefix="/public/share", tags=["Public Share"])
 logger = get_business_logger()
@@ -65,10 +67,10 @@ def get_or_generate_user_id(payload_user_id: str, request: Request) -> str:
     summary="获取访问 token"
 )
 def get_access_token(
-    share_token: str,
-    payload: release_share_schema.TokenRequest,
-    request: Request,
-    db: Session = Depends(get_db),
+        share_token: str,
+        payload: release_share_schema.TokenRequest,
+        request: Request,
+        db: Session = Depends(get_db),
 ):
     """获取访问 token
 
@@ -113,9 +115,9 @@ def get_access_token(
     response_model=None
 )
 def get_shared_release(
-    password: str = Query(None, description="访问密码（如果需要）"),
-    share_data: ShareTokenData = Depends(get_share_user_id),
-    db: Session = Depends(get_db),
+        password: str = Query(None, description="访问密码（如果需要）"),
+        share_data: ShareTokenData = Depends(get_share_user_id),
+        db: Session = Depends(get_db),
 ):
     """获取公开分享的发布版本信息
 
@@ -137,9 +139,9 @@ def get_shared_release(
     summary="验证访问密码"
 )
 def verify_password(
-    payload: release_share_schema.PasswordVerifyRequest,
-    share_data: ShareTokenData = Depends(get_share_user_id),
-    db: Session = Depends(get_db),
+        payload: release_share_schema.PasswordVerifyRequest,
+        share_data: ShareTokenData = Depends(get_share_user_id),
+        db: Session = Depends(get_db),
 ):
     """验证分享的访问密码
 
@@ -159,11 +161,11 @@ def verify_password(
     summary="获取嵌入代码"
 )
 def get_embed_code(
-    width: str = Query("100%", description="iframe 宽度"),
-    height: str = Query("600px", description="iframe 高度"),
-    request: Request = None,
-    share_data: ShareTokenData = Depends(get_share_user_id),
-    db: Session = Depends(get_db),
+        width: str = Query("100%", description="iframe 宽度"),
+        height: str = Query("600px", description="iframe 高度"),
+        request: Request = None,
+        share_data: ShareTokenData = Depends(get_share_user_id),
+        db: Session = Depends(get_db),
 ):
     """获取嵌入代码
 
@@ -183,7 +185,6 @@ def get_embed_code(
     return success(data=embed_code)
 
 
-
 # ---------- 会话管理接口 ----------
 
 @router.get(
@@ -191,11 +192,11 @@ def get_embed_code(
     summary="获取会话列表"
 )
 def list_conversations(
-    password: str = Query(None, description="访问密码"),
-    page: int = Query(1, ge=1),
-    pagesize: int = Query(20, ge=1, le=100),
-    share_data: ShareTokenData = Depends(get_share_user_id),
-    db: Session = Depends(get_db),
+        password: str = Query(None, description="访问密码"),
+        page: int = Query(1, ge=1),
+        pagesize: int = Query(20, ge=1, le=100),
+        share_data: ShareTokenData = Depends(get_share_user_id),
+        db: Session = Depends(get_db),
 ):
     """获取分享应用的会话列表
 
@@ -209,9 +210,9 @@ def list_conversations(
     from app.repositories.end_user_repository import EndUserRepository
     end_user_repo = EndUserRepository(db)
     new_end_user = end_user_repo.get_or_create_end_user(
-            app_id=share.app_id,
-            other_id=other_id
-        )
+        app_id=share.app_id,
+        other_id=other_id
+    )
     logger.debug(new_end_user.id)
     service = SharedChatService(db)
     conversations, total = service.list_conversations(
@@ -233,10 +234,10 @@ def list_conversations(
     summary="获取会话详情（含消息）"
 )
 def get_conversation(
-    conversation_id: uuid.UUID,
-    password: str = Query(None, description="访问密码"),
-    share_data: ShareTokenData = Depends(get_share_user_id),
-    db: Session = Depends(get_db),
+        conversation_id: uuid.UUID,
+        password: str = Query(None, description="访问密码"),
+        share_data: ShareTokenData = Depends(get_share_user_id),
+        db: Session = Depends(get_db),
 ):
     """获取会话详情和消息历史"""
     chat_service = SharedChatService(db)
@@ -266,10 +267,10 @@ def get_conversation(
     summary="发送消息（支持流式和非流式）"
 )
 async def chat(
-    payload: conversation_schema.ChatRequest,
-    share_data: ShareTokenData = Depends(get_share_user_id),
-    db: Session = Depends(get_db),
-    app_chat_service: Annotated[AppChatService, Depends(get_app_chat_service)] = None,
+        payload: conversation_schema.ChatRequest,
+        share_data: ShareTokenData = Depends(get_share_user_id),
+        db: Session = Depends(get_db),
+        app_chat_service: Annotated[AppChatService, Depends(get_app_chat_service)] = None,
 ):
     """发送消息并获取回复
 
@@ -313,12 +314,15 @@ async def chat(
         )
         end_user_id = str(new_end_user.id)
 
-        appid=share.app_id
+        appid = share.app_id
         """获取存储类型和工作空间的ID"""
 
-        # 直接通过 SQLAlchemy 查询 app
+        # 直接通过 SQLAlchemy 查询 app（仅查询未删除的应用）
         from app.models.app_model import App
-        app = db.query(App).filter(App.id == appid).first()
+        app = db.query(App).filter(
+            App.id == appid,
+            App.is_active.is_(True)
+        ).first()
         if not app:
             raise BusinessException("应用不存在", BizCode.APP_NOT_FOUND)
 
@@ -425,16 +429,16 @@ async def chat(
             # )
             async def event_generator():
                 async for event in app_chat_service.agnet_chat_stream(
-                    message=payload.message,
-                    conversation_id=conversation.id,  # 使用已创建的会话 ID
-                    user_id= str(new_end_user.id),  # 转换为字符串
-                    variables=payload.variables,
-                    web_search=payload.web_search,
-                    config=agent_config,
-                    memory=payload.memory,
-                    storage_type=storage_type,
-                    user_rag_memory_id=user_rag_memory_id,
-                    workspace_id=workspace_id
+                        message=payload.message,
+                        conversation_id=conversation.id,  # 使用已创建的会话 ID
+                        user_id=str(new_end_user.id),  # 转换为字符串
+                        variables=payload.variables,
+                        web_search=payload.web_search,
+                        config=agent_config,
+                        memory=payload.memory,
+                        storage_type=storage_type,
+                        user_rag_memory_id=user_rag_memory_id,
+                        workspace_id=workspace_id
                 ):
                     yield event
 
@@ -481,15 +485,15 @@ async def chat(
             async def event_generator():
                 async for event in app_chat_service.multi_agent_chat_stream(
 
-                    message=payload.message,
-                    conversation_id=conversation.id,  # 使用已创建的会话 ID
-                    user_id=str(new_end_user.id),  # 转换为字符串
-                    variables=payload.variables,
-                    config=config,
-                    web_search=payload.web_search,
-                    memory=payload.memory,
-                    storage_type=storage_type,
-                    user_rag_memory_id=user_rag_memory_id
+                        message=payload.message,
+                        conversation_id=conversation.id,  # 使用已创建的会话 ID
+                        user_id=str(new_end_user.id),  # 转换为字符串
+                        variables=payload.variables,
+                        config=config,
+                        web_search=payload.web_search,
+                        memory=payload.memory,
+                        storage_type=storage_type,
+                        user_rag_memory_id=user_rag_memory_id
                 ):
                     yield event
 
@@ -561,24 +565,27 @@ async def chat(
 
         # return success(data=conversation_schema.ChatResponse(**result))
     elif app_type == AppType.WORKFLOW:
-
         config = workflow_config_4_app_release(release)
+        if not config.id:
+            with get_db_read() as db:
+                source_config = WorkflowConfigRepository(db).get_by_app_id(release.app_id)
+                config.id = source_config.id
+        config.id = uuid.UUID(config.id)
         if payload.stream:
             async def event_generator():
-
                 async for event in app_chat_service.workflow_chat_stream(
-
-                    message=payload.message,
-                    conversation_id=conversation.id,  # 使用已创建的会话 ID
-                    user_id=end_user_id,  # 转换为字符串
-                    variables=payload.variables,
-                    config=config,
-                    web_search=payload.web_search,
-                    memory=payload.memory,
-                    storage_type=storage_type,
-                    user_rag_memory_id=user_rag_memory_id,
-                    app_id=release.app_id,
-                    workspace_id=workspace_id
+                        message=payload.message,
+                        conversation_id=conversation.id,  # 使用已创建的会话 ID
+                        user_id=end_user_id,  # 转换为字符串
+                        variables=payload.variables,
+                        config=config,
+                        web_search=payload.web_search,
+                        memory=payload.memory,
+                        storage_type=storage_type,
+                        user_rag_memory_id=user_rag_memory_id,
+                        app_id=release.app_id,
+                        workspace_id=workspace_id,
+                        release_id=release.id
                 ):
                     event_type = event.get("event", "message")
                     event_data = event.get("data", {})
@@ -610,7 +617,8 @@ async def chat(
             storage_type=storage_type,
             user_rag_memory_id=user_rag_memory_id,
             app_id=release.app_id,
-            workspace_id=workspace_id
+            workspace_id=workspace_id,
+            release_id=release.id
         )
         logger.debug(
             "工作流试运行返回结果",

api/app/controllers/service/app_api_controller.py

@@ -235,15 +235,16 @@ async def chat(
 
                     message=payload.message,
                     conversation_id=conversation.id,  # 使用已创建的会话 ID
-                    user_id=new_end_user.id,  # 转换为字符串
+                    user_id=end_user_id,  # 转换为字符串
                     variables=payload.variables,
                     config=config,
-                    web_search=payload.web_search,
-                    memory=payload.memory,
+                    web_search=web_search,
+                    memory=memory,
                     storage_type=storage_type,
                     user_rag_memory_id=user_rag_memory_id,
-                    app_id=app.app_id,
-                    workspace_id=workspace_id
+                    app_id=app.id,
+                    workspace_id=workspace_id,
+                    release_id=app.current_release.id,
                 ):
                     event_type = event.get("event", "message")
                     event_data = event.get("data", {})
@@ -267,15 +268,16 @@ async def chat(
 
             message=payload.message,
             conversation_id=conversation.id,  # 使用已创建的会话 ID
-            user_id=new_end_user.id,  # 转换为字符串
+            user_id=end_user_id,  # 转换为字符串
             variables=payload.variables,
             config=config,
-            web_search=payload.web_search,
-            memory=payload.memory,
+            web_search=web_search,
+            memory=memory,
             storage_type=storage_type,
             user_rag_memory_id=user_rag_memory_id,
-            app_id=app.app_id,
-            workspace_id=workspace_id
+            app_id=app.id,
+            workspace_id=workspace_id,
+            release_id=app.current_release.id
         )
         logger.debug(
             "工作流试运行返回结果",

api/app/controllers/service/memory_api_controller.py

@@ -39,7 +39,7 @@ async def write_memory_api_service(
     
     Stores memory content for the specified end user using the Memory API Service.
     """
-    logger.info(f"Memory write request - end_user_id: {payload.end_user_id}")
+    logger.info(f"Memory write request - end_user_id: {payload.end_user_id}, tenant_id: {api_key_auth.tenant_id}")
     
     memory_api_service = MemoryAPIService(db)
     

api/app/controllers/user_memory_controllers.py

@@ -5,13 +5,14 @@
 from typing import Optional
 import datetime
 from sqlalchemy.orm import Session
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends,Header
 
 from app.db import get_db
 from app.core.logging_config import get_api_logger
 from app.core.response_utils import success, fail
 from app.core.error_codes import BizCode
 from app.core.api_key_utils import timestamp_to_datetime
+from app.services.memory_base_service import Translation_English
 from app.services.user_memory_service import (
     UserMemoryService,
     analytics_memory_types,
@@ -20,7 +21,7 @@ from app.services.user_memory_service import (
 from app.services.memory_entity_relationship_service import MemoryEntityService,MemoryEmotion,MemoryInteraction
 from app.schemas.response_schema import ApiResponse
 from app.schemas.memory_storage_schema import GenerateCacheRequest
-
+from app.repositories.workspace_repository import WorkspaceRepository
 from app.schemas.end_user_schema import (
     EndUserProfileResponse,
     EndUserProfileUpdate,
@@ -44,6 +45,7 @@ router = APIRouter(
 @router.get("/analytics/memory_insight/report", response_model=ApiResponse)
 async def get_memory_insight_report_api(
     end_user_id: str,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
 ) -> dict:
@@ -53,10 +55,18 @@ async def get_memory_insight_report_api(
     此接口仅查询数据库中已缓存的记忆洞察数据，不执行生成操作。
     如需生成新的洞察报告，请使用专门的生成接口。
     """
+    workspace_id = current_user.current_workspace_id
+    workspace_repo = WorkspaceRepository(db)
+    workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
+
+    if workspace_models:
+        model_id = workspace_models.get("llm", None)
+    else:
+        model_id = None
     api_logger.info(f"记忆洞察报告查询请求: end_user_id={end_user_id}, user={current_user.username}")
     try:
         # 调用服务层获取缓存数据
-        result = await user_memory_service.get_cached_memory_insight(db, end_user_id)
+        result = await user_memory_service.get_cached_memory_insight(db, end_user_id,model_id,language_type)
 
         if result["is_cached"]:
             api_logger.info(f"成功返回缓存的记忆洞察报告: end_user_id={end_user_id}")
@@ -72,6 +82,7 @@ async def get_memory_insight_report_api(
 @router.get("/analytics/user_summary", response_model=ApiResponse)
 async def get_user_summary_api(
     end_user_id: str,
+    language_type: str = Header(default="zh", alias="X-Language-Type"),
     current_user: User = Depends(get_current_user),
     db: Session = Depends(get_db),
 ) -> dict:
@@ -81,10 +92,18 @@ async def get_user_summary_api(
     此接口仅查询数据库中已缓存的用户摘要数据，不执行生成操作。
     如需生成新的用户摘要，请使用专门的生成接口。
     """
+    workspace_id = current_user.current_workspace_id
+    workspace_repo = WorkspaceRepository(db)
+    workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
+
+    if workspace_models:
+        model_id = workspace_models.get("llm", None)
+    else:
+        model_id = None
     api_logger.info(f"用户摘要查询请求: end_user_id={end_user_id}, user={current_user.username}")
     try:
         # 调用服务层获取缓存数据
-        result = await user_memory_service.get_cached_user_summary(db, end_user_id)
+        result = await user_memory_service.get_cached_user_summary(db, end_user_id,model_id,language_type)
 
         if result["is_cached"]:
             api_logger.info(f"成功返回缓存的用户摘要: end_user_id={end_user_id}")
@@ -116,27 +135,27 @@ async def generate_cache_api(
         api_logger.warning(f"用户 {current_user.username} 尝试生成缓存但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
 
-    group_id = request.end_user_id
+    end_user_id = request.end_user_id
 
     api_logger.info(
         f"缓存生成请求: user={current_user.username}, workspace={workspace_id}, "
-        f"end_user_id={group_id if group_id else '全部用户'}"
+        f"end_user_id={end_user_id if end_user_id else '全部用户'}"
     )
 
     try:
-        if group_id:
+        if end_user_id:
             # 为单个用户生成
-            api_logger.info(f"开始为单个用户生成缓存: end_user_id={group_id}")
+            api_logger.info(f"开始为单个用户生成缓存: end_user_id={end_user_id}")
 
             # 生成记忆洞察
-            insight_result = await user_memory_service.generate_and_cache_insight(db, group_id, workspace_id)
+            insight_result = await user_memory_service.generate_and_cache_insight(db, end_user_id, workspace_id)
 
             # 生成用户摘要
-            summary_result = await user_memory_service.generate_and_cache_summary(db, group_id, workspace_id)
+            summary_result = await user_memory_service.generate_and_cache_summary(db, end_user_id, workspace_id)
 
             # 构建响应
             result = {
-                "end_user_id": group_id,
+                "end_user_id": end_user_id,
                 "insight_success": insight_result["success"],
                 "summary_success": summary_result["success"],
                 "errors": []
@@ -156,9 +175,9 @@ async def generate_cache_api(
 
             # 记录结果
             if result["insight_success"] and result["summary_success"]:
-                api_logger.info(f"成功为用户 {group_id} 生成缓存")
+                api_logger.info(f"成功为用户 {end_user_id} 生成缓存")
             else:
-                api_logger.warning(f"用户 {group_id} 的缓存生成部分失败: {result['errors']}")
+                api_logger.warning(f"用户 {end_user_id} 的缓存生成部分失败: {result['errors']}")
 
             return success(data=result, msg="生成完成")
 
@@ -253,7 +272,6 @@ async def get_graph_data_api(
             depth=depth,
             center_node_id=center_node_id
         )
-
         # 检查是否有错误消息
         if "message" in result and result["statistics"]["total_nodes"] == 0:
             api_logger.warning(f"图数据查询返回空结果: {result.get('message')}")
@@ -278,7 +296,13 @@ async def get_end_user_profile(
     db: Session = Depends(get_db),
 ) -> dict:
     workspace_id = current_user.current_workspace_id
+    workspace_repo = WorkspaceRepository(db)
+    workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
 
+    if workspace_models:
+        model_id = workspace_models.get("llm", None)
+    else:
+        model_id = None
     # 检查用户是否已选择工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试查询用户信息但未选择工作空间")
@@ -296,7 +320,6 @@ async def get_end_user_profile(
         if not end_user:
             api_logger.warning(f"终端用户不存在: end_user_id={end_user_id}")
             return fail(BizCode.INVALID_PARAMETER, "终端用户不存在", f"end_user_id={end_user_id}")
-
         # 构建响应数据
         profile_data = EndUserProfileResponse(
             id=end_user.id,
@@ -328,12 +351,11 @@ async def update_end_user_profile(
 
     该接口可以更新用户的姓名、职位、部门、联系方式、电话和入职日期等信息。
     所有字段都是可选的，只更新提供的字段。
-
     """
     workspace_id = current_user.current_workspace_id
     end_user_id = profile_update.end_user_id
 
-    # 检查用户是否已选择工作空间
+    # 验证工作空间
     if workspace_id is None:
         api_logger.warning(f"用户 {current_user.username} 尝试更新用户信息但未选择工作空间")
         return fail(BizCode.INVALID_PARAMETER, "请先切换到一个工作空间", "current_workspace_id is None")
@@ -343,65 +365,41 @@ async def update_end_user_profile(
         f"workspace={workspace_id}"
     )
 
-    try:
-        # 查询终端用户
-        end_user = db.query(EndUser).filter(EndUser.id == end_user_id).first()
+    # 调用 Service 层处理业务逻辑
+    result = user_memory_service.update_end_user_profile(db, end_user_id, profile_update)
 
-        if not end_user:
-            api_logger.warning(f"终端用户不存在: end_user_id={end_user_id}")
-            return fail(BizCode.INVALID_PARAMETER, "终端用户不存在", f"end_user_id={end_user_id}")
-
-        # 更新字段（只更新提供的字段，排除 end_user_id）
-        # 允许 None 值来重置字段（如 hire_date）
-        update_data = profile_update.model_dump(exclude_unset=True, exclude={'end_user_id'})
-
-        # 特殊处理 hire_date：如果提供了时间戳，转换为 DateTime
-        if 'hire_date' in update_data:
-            hire_date_timestamp = update_data['hire_date']
-            if hire_date_timestamp is not None:
-                update_data['hire_date'] = timestamp_to_datetime(hire_date_timestamp)
-            # 如果是 None，保持 None（允许清空）
-
-        for field, value in update_data.items():
-            setattr(end_user, field, value)
-
-        # 更新 updated_at 时间戳
-        end_user.updated_at = datetime.datetime.now()
-
-        # 更新 updatetime_profile 为当前时间
-        end_user.updatetime_profile = datetime.datetime.now()
-
-        # 提交更改
-        db.commit()
-        db.refresh(end_user)
-
-        # 构建响应数据
-        profile_data = EndUserProfileResponse(
-            id=end_user.id,
-            other_name=end_user.other_name,
-            position=end_user.position,
-            department=end_user.department,
-            contact=end_user.contact,
-            phone=end_user.phone,
-            hire_date=end_user.hire_date,
-            updatetime_profile=end_user.updatetime_profile
-        )
-
-        api_logger.info(f"成功更新用户信息: end_user_id={end_user_id}, updated_fields={list(update_data.keys())}")
-        return success(data=UserMemoryService.convert_profile_to_dict_with_timestamp(profile_data), msg="更新成功")
-
-    except Exception as e:
-        db.rollback()
-        api_logger.error(f"用户信息更新失败: end_user_id={end_user_id}, error={str(e)}")
-        return fail(BizCode.INTERNAL_ERROR, "用户信息更新失败", str(e))
+    if result["success"]:
+        api_logger.info(f"成功更新用户信息: end_user_id={end_user_id}")
+        return success(data=result["data"], msg="更新成功")
+    else:
+        error_msg = result["error"]
+        api_logger.error(f"用户信息更新失败: end_user_id={end_user_id}, error={error_msg}")
+        
+        # 根据错误类型映射到合适的业务错误码
+        if error_msg == "终端用户不存在":
+            return fail(BizCode.USER_NOT_FOUND, "终端用户不存在", error_msg)
+        elif error_msg == "无效的用户ID格式":
+            return fail(BizCode.INVALID_USER_ID, "无效的用户ID格式", error_msg)
+        else:
+            # 只有未预期的错误才使用 INTERNAL_ERROR
+            return fail(BizCode.INTERNAL_ERROR, "用户信息更新失败", error_msg)
 
 @router.get("/memory_space/timeline_memories", response_model=ApiResponse)
-async def memory_space_timeline_of_shared_memories(id: str, label: str,
+async def memory_space_timeline_of_shared_memories(id: str, label: str,language_type: str = Header(default="zh", alias="X-Language-Type"),
                                       current_user: User = Depends(get_current_user),
                                       db: Session = Depends(get_db),
                                       ):
+    workspace_id=current_user.current_workspace_id
+    workspace_repo = WorkspaceRepository(db)
+    workspace_models = workspace_repo.get_workspace_models_configs(workspace_id)
+
+    if workspace_models:
+        model_id = workspace_models.get("llm", None)
+    else:
+        model_id = None
     MemoryEntity = MemoryEntityService(id, label)
-    timeline_memories_result = await MemoryEntity.get_timeline_memories_server()
+    timeline_memories_result = await MemoryEntity.get_timeline_memories_server(model_id, language_type)
+
     return success(data=timeline_memories_result, msg="共同记忆时间线")
 @router.get("/memory_space/relationship_evolution", response_model=ApiResponse)
 async def memory_space_relationship_evolution(id: str, label: str,

api/app/controllers/workflow_controller.py

@@ -54,7 +54,7 @@ async def create_workflow_config(
         app = db.query(App).filter(
             App.id == app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:
@@ -214,7 +214,7 @@ async def delete_workflow_config(
         app = db.query(App).filter(
             App.id == app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:
@@ -259,7 +259,7 @@ async def validate_workflow_config(
         app = db.query(App).filter(
             App.id == app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:
@@ -329,7 +329,7 @@ async def get_workflow_executions(
         app = db.query(App).filter(
             App.id == app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:
@@ -389,7 +389,7 @@ async def get_workflow_execution(
         app = db.query(App).filter(
             App.id == execution.app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:
@@ -440,7 +440,7 @@ async def run_workflow(
         app = db.query(App).filter(
             App.id == app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:
@@ -578,7 +578,7 @@ async def cancel_workflow_execution(
         app = db.query(App).filter(
             App.id == execution.app_id,
             App.workspace_id == current_user.current_workspace_id,
-            App.is_active == True
+            App.is_active.is_(True)
         ).first()
 
         if not app:

api/app/core/agent/langchain_agent.py

@@ -7,27 +7,21 @@ LangChain Agent 封装
 - 支持流式输出
 - 使用 RedBearLLM 支持多提供商
 """
-import os
+
 import time
 from typing import Any, AsyncGenerator, Dict, List, Optional, Sequence
 
-
+from app.core.memory.agent.langgraph_graph.write_graph import  write_long_term
 from app.db import get_db
 from app.core.logging_config import get_business_logger
-from app.core.memory.agent.utils.redis_tool import store
 from app.core.models import RedBearLLM, RedBearModelConfig
 from app.models.models_model import ModelType
-from app.repositories.memory_short_repository import LongTermMemoryRepository
 from app.services.memory_agent_service import (
     get_end_user_connected_config,
 )
-from app.services.memory_konwledges_server import write_rag
-from app.services.task_service import get_task_memory_write_result
-from app.tasks import write_message_task
 from langchain.agents import create_agent
 from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
 from langchain_core.tools import BaseTool
-
 logger = get_business_logger()
 
 
@@ -104,7 +98,7 @@ class LangChainAgent:
                 "streaming": streaming,
                 "tool_count": len(self.tools),
                 "tool_names": [tool.name for tool in self.tools] if self.tools else [],
-                "tool_count": len(self.tools)
+                # "tool_count": len(self.tools)
             }
         )
 
@@ -143,46 +137,7 @@ class LangChainAgent:
             user_content = f"参考信息：\n{context}\n\n用户问题：\n{user_content}"
 
         messages.append(HumanMessage(content=user_content))
-
         return messages
-    async def term_memory_save(self,messages,end_user_end,aimessages):
-        '''短长期存储redis，为不影响正常使用6句一段话，存储用户名加一个前缀，当数据存够6条返回给neo4j'''
-        end_user_end=f"Term_{end_user_end}"
-        print(messages)
-        print(aimessages)
-        session_id = store.save_session(
-                        userid=end_user_end,
-                        messages=messages,
-                        apply_id=end_user_end,
-                        group_id=end_user_end,
-                        aimessages=aimessages
-                    )
-        store.delete_duplicate_sessions()
-        # logger.info(f'Redis_Agent:{end_user_end};{session_id}')
-        return session_id
-    async def term_memory_redis_read(self,end_user_end):
-        end_user_end = f"Term_{end_user_end}"
-        history = store.find_user_apply_group(end_user_end, end_user_end, end_user_end)
-        # logger.info(f'Redis_Agent:{end_user_end};{history}')
-        messagss_list=[]
-        retrieved_content=[]
-        for messages in history:
-            query = messages.get("Query")
-            aimessages = messages.get("Answer")
-            messagss_list.append(f'用户:{query}。AI回复:{aimessages}')
-            retrieved_content.append({query: aimessages})
-        return messagss_list,retrieved_content
-
-
-    async def write(self,storage_type,end_user_id,message,user_rag_memory_id,actual_end_user_id,content,actual_config_id):
-        if storage_type == "rag":
-            await write_rag(end_user_id, message, user_rag_memory_id)
-            logger.info(f'RAG_Agent:{end_user_id};{user_rag_memory_id}')
-        else:
-            write_id = write_message_task.delay(actual_end_user_id, content, actual_config_id, storage_type,
-                                                user_rag_memory_id)
-            write_status = get_task_memory_write_result(str(write_id))
-            logger.info(f'Agent:{actual_end_user_id};{write_status}')
 
     async def chat(
             self,
@@ -227,29 +182,6 @@ class LangChainAgent:
         actual_end_user_id = end_user_id if end_user_id is not None else "unknown"
         logger.info(f'写入类型{storage_type,str(end_user_id), message, str(user_rag_memory_id)}')
         print(f'写入类型{storage_type,str(end_user_id), message, str(user_rag_memory_id)}')
-
-        history_term_memory_result = await self.term_memory_redis_read(end_user_id)
-        history_term_memory = history_term_memory_result[0]
-        db_for_memory = next(get_db())
-        if memory_flag:
-            if len(history_term_memory)>=4 and storage_type != "rag":
-                history_term_memory = ';'.join(history_term_memory)
-                retrieved_content = history_term_memory_result[1]
-                print(retrieved_content)
-                # 为长期记忆操作获取新的数据库连接
-                try:
-                    repo = LongTermMemoryRepository(db_for_memory)
-                    repo.upsert(end_user_id, retrieved_content)
-                    logger.info(
-                        f'写入短长期：{storage_type, str(end_user_id), history_term_memory, str(user_rag_memory_id)}')
-                except Exception as e:
-                    logger.error(f"Failed to write to LongTermMemory: {e}")
-                    raise
-                finally:
-                    db_for_memory.close()
-
-                await self.write(storage_type,end_user_id,history_term_memory,user_rag_memory_id,actual_end_user_id,history_term_memory,actual_config_id)
-            await self.write(storage_type,end_user_id,message,user_rag_memory_id,actual_end_user_id,message,actual_config_id)
         try:
             # 准备消息列表
             messages = self._prepare_messages(message, history, context)
@@ -270,15 +202,17 @@ class LangChainAgent:
             # 获取最后的 AI 消息
             output_messages = result.get("messages", [])
             content = ""
+            total_tokens = 0
             for msg in reversed(output_messages):
                 if isinstance(msg, AIMessage):
                     content = msg.content
+                    response_meta = msg.response_metadata if hasattr(msg, 'response_metadata') else None
+                    total_tokens = response_meta.get("token_usage", {}).get("total_tokens", 0) if response_meta else 0
                     break
 
             elapsed_time = time.time() - start_time
             if memory_flag:
-                await self.write(storage_type,end_user_id,content,user_rag_memory_id,actual_end_user_id,content,actual_config_id)
-                await self.term_memory_save(message_chat,end_user_id,content)
+                await write_long_term(storage_type, end_user_id, message_chat, content, user_rag_memory_id, actual_config_id)
             response = {
                 "content": content,
                 "model": self.model_name,
@@ -286,7 +220,7 @@ class LangChainAgent:
                 "usage": {
                     "prompt_tokens": 0,
                     "completion_tokens": 0,
-                    "total_tokens": 0
+                    "total_tokens": total_tokens
                 }
             }
 
@@ -347,26 +281,8 @@ class LangChainAgent:
             except Exception as e:
                 logger.warning(f"Failed to get db session: {e}")
 
-        history_term_memory_result = await self.term_memory_redis_read(end_user_id)
-        history_term_memory = history_term_memory_result[0]
-        if memory_flag:
-            if len(history_term_memory) >= 4 and storage_type != "rag":
-                history_term_memory = ';'.join(history_term_memory)
-                retrieved_content = history_term_memory_result[1]
-                db_for_memory = next(get_db())
-                try:
-                    repo = LongTermMemoryRepository(db_for_memory)
-                    repo.upsert(end_user_id, retrieved_content)
-                    logger.info(
-                        f'写入短长期：{storage_type, str(end_user_id), history_term_memory, str(user_rag_memory_id)}')
-                    await self.write(storage_type, end_user_id, history_term_memory, user_rag_memory_id, end_user_id,
-                                     history_term_memory, actual_config_id)
-                except Exception as e:
-                    logger.error(f"Failed to write to long term memory: {e}")
-                finally:
-                    db_for_memory.close()
 
-            await self.write(storage_type, end_user_id, message, user_rag_memory_id, end_user_id, message, actual_config_id)
+            # 注意：不在这里写入用户消息，等 AI 回复后一起写入
         try:
             # 准备消息列表
             messages = self._prepare_messages(message, history, context)
@@ -380,7 +296,7 @@ class LangChainAgent:
 
             # 统一使用 agent 的 astream_events 实现流式输出
             logger.debug("使用 Agent astream_events 实现流式输出")
-            full_content=''
+            full_content = ''
             try:
                 async for event in self.agent.astream_events(
                     {"messages": messages},
@@ -417,10 +333,17 @@ class LangChainAgent:
                         logger.debug(f"工具调用结束: {event.get('name')}")
                 
                 logger.debug(f"Agent 流式完成，共 {chunk_count} 个事件")
+                # 统计token消耗
+                output_messages = event.get("data", {}).get("output", {}).get("messages", [])
+                for msg in reversed(output_messages):
+                    if isinstance(msg, AIMessage):
+                        response_meta = msg.response_metadata if hasattr(msg, 'response_metadata') else None
+                        total_tokens = response_meta.get("token_usage", {}).get("total_tokens",
+                                                                                0) if response_meta else 0
+                        yield total_tokens
+                        break
                 if memory_flag:
-                    await self.write(storage_type, end_user_id,full_content, user_rag_memory_id, end_user_id,full_content, actual_config_id)
-                    await self.term_memory_save(message_chat, end_user_id, full_content)
-                
+                    await write_long_term(storage_type, end_user_id, message_chat, full_content, user_rag_memory_id, actual_config_id)
             except Exception as e:
                 logger.error(f"Agent astream_events 失败: {str(e)}", exc_info=True)
                 raise

api/app/core/config.py

@@ -9,6 +9,25 @@ load_dotenv()
 
 
 class Settings:
+    # ========================================================================
+    # Deployment Mode Configuration
+    # ========================================================================
+    # community: 社区版（开源，功能受限）
+    # cloud: SaaS 云服务版（全功能，按量计费）
+    # enterprise: 企业私有化版（License 控制）
+    DEPLOYMENT_MODE: str = os.getenv("DEPLOYMENT_MODE", "community")
+    
+    # License 配置（企业版）
+    LICENSE_FILE: str = os.getenv("LICENSE_FILE", "/etc/app/license.json")
+    LICENSE_SERVER_URL: str = os.getenv("LICENSE_SERVER_URL", "https://license.yourcompany.com")
+    
+    # 计费服务配置（SaaS 版）
+    BILLING_SERVICE_URL: str = os.getenv("BILLING_SERVICE_URL", "")
+    
+    # 基础 URL（用于 SSO 回调等）
+    BASE_URL: str = os.getenv("BASE_URL", "http://localhost:8000")
+    FRONTEND_URL: str = os.getenv("FRONTEND_URL", "http://localhost:3000")
+    
     ENABLE_SINGLE_WORKSPACE: bool = os.getenv("ENABLE_SINGLE_WORKSPACE", "true").lower() == "true"
     # API Keys Configuration
     OPENAI_API_KEY: str = os.getenv("OPENAI_API_KEY", "")
@@ -38,6 +57,7 @@ class Settings:
     REDIS_PORT: int = int(os.getenv("REDIS_PORT", "6379"))
     REDIS_DB: int = int(os.getenv("REDIS_DB", "1"))
     REDIS_PASSWORD: str = os.getenv("REDIS_PASSWORD", "")
+    
 
     # ElasticSearch configuration
     ELASTICSEARCH_HOST: str = os.getenv("ELASTICSEARCH_HOST", "https://127.0.0.1")
@@ -71,10 +91,30 @@ class Settings:
 
     # Single Sign-On configuration
     ENABLE_SINGLE_SESSION: bool = os.getenv("ENABLE_SINGLE_SESSION", "false").lower() == "true"
+    
+    # SSO 免登配置
+    SSO_TOKEN_EXPIRE_SECONDS: int = int(os.getenv("SSO_TOKEN_EXPIRE_SECONDS", "300"))
+    SSO_TRUSTED_SOURCES_CONFIG: str = os.getenv("SSO_TRUSTED_SOURCES_CONFIG", "{}")
 
     # File Upload
     MAX_FILE_SIZE: int = int(os.getenv("MAX_FILE_SIZE", "52428800"))
     FILE_PATH: str = os.getenv("FILE_PATH", "/files")
+    FILE_URL_EXPIRES: int = int(os.getenv("FILE_URL_EXPIRES", "3600"))
+
+    # Storage Configuration
+    STORAGE_TYPE: str = os.getenv("STORAGE_TYPE", "local")
+
+    # Aliyun OSS Configuration
+    OSS_ENDPOINT: str = os.getenv("OSS_ENDPOINT", "")
+    OSS_ACCESS_KEY_ID: str = os.getenv("OSS_ACCESS_KEY_ID", "")
+    OSS_ACCESS_KEY_SECRET: str = os.getenv("OSS_ACCESS_KEY_SECRET", "")
+    OSS_BUCKET_NAME: str = os.getenv("OSS_BUCKET_NAME", "")
+
+    # AWS S3 Configuration
+    S3_REGION: str = os.getenv("S3_REGION", "")
+    S3_ACCESS_KEY_ID: str = os.getenv("S3_ACCESS_KEY_ID", "")
+    S3_SECRET_ACCESS_KEY: str = os.getenv("S3_SECRET_ACCESS_KEY", "")
+    S3_BUCKET_NAME: str = os.getenv("S3_BUCKET_NAME", "")
 
     # VOLC ASR settings
     VOLC_APP_KEY: str = os.getenv("VOLC_APP_KEY", "")
@@ -90,6 +130,7 @@ class Settings:
 
     # Server Configuration
     SERVER_IP: str = os.getenv("SERVER_IP", "127.0.0.1")
+    FILE_LOCAL_SERVER_URL : str = os.getenv("FILE_LOCAL_SERVER_URL", "http://localhost:8000/api")
 
     # ========================================================================
     # Internal Configuration (not in .env, used by application code)
@@ -116,6 +157,11 @@ class Settings:
         if origin.strip()
     ]
 
+    # Language Configuration
+    # Supported values: "zh" (Chinese), "en" (English)
+    # This controls the language used for memory summary titles and other generated content
+    DEFAULT_LANGUAGE: str = os.getenv("DEFAULT_LANGUAGE", "zh")
+
     # Logging settings
     LOG_LEVEL: str = os.getenv("LOG_LEVEL", "INFO")
     LOG_FORMAT: str = os.getenv("LOG_FORMAT", "%(asctime)s - %(name)s - %(levelname)s - %(message)s")
@@ -146,6 +192,7 @@ class Settings:
     # Celery configuration (internal)
     CELERY_BROKER: int = int(os.getenv("CELERY_BROKER", "1"))
     CELERY_BACKEND: int = int(os.getenv("CELERY_BACKEND", "2"))
+
     REFLECTION_INTERVAL_SECONDS: float = float(os.getenv("REFLECTION_INTERVAL_SECONDS", "300"))
     HEALTH_CHECK_SECONDS: float = float(os.getenv("HEALTH_CHECK_SECONDS", "600"))
     MEMORY_INCREMENT_INTERVAL_HOURS: float = float(os.getenv("MEMORY_INCREMENT_INTERVAL_HOURS", "24"))
@@ -166,7 +213,7 @@ class Settings:
     ENABLE_TOOL_MANAGEMENT: bool = os.getenv("ENABLE_TOOL_MANAGEMENT", "true").lower() == "true"
 
     # official environment system version
-    SYSTEM_VERSION: str = os.getenv("SYSTEM_VERSION", "v0.2.0")
+    SYSTEM_VERSION: str = os.getenv("SYSTEM_VERSION", "v0.2.1")
 
     # workflow config
     WORKFLOW_NODE_TIMEOUT: int = int(os.getenv("WORKFLOW_NODE_TIMEOUT", 600))

api/app/core/memory/agent/langgraph_graph/__init__.py

@@ -1,16 +0,0 @@
-"""
-LangGraph Graph package for memory agent.
-
-This package provides the LangGraph workflow orchestrator with modular
-node implementations, routing logic, and state management.
-
-Package structure:
-- read_graph: Main graph factory for read operations
-- write_graph: Main graph factory for write operations
-- nodes: LangGraph node implementations
-- routing: State routing logic
-- state: State management utilities
-"""
-from app.core.memory.agent.langgraph_graph.read_graph import make_read_graph
-
-__all__ = ['make_read_graph']

api/app/core/memory/agent/langgraph_graph/nodes/__init__.py

@@ -4,7 +4,7 @@ LangGraph node implementations.
 This module contains custom node implementations for the LangGraph workflow.
 """
 
-from app.core.memory.agent.langgraph_graph.nodes.tool_node import ToolExecutionNode
-from app.core.memory.agent.langgraph_graph.nodes.input_node import create_input_message
-
-__all__ = ["ToolExecutionNode", "create_input_message"]
+# from app.core.memory.agent.langgraph_graph.nodes.tool_node import ToolExecutionNode
+# from app.core.memory.agent.langgraph_graph.nodes.input_node import create_input_message
+#
+# __all__ = ["ToolExecutionNode", "create_input_message"]

api/app/core/memory/agent/langgraph_graph/nodes/data_nodes.py

@@ -0,0 +1,16 @@
+from app.core.memory.agent.utils.llm_tools import ReadState, WriteState
+
+
+def content_input_node(state: ReadState) -> ReadState:
+    """开始节点 - 提取内容并保持状态信息"""
+
+    content = state['messages'][0].content if state.get('messages') else ''
+    # 返回内容并保持所有状态信息
+    return {"data": content}
+
+def content_input_write(state: WriteState) -> WriteState:
+    """开始节点 - 提取内容并保持状态信息"""
+
+    content = state['messages'][0].content if state.get('messages') else ''
+    # 返回内容并保持所有状态信息
+    return {"data": content}

api/app/core/memory/agent/langgraph_graph/nodes/input_node.py

@@ -1,150 +0,0 @@
-"""
-Input node for LangGraph workflow entry point.
-
-This module provides the create_input_message function which processes initial
-user input with multimodal support and creates the first tool call message.
-"""
-
-import logging
-import re
-import uuid
-from datetime import datetime
-from typing import Any, Dict
-
-from app.core.memory.agent.utils.multimodal import MultimodalProcessor
-from app.schemas.memory_config_schema import MemoryConfig
-from langchain_core.messages import AIMessage
-
-logger = logging.getLogger(__name__)
-
-
-async def create_input_message(
-    state: Dict[str, Any],
-    tool_name: str,
-    session_id: str,
-    search_switch: str,
-    apply_id: str,
-    group_id: str,
-    multimodal_processor: MultimodalProcessor,
-    memory_config: MemoryConfig,
-) -> Dict[str, Any]:
-    """
-    Create initial tool call message from user input.
-    
-    This function:
-    1. Extracts the last message content from state
-    2. Processes multimodal inputs (images/audio) using the multimodal processor
-    3. Generates a unique message ID
-    4. Extracts namespace from session_id
-    5. Handles verified_data extraction for backward compatibility
-    6. Returns AIMessage with complete tool_calls structure
-    
-    Args:
-        state: LangGraph state dictionary containing messages
-        tool_name: Name of the tool to invoke (typically "Split_The_Problem")
-        session_id: Session identifier (format: "call_id_{namespace}")
-        search_switch: Search routing parameter
-        apply_id: Application identifier
-        group_id: Group identifier
-        multimodal_processor: Processor for handling image/audio inputs
-        memory_config: MemoryConfig object containing all configuration
-        
-    Returns:
-        State update with AIMessage containing tool_call
-        
-    Examples:
-        >>> state = {"messages": [HumanMessage(content="What is AI?")]}
-        >>> result = await create_input_message(
-        ...     state, "Split_The_Problem", "call_id_user123", "0", "app1", "group1", processor, config
-        ... )
-        >>> result["messages"][0].tool_calls[0]["name"]
-        'Split_The_Problem'
-    """
-    messages = state.get("messages", [])
-    
-    # Extract last message content
-    if messages:
-        last_message = messages[-1].content if hasattr(messages[-1], 'content') else str(messages[-1])
-    else:
-        logger.warning("[create_input_message] No messages in state, using empty string")
-        last_message = ""
-    
-    logger.debug(f"[create_input_message] Original input: {last_message[:100]}...")
-    
-    # Process multimodal input (images/audio)
-    try:
-        processed_content = await multimodal_processor.process_input(last_message)
-        if processed_content != last_message:
-            logger.info(
-                f"[create_input_message] Multimodal processing converted input "
-                f"from {len(last_message)} to {len(processed_content)} chars"
-            )
-        last_message = processed_content
-    except Exception as e:
-        logger.error(
-            f"[create_input_message] Multimodal processing failed: {e}",
-            exc_info=True
-        )
-        # Continue with original content
-    
-    # Generate unique message ID
-    uuid_str = uuid.uuid4()
-    time_str = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-    
-    # Extract namespace from session_id
-    # Expected format: "call_id_{namespace}" or similar
-    try:
-        namespace = str(session_id).split('_id_')[1]
-    except (IndexError, AttributeError):
-        logger.warning(
-            f"[create_input_message] Could not extract namespace from session_id: {session_id}"
-        )
-        namespace = "unknown"
-    
-    # Handle verified_data extraction (backward compatibility)
-    # This regex-based extraction is kept for compatibility with existing data formats
-    if 'verified_data' in str(last_message):
-        try:
-            messages_last = str(last_message).replace('\\n', '').replace('\\', '')
-            query_match = re.findall(r'"query": "(.*?)",', messages_last)
-            if query_match:
-                last_message = query_match[0]
-                logger.debug(
-                    f"[create_input_message] Extracted query from verified_data: {last_message}"
-                )
-        except Exception as e:
-            logger.warning(
-                f"[create_input_message] Failed to extract query from verified_data: {e}"
-            )
-    
-    # Construct tool call message
-    tool_call_id = f"{session_id}_{uuid_str}"
-    
-    logger.info(
-        f"[create_input_message] Creating tool call for '{tool_name}' "
-        f"with ID: {tool_call_id}"
-    )
-    
-    # Build tool arguments
-    tool_args = {
-        "sentence": last_message,
-        "sessionid": session_id,
-        "messages_id": str(uuid_str),
-        "search_switch": search_switch,
-        "apply_id": apply_id,
-        "group_id": group_id,
-        "memory_config": memory_config,
-    }
-    
-    return {
-        "messages": [
-            AIMessage(
-                content="",
-                tool_calls=[{
-                    "name": tool_name,
-                    "args": tool_args,
-                    "id": tool_call_id
-                }]
-            )
-        ]
-    }

api/app/core/memory/agent/langgraph_graph/nodes/problem_nodes.py

@@ -0,0 +1,249 @@
+import os
+import json
+import time
+from app.core.logging_config import get_agent_logger
+from app.db import get_db
+
+from app.core.memory.agent.models.problem_models import ProblemExtensionResponse
+from app.core.memory.agent.utils.llm_tools import (
+    PROJECT_ROOT_,
+    ReadState,
+)
+from app.core.memory.agent.utils.redis_tool import store
+from app.core.memory.agent.utils.session_tools import SessionService
+from app.core.memory.agent.utils.template_tools import TemplateService
+from app.core.memory.agent.services.optimized_llm_service import LLMServiceMixin
+
+template_root = os.path.join(PROJECT_ROOT_, 'memory', 'agent', 'utils', 'prompt')
+db_session = next(get_db())
+logger = get_agent_logger(__name__)
+
+
+class ProblemNodeService(LLMServiceMixin):
+    """问题处理节点服务类"""
+
+    def __init__(self):
+        super().__init__()
+        self.template_service = TemplateService(template_root)
+
+
+# 创建全局服务实例
+problem_service = ProblemNodeService()
+
+
+async def Split_The_Problem(state: ReadState) -> ReadState:
+    """问题分解节点"""
+    # 从状态中获取数据
+    content = state.get('data', '')
+    end_user_id = state.get('end_user_id', '')
+    memory_config = state.get('memory_config', None)
+
+    history = await SessionService(store).get_history(end_user_id, end_user_id, end_user_id)
+
+    # 生成 JSON schema 以指导 LLM 输出正确格式
+    json_schema = ProblemExtensionResponse.model_json_schema()
+
+    system_prompt = await problem_service.template_service.render_template(
+        template_name='problem_breakdown_prompt.jinja2',
+        operation_name='split_the_problem',
+        history=history,
+        sentence=content,
+        json_schema=json_schema
+    )
+
+    try:
+        # 使用优化的LLM服务
+        structured = await problem_service.call_llm_structured(
+            state=state,
+            db_session=db_session,
+            system_prompt=system_prompt,
+            response_model=ProblemExtensionResponse,
+            fallback_value=[]
+        )
+
+        # 添加更详细的日志记录
+        logger.info(f"Split_The_Problem: 开始处理问题分解，内容长度: {len(content)}")
+
+        # 验证结构化响应
+        if not structured or not hasattr(structured, 'root'):
+            logger.warning("Split_The_Problem: 结构化响应为空或格式不正确")
+            split_result = json.dumps([], ensure_ascii=False)
+        elif not structured.root:
+            logger.warning("Split_The_Problem: 结构化响应的root为空")
+            split_result = json.dumps([], ensure_ascii=False)
+        else:
+            split_result = json.dumps(
+                [item.model_dump() for item in structured.root],
+                ensure_ascii=False
+            )
+
+        split_result_dict = []
+        for index, item in enumerate(json.loads(split_result)):
+            split_data = {
+                "id": f"Q{index + 1}",
+                "question": item['extended_question'],
+                "type": item['type'],
+                "reason": item['reason']
+            }
+            split_result_dict.append(split_data)
+
+        logger.info(f"Split_The_Problem: 成功生成 {len(structured.root) if structured.root else 0} 个分解项")
+
+        result = {
+            "context": split_result,
+            "original": content,
+            "_intermediate": {
+                "type": "problem_split",
+                "title": "问题拆分",
+                "data": split_result_dict,
+                "original_query": content
+            }
+        }
+
+    except Exception as e:
+        logger.error(
+            f"Split_The_Problem failed: {e}",
+            exc_info=True
+        )
+
+        # 提供更详细的错误信息
+        error_details = {
+            "error_type": type(e).__name__,
+            "error_message": str(e),
+            "content_length": len(content),
+            "llm_model_id": memory_config.llm_model_id if memory_config else None
+        }
+
+        logger.error(f"Split_The_Problem error details: {error_details}")
+
+        # 创建默认的空结果
+        result = {
+            "context": json.dumps([], ensure_ascii=False),
+            "original": content,
+            "error": str(e),
+            "_intermediate": {
+                "type": "problem_split",
+                "title": "问题拆分",
+                "data": [],
+                "original_query": content,
+                "error": error_details
+            }
+        }
+
+    # 返回更新后的状态，包含spit_context字段
+    return {"spit_data": result}
+
+
+async def Problem_Extension(state: ReadState) -> ReadState:
+    """问题扩展节点"""
+    # 获取原始数据和分解结果
+    start = time.time()
+    content = state.get('data', '')
+    data = state.get('spit_data', '')['context']
+    end_user_id = state.get('end_user_id', '')
+    storage_type = state.get('storage_type', '')
+    user_rag_memory_id = state.get('user_rag_memory_id', '')
+    memory_config = state.get('memory_config', None)
+
+    databasets = {}
+    try:
+        data = json.loads(data)
+        for i in data:
+            databasets[i['extended_question']] = i['type']
+    except (json.JSONDecodeError, KeyError, TypeError) as e:
+        logger.error(f"Problem_Extension: 数据解析失败: {e}")
+        # 使用空字典作为fallback
+        databasets = {}
+        data = []
+
+    history = await SessionService(store).get_history(end_user_id, end_user_id, end_user_id)
+
+    # 生成 JSON schema 以指导 LLM 输出正确格式
+    json_schema = ProblemExtensionResponse.model_json_schema()
+
+    system_prompt = await problem_service.template_service.render_template(
+        template_name='Problem_Extension_prompt.jinja2',
+        operation_name='problem_extension',
+        history=history,
+        questions=databasets,
+        json_schema=json_schema
+    )
+
+    try:
+        # 使用优化的LLM服务
+        response_content = await problem_service.call_llm_structured(
+            state=state,
+            db_session=db_session,
+            system_prompt=system_prompt,
+            response_model=ProblemExtensionResponse,
+            fallback_value=[]
+        )
+
+        logger.info(f"Problem_Extension: 开始处理问题扩展，问题数量: {len(databasets)}")
+
+        # 验证结构化响应
+        if not response_content or not hasattr(response_content, 'root'):
+            logger.warning("Problem_Extension: 结构化响应为空或格式不正确")
+            aggregated_dict = {}
+        elif not response_content.root:
+            logger.warning("Problem_Extension: 结构化响应的root为空")
+            aggregated_dict = {}
+        else:
+            # Aggregate results by original question
+            aggregated_dict = {}
+            for item in response_content.root:
+                try:
+                    key = getattr(item, "original_question", None) or (
+                        item.get("original_question") if isinstance(item, dict) else None
+                    )
+                    value = getattr(item, "extended_question", None) or (
+                        item.get("extended_question") if isinstance(item, dict) else None
+                    )
+                    if not key or not value:
+                        logger.warning(f"Problem_Extension: 跳过无效项: key={key}, value={value}")
+                        continue
+                    aggregated_dict.setdefault(key, []).append(value)
+                except Exception as item_error:
+                    logger.warning(f"Problem_Extension: 处理项目时出错: {item_error}")
+                    continue
+
+        logger.info(f"Problem_Extension: 成功生成 {len(aggregated_dict)} 个扩展问题组")
+
+    except Exception as e:
+        logger.error(
+            f"LLM call failed for Problem_Extension: {e}",
+            exc_info=True
+        )
+
+        # 提供更详细的错误信息
+        error_details = {
+            "error_type": type(e).__name__,
+            "error_message": str(e),
+            "questions_count": len(databasets),
+            "llm_model_id": memory_config.llm_model_id if memory_config else None
+        }
+
+        logger.error(f"Problem_Extension error details: {error_details}")
+        aggregated_dict = {}
+
+    logger.info("Problem extension")
+    logger.info(f"Problem extension result: {aggregated_dict}")
+
+    # Emit intermediate output for frontend
+    print(time.time() - start)
+    result = {
+        "context": aggregated_dict,
+        "original": data,
+        "storage_type": storage_type,
+        "user_rag_memory_id": user_rag_memory_id,
+        "_intermediate": {
+            "type": "problem_extension",
+            "title": "问题扩展",
+            "data": aggregated_dict,
+            "original_query": content,
+            "storage_type": storage_type,
+            "user_rag_memory_id": user_rag_memory_id
+        }
+    }
+
+    return {"problem_extension": result}

api/app/core/memory/agent/langgraph_graph/nodes/retrieve_nodes.py

@@ -0,0 +1,417 @@
+# ===== 标准库 =====
+import asyncio
+import json
+import os
+
+# ===== 第三方库 =====
+from langchain.agents import create_agent
+from langchain_openai import ChatOpenAI
+from app.core.logging_config import get_agent_logger
+from app.db import get_db, get_db_context
+
+from app.schemas import model_schema
+from app.services.memory_config_service import MemoryConfigService
+from app.services.model_service import ModelConfigService
+
+from app.core.memory.agent.services.search_service import SearchService
+from app.core.memory.agent.utils.llm_tools import (
+    COUNTState,
+    ReadState,
+    deduplicate_entries,
+    merge_to_key_value_pairs,
+)
+from app.core.memory.agent.langgraph_graph.tools.tool import (
+    create_hybrid_retrieval_tool_sync,
+    create_time_retrieval_tool,
+    extract_tool_message_content,
+)
+
+from app.core.rag.nlp.search import knowledge_retrieval
+
+logger = get_agent_logger(__name__)
+db = next(get_db())
+
+
+
+async def rag_config(state):
+    user_rag_memory_id = state.get('user_rag_memory_id', '')
+    kb_config = {
+        "knowledge_bases": [
+            {
+                "kb_id": user_rag_memory_id,
+                "similarity_threshold": 0.7,
+                "vector_similarity_weight": 0.5,
+                "top_k": 10,
+                "retrieve_type": "participle"
+            }
+        ],
+        "merge_strategy": "weight",
+        "reranker_id": os.getenv('reranker_id'),
+        "reranker_top_k": 10
+    }
+    return kb_config
+async def rag_knowledge(state,question):
+    kb_config = await rag_config(state)
+    end_user_id = state.get('end_user_id', '')
+    user_rag_memory_id=state.get("user_rag_memory_id",'')
+    retrieve_chunks_result = knowledge_retrieval(question, kb_config, [str(end_user_id)])
+    try:
+        retrieval_knowledge = [i.page_content for i in retrieve_chunks_result]
+        clean_content = '\n\n'.join(retrieval_knowledge)
+        cleaned_query = question
+        raw_results = clean_content
+        logger.info(f" Using RAG storage with memory_id={user_rag_memory_id}")
+    except  Exception :
+        retrieval_knowledge=[]
+        clean_content = ''
+        raw_results = ''
+        cleaned_query = question
+        logger.info(f"No content retrieved from knowledge base: {user_rag_memory_id}")
+    return retrieval_knowledge,clean_content,cleaned_query,raw_results
+
+
+async def llm_infomation(state: ReadState) -> ReadState:
+    memory_config = state.get('memory_config', None)
+    model_id = memory_config.llm_model_id
+    tenant_id = memory_config.tenant_id
+
+    # 使用现有的 memory_config 而不是重新查询数据库
+    # 或者使用线程安全的数据库访问
+    with get_db_context() as db:
+        result_orm = ModelConfigService.get_model_by_id(db=db, model_id=model_id, tenant_id=tenant_id)
+        result_pydantic = model_schema.ModelConfig.model_validate(result_orm)
+    return result_pydantic
+
+
+async def clean_databases(data) -> str:
+    """
+    简化的数据库搜索结果清理函数
+    
+    Args:
+        data: 搜索结果数据
+        
+    Returns:
+        清理后的内容字符串
+    """
+    try:
+        # 解析JSON字符串
+        if isinstance(data, str):
+            try:
+                data = json.loads(data)
+            except json.JSONDecodeError:
+                return data
+
+        if not isinstance(data, dict):
+            return str(data)
+
+        # 获取结果数据
+        # with open("搜索结果.json","w",encoding='utf-8') as f:
+        #     f.write(json.dumps(data, indent=4, ensure_ascii=False))
+        results = data.get('results', data)
+        if not isinstance(results, dict):
+            return str(results)
+
+        # 收集所有内容
+        content_list = []
+        
+        # 处理重排序结果
+        reranked = results.get('reranked_results', {})
+        if reranked:
+            for category in ['summaries', 'statements', 'chunks', 'entities']:
+                items = reranked.get(category, [])
+                if isinstance(items, list):
+                    content_list.extend(items)
+        # 处理时间搜索结果
+        time_search = results.get('time_search', {})
+        if time_search:
+            if isinstance(time_search, dict):
+                statements = time_search.get('statements', time_search.get('time_search', []))
+                if isinstance(statements, list):
+                    content_list.extend(statements)
+            elif isinstance(time_search, list):
+                content_list.extend(time_search)
+
+        # 提取文本内容
+        text_parts = []
+        for item in content_list:
+            if isinstance(item, dict):
+                text = item.get('statement') or item.get('content', '')
+                if text:
+                    text_parts.append(text)
+            elif isinstance(item, str):
+                text_parts.append(item)
+
+
+        return '\n'.join(text_parts).strip()
+
+    except Exception as e:
+        logger.error(f"clean_databases failed: {e}", exc_info=True)
+        return str(data)
+
+
+async def retrieve_nodes(state: ReadState) -> ReadState:
+
+    '''
+
+    模型信息
+    '''
+
+    problem_extension=state.get('problem_extension', '')['context']
+    storage_type=state.get('storage_type', '')
+    user_rag_memory_id=state.get('user_rag_memory_id', '')
+    end_user_id=state.get('end_user_id', '')
+    memory_config = state.get('memory_config', None)
+    original=state.get('data', '')
+    problem_list=[]
+    for key,values in problem_extension.items():
+        for data in values:
+            problem_list.append(data)
+    logger.info(f"Retrieve: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
+    # 创建异步任务处理单个问题
+    async def process_question_nodes(idx, question):
+        try:
+            # Prepare search parameters based on storage type
+            search_params = {
+                "end_user_id": end_user_id,
+                "question": question,
+                "return_raw_results": True
+            }
+            if storage_type == "rag" and user_rag_memory_id:
+                retrieval_knowledge, clean_content, cleaned_query, raw_results = await rag_knowledge(state, question)
+            else:
+                clean_content, cleaned_query, raw_results = await SearchService().execute_hybrid_search(
+                    **search_params, memory_config=memory_config
+                )
+
+            return {
+                "Query_small": cleaned_query,
+                "Result_small": clean_content,
+                "_intermediate": {
+                    "type": "search_result",
+                    "query": cleaned_query,
+                    "raw_results": raw_results,
+                    "index": idx + 1,
+                    "total": len(problem_list)
+                }
+            }
+
+        except Exception as e:
+            logger.error(
+                f"Retrieve: hybrid_search failed for question '{question}': {e}",
+                exc_info=True
+            )
+            # Return empty result for this question
+            return {
+                "Query_small": question,
+                "Result_small": "",
+                "_intermediate": {
+                    "type": "search_result",
+                    "query": question,
+                    "raw_results": [],
+                    "index": idx + 1,
+                    "total": len(problem_list)
+                }
+            }
+
+    # 并发处理所有问题
+    tasks = [process_question_nodes(idx, question) for idx, question in enumerate(problem_list)]
+    databases_anser = await asyncio.gather(*tasks)
+    databases_data = {
+        "Query": original,
+        "Expansion_issue": databases_anser
+    }
+
+    # Collect intermediate outputs before deduplication
+    intermediate_outputs = []
+    for item in databases_anser:
+        if '_intermediate' in item:
+            intermediate_outputs.append(item['_intermediate'])
+
+    # Deduplicate and merge results
+    deduplicated_data = deduplicate_entries(databases_data['Expansion_issue'])
+    deduplicated_data_merged = merge_to_key_value_pairs(
+        deduplicated_data,
+        'Query_small',
+        'Result_small'
+    )
+
+    # Restructure for Verify/Retrieve_Summary compatibility
+    keys, val = [], []
+    for item in deduplicated_data_merged:
+        for items_key, items_value in item.items():
+            keys.append(items_key)
+            val.append(items_value)
+
+    send_verify = []
+    for i, j in zip(keys, val, strict=False):
+        if j!=['']:
+            send_verify.append({
+                "Query_small": i,
+                "Answer_Small": j
+            })
+
+    dup_databases = {
+        "Query": original,
+        "Expansion_issue": send_verify,
+        "_intermediate_outputs": intermediate_outputs  # Preserve intermediate outputs
+    }
+
+    logger.info(f"Collected {len(intermediate_outputs)} intermediate outputs from search results")
+    return {'retrieve':dup_databases}
+
+
+
+
+async def retrieve(state: ReadState) -> ReadState:
+    # 从state中获取end_user_id
+    import time
+    start=time.time()
+    problem_extension = state.get('problem_extension', '')['context']
+    storage_type = state.get('storage_type', '')
+    user_rag_memory_id = state.get('user_rag_memory_id', '')
+    end_user_id = state.get('end_user_id', '')
+    memory_config = state.get('memory_config', None)
+    original = state.get('data', '')
+    problem_list = []
+    for key, values in problem_extension.items():
+        for data in values:
+            problem_list.append(data)
+    logger.info(f"Retrieve: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
+    databases_anser = []
+
+    async def get_llm_info():
+        with get_db_context() as db:  # 使用同步数据库上下文管理器
+            config_service = MemoryConfigService(db)
+            return await llm_infomation(state)
+    llm_config = await get_llm_info()
+    api_key_obj = llm_config.api_keys[0]
+    api_key = api_key_obj.api_key
+    api_base = api_key_obj.api_base
+    model_name = api_key_obj.model_name
+    llm = ChatOpenAI(
+        model=model_name,
+        api_key=api_key,
+        base_url=api_base,
+        temperature=0.2,
+    )
+
+    time_retrieval_tool = create_time_retrieval_tool(end_user_id)
+    search_params = { "end_user_id": end_user_id, "return_raw_results": True }
+    hybrid_retrieval=create_hybrid_retrieval_tool_sync(memory_config, **search_params)
+    agent = create_agent(
+        llm,
+        tools=[time_retrieval_tool,hybrid_retrieval],
+        system_prompt=f"我是检索专家，可以根据适合的工具进行检索。当前使用的end_user_id是: {end_user_id}"
+    )
+
+    # 创建异步任务处理单个问题
+    import asyncio
+
+    # 在模块级别定义信号量，限制最大并发数
+    SEMAPHORE = asyncio.Semaphore(5)  # 限制最多5个并发数据库操作
+
+    async def process_question(idx, question):
+        async with SEMAPHORE:  # 限制并发
+            try:
+                if storage_type == "rag" and user_rag_memory_id:
+                    retrieval_knowledge, clean_content, cleaned_query, raw_results = await rag_knowledge(state, question)
+                else:
+                    cleaned_query = question
+                    # 使用 asyncio 在线程池中运行同步的 agent.invoke
+                    import asyncio
+                    response = await asyncio.get_event_loop().run_in_executor(
+                        None,
+                        lambda: agent.invoke({"messages": question})
+                    )
+                    tool_results = extract_tool_message_content(response)
+                    if tool_results == None:
+                        raw_results = []
+                        clean_content = ''
+                    else:
+                        raw_results = tool_results['content']
+                        clean_content = await clean_databases(raw_results)
+
+                        try:
+                            raw_results = raw_results['results']
+                        except Exception:
+                            raw_results = []
+
+                return {
+                    "Query_small": cleaned_query,
+                    "Result_small": clean_content,
+                    "_intermediate": {
+                        "type": "search_result",
+                        "query": cleaned_query,
+                        "raw_results": raw_results,
+                        "index": idx + 1,
+                        "total": len(problem_list)
+                    }
+                }
+
+            except Exception as e:
+                logger.error(
+                    f"Retrieve: hybrid_search failed for question '{question}': {e}",
+                    exc_info=True
+                )
+                # Return empty result for this question
+                return {
+                    "Query_small": question,
+                    "Result_small": "",
+                    "_intermediate": {
+                        "type": "search_result",
+                        "query": question,
+                        "raw_results": [],
+                        "index": idx + 1,
+                        "total": len(problem_list)
+                    }
+                }
+
+    # 并发处理所有问题
+    import asyncio
+    tasks = [process_question(idx, question) for idx, question in enumerate(problem_list)]
+    databases_anser = await asyncio.gather(*tasks)
+    databases_data = {
+        "Query": original,
+        "Expansion_issue": databases_anser
+    }
+
+    # Collect intermediate outputs before deduplication
+    intermediate_outputs = []
+    for item in databases_anser:
+        if '_intermediate' in item:
+            intermediate_outputs.append(item['_intermediate'])
+
+    # Deduplicate and merge results
+    deduplicated_data = deduplicate_entries(databases_data['Expansion_issue'])
+    deduplicated_data_merged = merge_to_key_value_pairs(
+        deduplicated_data,
+        'Query_small',
+        'Result_small'
+    )
+
+    # Restructure for Verify/Retrieve_Summary compatibility
+    keys, val = [], []
+    for item in deduplicated_data_merged:
+        for items_key, items_value in item.items():
+            keys.append(items_key)
+            val.append(items_value)
+
+    send_verify = []
+    for i, j in zip(keys, val, strict=False):
+        if j != ['']:
+            send_verify.append({
+                "Query_small": i,
+                "Answer_Small": j
+            })
+
+    dup_databases = {
+        "Query": original,
+        "Expansion_issue": send_verify,
+        "_intermediate_outputs": intermediate_outputs  # Preserve intermediate outputs
+    }
+    # with open('retrieve_text.json', 'w') as f:
+    #     json.dump(dup_databases, f, indent=4)
+    logger.info(f"Collected {len(intermediate_outputs)} intermediate outputs from search results")
+    return {'retrieve': dup_databases}
+
+

api/app/core/memory/agent/langgraph_graph/nodes/summary_nodes.py

@@ -0,0 +1,320 @@
+
+
+import os
+import time
+
+from app.core.logging_config import get_agent_logger, log_time
+from app.core.memory.agent.models.summary_models import (
+    RetrieveSummaryResponse,
+    SummaryResponse,
+)
+from app.core.memory.agent.services.optimized_llm_service import LLMServiceMixin
+from app.core.memory.agent.services.search_service import SearchService
+from app.core.memory.agent.utils.llm_tools import (
+    PROJECT_ROOT_,
+    ReadState,
+)
+from app.core.memory.agent.utils.redis_tool import store
+from app.core.memory.agent.utils.session_tools import SessionService
+from app.core.memory.agent.utils.template_tools import TemplateService
+from app.db import get_db
+
+template_root = os.path.join(PROJECT_ROOT_, 'memory', 'agent', 'utils', 'prompt')
+logger = get_agent_logger(__name__)
+db_session = next(get_db())
+
+class SummaryNodeService(LLMServiceMixin):
+    """总结节点服务类"""
+    
+    def __init__(self):
+        super().__init__()
+        self.template_service = TemplateService(template_root)
+
+# 创建全局服务实例
+summary_service = SummaryNodeService()
+
+async def summary_history(state: ReadState) -> ReadState:
+    end_user_id = state.get("end_user_id", '')
+    history = await SessionService(store).get_history(end_user_id, end_user_id, end_user_id)
+    return history
+
+async def summary_llm(state: ReadState, history, retrieve_info, template_name, operation_name, response_model,search_mode) -> str:
+    """
+    增强的summary_llm函数，包含更好的错误处理和数据验证
+    """
+    data = state.get("data", '')
+    
+    # 构建系统提示词
+    if str(search_mode) == "0":
+        system_prompt = await summary_service.template_service.render_template(
+            template_name=template_name,
+            operation_name=operation_name,
+            data=retrieve_info,
+            query=data
+        )
+    else:
+        system_prompt = await summary_service.template_service.render_template(
+            template_name=template_name,
+            operation_name=operation_name,
+            query=data,
+            history=history,
+            retrieve_info=retrieve_info
+        )
+    try:
+        # 使用优化的LLM服务进行结构化输出
+        structured = await summary_service.call_llm_structured(
+            state=state,
+            db_session=db_session,
+            system_prompt=system_prompt,
+            response_model=response_model,
+            fallback_value=None
+        )
+        # 验证结构化响应
+        if structured is None:
+            logger.warning(f"LLM返回None，使用默认回答")
+            return "信息不足，无法回答"
+        
+        # 根据操作类型提取答案
+        if operation_name == "summary":
+            aimessages = getattr(structured, 'query_answer', None) or "信息不足，无法回答"
+        else:
+            # 处理RetrieveSummaryResponse
+            if hasattr(structured, 'data') and structured.data:
+                aimessages = getattr(structured.data, 'query_answer', None) or "信息不足，无法回答"
+            else:
+                logger.warning(f"结构化响应缺少data字段")
+                aimessages = "信息不足，无法回答"
+        
+        # 验证答案不为空
+        if not aimessages or aimessages.strip() == "":
+            aimessages = "信息不足，无法回答"
+            
+        return aimessages
+        
+    except Exception as e:
+        logger.error(f"结构化输出失败: {e}", exc_info=True)
+        
+        # 尝试非结构化输出作为fallback
+        try:
+            logger.info("尝试非结构化输出作为fallback")
+            response = await summary_service.call_llm_simple(
+                state=state,
+                db_session=db_session,
+                system_prompt=system_prompt,
+                fallback_message="信息不足，无法回答"
+            )
+            
+            if response and response.strip():
+                # 简单清理响应
+                cleaned_response = response.strip()
+                # 移除可能的JSON标记
+                if cleaned_response.startswith('```'):
+                    lines = cleaned_response.split('\n')
+                    cleaned_response = '\n'.join(lines[1:-1])
+                
+                return cleaned_response
+            else:
+                return "信息不足，无法回答"
+                
+        except Exception as fallback_error:
+            logger.error(f"Fallback也失败: {fallback_error}")
+            return "信息不足，无法回答"
+
+async def summary_redis_save(state: ReadState,aimessages) -> ReadState:
+    data = state.get("data", '')
+    end_user_id = state.get("end_user_id", '')
+    await SessionService(store).save_session(
+        user_id=end_user_id,
+        query=data,
+        apply_id=end_user_id,
+        end_user_id=end_user_id,
+        ai_response=aimessages
+    )
+    await SessionService(store).cleanup_duplicates()
+    logger.info(f"sessionid: {aimessages} 写入成功")
+async def summary_prompt(state: ReadState,aimessages,raw_results) -> ReadState:
+    storage_type=state.get("storage_type",'')
+    user_rag_memory_id=state.get("user_rag_memory_id",'')
+    data=state.get("data", '')
+    input_summary = {
+        "status": "success",
+        "summary_result": aimessages,
+        "storage_type": storage_type,
+        "user_rag_memory_id": user_rag_memory_id,
+        "_intermediate": {
+            "type": "input_summary",
+            "title": "快速答案",
+            "summary": aimessages,
+            "query": data,
+            "raw_results": raw_results,
+            "search_mode": "quick_search",
+            "storage_type": storage_type,
+            "user_rag_memory_id": user_rag_memory_id
+        }
+    }
+    retrieve={
+        "status": "success",
+        "summary_result": aimessages,
+        "storage_type": storage_type,
+        "user_rag_memory_id": user_rag_memory_id,
+        "_intermediate": {
+            "type": "retrieval_summary",
+            "title":"快速检索",
+            "summary": aimessages,
+            "query": data,
+            "storage_type": storage_type,
+            "user_rag_memory_id": user_rag_memory_id
+        }
+    }
+
+    return input_summary,retrieve
+
+async def Input_Summary(state: ReadState) -> ReadState:
+    start=time.time()
+    storage_type=state.get("storage_type",'')
+    memory_config = state.get('memory_config', None)
+    user_rag_memory_id=state.get("user_rag_memory_id",'')
+    data=state.get("data", '')
+    end_user_id=state.get("end_user_id", '')
+    logger.info(f"Input_Summary: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
+    history = await summary_history( state)
+    search_params = {
+        "end_user_id": end_user_id,
+        "question": data,
+        "return_raw_results": True,
+        "include": ["summaries"]  # Only search summary nodes for faster performance
+    }
+
+    try:
+        retrieve_info, question, raw_results = await SearchService().execute_hybrid_search(**search_params, memory_config=memory_config)
+    except Exception as e:
+        logger.error( f"Input_Summary: hybrid_search failed, using empty results: {e}",  exc_info=True )
+        retrieve_info, question, raw_results = "", data, []
+
+
+    try:
+        # aimessages=await summary_llm(state,history,retrieve_info,'Retrieve_Summary_prompt.jinja2',
+        #                              'input_summary',RetrieveSummaryResponse)
+        # logger.info(f"快速答案总结==>>:{storage_type}--{user_rag_memory_id}--{aimessages}")
+        summary_result = await summary_prompt(state, retrieve_info, retrieve_info)
+        summary = summary_result[0]
+    except Exception as e:
+        logger.error( f"Input_Summary failed: {e}", exc_info=True )
+        summary= {
+            "status": "fail",
+            "summary_result": "信息不足，无法回答",
+            "storage_type": storage_type,
+            "user_rag_memory_id": user_rag_memory_id,
+            "error": str(e)
+        }
+    end = time.time()
+    try:
+        duration = end - start
+    except Exception:
+        duration = 0.0
+    log_time('检索', duration)
+    return {"summary":summary}
+
+async def Retrieve_Summary(state: ReadState)-> ReadState:
+    retrieve=state.get("retrieve", '')
+    history = await summary_history( state)
+    import json
+    with open("检索.json","w",encoding='utf-8') as f:
+        f.write(json.dumps(retrieve, indent=4, ensure_ascii=False))
+    retrieve=retrieve.get("Expansion_issue", [])
+    start=time.time()
+    retrieve_info_str=[]
+    for data in retrieve:
+        if data=='':
+            retrieve_info_str=''
+        else:
+            for key, value in data.items():
+                if key=='Answer_Small':
+                    for i in value:
+                        retrieve_info_str.append(i)
+    retrieve_info_str=list(set(retrieve_info_str))
+    retrieve_info_str='\n'.join(retrieve_info_str)
+
+    aimessages=await  summary_llm(state,history,retrieve_info_str,
+                                  'direct_summary_prompt.jinja2','retrieve_summary',RetrieveSummaryResponse,"1")
+    if '信息不足，无法回答' not in str(aimessages) or str(aimessages) != "":
+        await summary_redis_save(state, aimessages)
+    if aimessages == '':
+        aimessages = '信息不足，无法回答'
+    logger.info(f"Summary after retrieval: {aimessages}")
+    end = time.time()
+    try:
+        duration = end - start
+    except Exception:
+        duration = 0.0
+    log_time('Retrieval summary', duration)
+    
+    # 修复协程调用 - 先await，然后访问返回值
+    summary_result = await summary_prompt(state, aimessages, retrieve_info_str)
+    summary = summary_result[1]
+    return {"summary":summary}
+
+
+async def Summary(state: ReadState)-> ReadState:
+    start=time.time()
+    query = state.get("data", '')
+    verify=state.get("verify", '')
+    verify_expansion_issue=verify.get("verified_data", '')
+    retrieve_info_str=''
+    for data in verify_expansion_issue:
+        for key, value in data.items():
+            if key=='answer_small':
+                for i in value:
+                    retrieve_info_str+=i+'\n'
+    history=await summary_history(state)
+
+    data = {
+        "query": query,
+        "history": history,
+        "retrieve_info": retrieve_info_str
+    }
+    aimessages=await  summary_llm(state,history,data,
+                                  'summary_prompt.jinja2','summary',SummaryResponse,0)
+
+    if '信息不足，无法回答' not in str(aimessages) or str(aimessages) != "":
+        await summary_redis_save(state, aimessages)
+    if aimessages == '':
+        aimessages = '信息不足，无法回答'
+    try:
+        duration = time.time() - start
+    except Exception:
+        duration = 0.0
+    log_time('Retrieval summary', duration)
+
+    # 修复协程调用 - 先await，然后访问返回值
+    summary_result = await summary_prompt(state, aimessages, retrieve_info_str)
+    summary = summary_result[1]
+    return {"summary":summary}
+
+async def Summary_fails(state: ReadState)-> ReadState:
+    storage_type=state.get("storage_type", '')
+    user_rag_memory_id=state.get("user_rag_memory_id", '')
+    history = await summary_history(state)
+    query = state.get("data", '')
+    verify = state.get("verify", '')
+    verify_expansion_issue = verify.get("verified_data", '')
+    retrieve_info_str = ''
+    for data in verify_expansion_issue:
+        for key, value in data.items():
+            if key == 'answer_small':
+                for i in value:
+                    retrieve_info_str += i + '\n'
+    data = {
+        "query": query,
+        "history": history,
+        "retrieve_info": retrieve_info_str
+    }
+    aimessages = await  summary_llm(state, history, data,
+                                 'fail_summary_prompt.jinja2', 'summary', SummaryResponse, 0)
+    result= {
+        "status": "success",
+        "summary_result": aimessages,
+        "storage_type": storage_type,
+        "user_rag_memory_id": user_rag_memory_id
+    }
+    return {"summary":result}

api/app/core/memory/agent/langgraph_graph/nodes/tool_node.py

@@ -1,234 +0,0 @@
-"""
-Tool execution node for LangGraph workflow.
-
-This module provides the ToolExecutionNode class which wraps tool execution
-with parameter transformation logic using the ParameterBuilder service.
-"""
-
-import logging
-import time
-from typing import Any, Callable, Dict
-
-from app.core.memory.agent.langgraph_graph.state.extractors import (
-    extract_content_payload,
-    extract_tool_call_id,
-)
-from app.core.memory.agent.mcp_server.services.parameter_builder import ParameterBuilder
-from app.schemas.memory_config_schema import MemoryConfig
-from langchain_core.messages import AIMessage
-from langgraph.prebuilt import ToolNode
-
-logger = logging.getLogger(__name__)
-
-
-class ToolExecutionNode:
-    """
-    Custom LangGraph node that wraps tool execution with parameter transformation.
-    
-    This node extracts content from previous tool results, transforms parameters
-    based on tool type using ParameterBuilder, and invokes the tool with the
-    correct argument structure.
-    
-    Attributes:
-        tool_node: LangGraph ToolNode wrapping the actual tool
-        id: Node identifier for message IDs
-        tool_name: Name of the tool being executed
-        namespace: Namespace for session management
-        search_switch: Search routing parameter
-        apply_id: Application identifier
-        group_id: Group identifier
-        parameter_builder: Service for building tool-specific arguments
-        memory_config: MemoryConfig object containing all configuration
-    """
-
-    def __init__(
-        self,
-        tool: Callable,
-        node_id: str,
-        namespace: str,
-        search_switch: str,
-        apply_id: str,
-        group_id: str,
-        parameter_builder: ParameterBuilder,
-        storage_type: str,
-        user_rag_memory_id: str,
-        memory_config: MemoryConfig,
-    ):
-        """
-        Initialize the tool execution node.
-        
-        Args:
-            tool: The tool function to execute
-            node_id: Identifier for this node (used in message IDs)
-            namespace: Namespace for session management
-            search_switch: Search routing parameter
-            apply_id: Application identifier
-            group_id: Group identifier
-            parameter_builder: Service for building tool-specific arguments
-            storage_type: Storage type for the workspace
-            user_rag_memory_id: User RAG memory identifier
-            memory_config: MemoryConfig object containing all configuration
-        """
-        self.tool_node = ToolNode([tool])
-        self.id = node_id
-        self.tool_name = tool.name if hasattr(tool, 'name') else str(tool)
-        self.namespace = namespace
-        self.search_switch = search_switch
-        self.apply_id = apply_id
-        self.group_id = group_id
-        self.parameter_builder = parameter_builder
-        self.storage_type = storage_type
-        self.user_rag_memory_id = user_rag_memory_id
-        self.memory_config = memory_config
-
-        logger.info(
-            f"[ToolExecutionNode] Initialized node '{self.id}' for tool '{self.tool_name}'"
-        )
-    
-    async def __call__(self, state: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        Execute the tool with transformed parameters.
-        
-        This method:
-        1. Extracts the last message from state
-        2. Extracts tool call ID using state extractors
-        3. Extracts content payload using state extractors
-        4. Builds tool arguments using parameter builder
-        5. Constructs AIMessage with tool_calls
-        6. Invokes the tool and returns the result
-        
-        Args:
-            state: LangGraph state dictionary
-            
-        Returns:
-            Updated state with tool result in messages
-        """
-        messages = state.get("messages", [])
-        logger.debug( self.tool_name)
-        
-        if not messages:
-            logger.warning(f"[ToolExecutionNode] {self.id} - No messages in state")
-            return {"messages": [AIMessage(content="Error: No messages in state")]}
-        
-        last_message = messages[-1]
-        logger.debug(
-            f"[ToolExecutionNode] {self.id} - Processing message at {time.time()}"
-        )
-        
-        try:
-            # Extract tool call ID using state extractors
-            tool_call_id = extract_tool_call_id(last_message)
-            logger.debug(f"[ToolExecutionNode] {self.id} - Extracted tool_call_id: {tool_call_id}")
-            
-        except ValueError as e:
-            logger.error(
-                f"[ToolExecutionNode] {self.id} - Failed to extract tool call ID: {e}"
-            )
-            return {"messages": [AIMessage(content=f"Error: {str(e)}")]}
-        
-        try:
-            # Extract content payload using state extractors
-            content = extract_content_payload(last_message)
-            logger.debug(
-                f"[ToolExecutionNode] {self.id} - Extracted content type: {type(content)}, content_keys: {list(content.keys()) if isinstance(content, dict) else 'N/A'}"
-            )
-            # Log raw message content for debugging
-            if hasattr(last_message, 'content'):
-                raw = last_message.content
-                logger.debug(f"[ToolExecutionNode] {self.id} - Raw message content (first 500 chars): {str(raw)[:500]}")
-            
-        except Exception as e:
-            logger.error(
-                f"[ToolExecutionNode] {self.id} - Failed to extract content: {e}",
-                exc_info=True
-            )
-            content = {}
-        
-        try:
-            # Build tool arguments using parameter builder
-            tool_args = self.parameter_builder.build_tool_args(
-                tool_name=self.tool_name,
-                content=content,
-                tool_call_id=tool_call_id,
-                search_switch=self.search_switch,
-                apply_id=self.apply_id,
-                group_id=self.group_id,
-                memory_config=self.memory_config,
-                storage_type=self.storage_type,
-                user_rag_memory_id=self.user_rag_memory_id,
-            )
-            logger.debug(
-                f"[ToolExecutionNode] {self.id} - Built tool args with keys: {list(tool_args.keys())}"
-            )
-            
-        except Exception as e:
-            logger.error(
-                f"[ToolExecutionNode] {self.id} - Failed to build tool args: {e}",
-                exc_info=True
-            )
-            return {"messages": [AIMessage(content=f"Error building arguments: {str(e)}")]}
-        
-        # Construct tool input message
-        tool_input = {
-            "messages": [
-                AIMessage(
-                    content="",
-                    tool_calls=[{
-                        "name": self.tool_name,
-                        "args": tool_args,
-                        "id": f"{self.id}_{tool_call_id}",
-                    }]
-                )
-            ]
-        }
-        
-        try:
-            # Invoke the tool
-            result = await self.tool_node.ainvoke(tool_input)
-            
-            logger.debug(
-                f"[ToolExecutionNode] {self.id} - Tool execution completed"
-            )
-            
-            # Check for error in tool response
-            error_entry = None
-            if result and "messages" in result:
-                for msg in result["messages"]:
-                    if hasattr(msg, 'content'):
-                        try:
-                            import json
-                            content = msg.content
-                            if isinstance(content, str):
-                                parsed = json.loads(content)
-                                if isinstance(parsed, dict) and "error" in parsed:
-                                    error_msg = parsed["error"]
-                                    logger.warning(
-                                        f"[ToolExecutionNode] {self.id} - Tool returned error: {error_msg}"
-                                    )
-                                    error_entry = {"tool": self.tool_name, "error": error_msg, "node_id": self.id}
-                        except (json.JSONDecodeError, TypeError):
-                            pass
-            
-            # Return result with error tracking if error was found
-            if error_entry:
-                result["errors"] = [error_entry]
-            
-            return result
-            
-        except Exception as e:
-            logger.error(
-                f"[ToolExecutionNode] {self.id} - Tool execution failed: {e}",
-                exc_info=True
-            )
-            # Track error in state and return error message
-            from langchain_core.messages import ToolMessage
-            error_entry = {"tool": self.tool_name, "error": str(e), "node_id": self.id}
-            return {
-                "messages": [
-                    ToolMessage(
-                        content=f"Error executing tool: {str(e)}",
-                        tool_call_id=f"{self.id}_{tool_call_id}"
-                    )
-                ],
-                "errors": [error_entry]
-            }

api/app/core/memory/agent/langgraph_graph/nodes/verification_nodes.py

@@ -0,0 +1,155 @@
+import os
+from app.core.logging_config import get_agent_logger
+from app.db import get_db
+
+from app.core.memory.agent.models.verification_models import VerificationResult
+from app.core.memory.agent.utils.llm_tools import (
+    PROJECT_ROOT_,
+    ReadState,
+)
+from app.core.memory.agent.utils.redis_tool import store
+from app.core.memory.agent.utils.session_tools import SessionService
+from app.core.memory.agent.utils.template_tools import TemplateService
+from app.core.memory.agent.services.optimized_llm_service import LLMServiceMixin
+
+template_root = os.path.join(PROJECT_ROOT_, 'memory', 'agent', 'utils', 'prompt')
+db_session = next(get_db())
+logger = get_agent_logger(__name__)
+
+class VerificationNodeService(LLMServiceMixin):
+    """验证节点服务类"""
+    
+    def __init__(self):
+        super().__init__()
+        self.template_service = TemplateService(template_root)
+
+# 创建全局服务实例
+verification_service = VerificationNodeService()
+
+async def Verify_prompt(state: ReadState, messages_deal: VerificationResult):
+    """处理验证结果并生成输出格式"""
+    storage_type = state.get('storage_type', '')
+    user_rag_memory_id = state.get('user_rag_memory_id', '')
+    data = state.get('data', '')
+    
+    # 将 VerificationItem 对象转换为字典列表
+    verified_data = []
+    if messages_deal.expansion_issue:
+        for item in messages_deal.expansion_issue:
+            if hasattr(item, 'model_dump'):
+                verified_data.append(item.model_dump())
+            elif isinstance(item, dict):
+                verified_data.append(item)
+    
+    Verify_result = {
+        "status": messages_deal.split_result,
+        "verified_data": verified_data,
+        "storage_type": storage_type,
+        "user_rag_memory_id": user_rag_memory_id,
+        "_intermediate": {
+            "type": "verification",
+            "title": "Data Verification",
+            "result": messages_deal.split_result,
+            "reason": messages_deal.reason or "验证完成",
+            "query": messages_deal.query,
+            "verified_count": len(verified_data),
+            "storage_type": storage_type,
+            "user_rag_memory_id": user_rag_memory_id
+        }
+    }
+    return Verify_result
+async def Verify(state: ReadState):
+    logger.info("=== Verify 节点开始执行 ===")
+    try:
+        content = state.get('data', '')
+        end_user_id = state.get('end_user_id', '')
+        memory_config = state.get('memory_config', None)
+        
+        logger.info(f"Verify: content={content[:50] if content else 'empty'}..., end_user_id={end_user_id}")
+
+        history = await SessionService(store).get_history(end_user_id, end_user_id, end_user_id)
+        logger.info(f"Verify: 获取历史记录完成，history length={len(history)}")
+
+        retrieve = state.get("retrieve", {})
+        logger.info(f"Verify: retrieve data type={type(retrieve)}, keys={retrieve.keys() if isinstance(retrieve, dict) else 'N/A'}")
+        
+        retrieve_expansion = retrieve.get("Expansion_issue", []) if isinstance(retrieve, dict) else []
+        logger.info(f"Verify: Expansion_issue length={len(retrieve_expansion)}")
+        
+        messages = {
+            "Query": content,
+            "Expansion_issue": retrieve_expansion
+        }
+
+        logger.info("Verify: 开始渲染模板")
+        
+        # 生成 JSON schema 以指导 LLM 输出正确格式
+        json_schema = VerificationResult.model_json_schema()
+        
+        system_prompt = await verification_service.template_service.render_template(
+            template_name='split_verify_prompt.jinja2',
+            operation_name='split_verify_prompt',
+            history=history,
+            sentence=messages,
+            json_schema=json_schema
+        )
+        logger.info(f"Verify: 模板渲染完成，prompt length={len(system_prompt)}")
+        
+        # 使用优化的LLM服务，添加超时保护
+        logger.info("Verify: 开始调用 LLM")
+        try:
+            # 添加 asyncio.wait_for 超时包裹，防止无限等待
+            # 超时时间设置为 150 秒（比 LLM 配置的 120 秒稍长）
+            import asyncio
+            structured = await asyncio.wait_for(
+                verification_service.call_llm_structured(
+                    state=state,
+                    db_session=db_session,
+                    system_prompt=system_prompt,
+                    response_model=VerificationResult,
+                    fallback_value={
+                        "query": content,
+                        "history": history if isinstance(history, list) else [],
+                        "expansion_issue": [],
+                        "split_result": "failed",
+                        "reason": "验证失败或超时"
+                    }
+                ),
+                timeout=150.0  # 150秒超时
+            )
+            logger.info(f"Verify: LLM 调用完成，result={structured}")
+        except asyncio.TimeoutError:
+            logger.error("Verify: LLM 调用超时（150秒），使用 fallback 值")
+            structured = VerificationResult(
+                query=content,
+                history=history if isinstance(history, list) else [],
+                expansion_issue=[],
+                split_result="failed",
+                reason="LLM调用超时"
+            )
+        
+        result = await Verify_prompt(state, structured)
+        logger.info("=== Verify 节点执行完成 ===")
+        return {"verify": result}
+        
+    except Exception as e:
+        logger.error(f"Verify 节点执行失败: {e}", exc_info=True)
+        # 返回失败的验证结果
+        return {
+            "verify": {
+                "status": "failed",
+                "verified_data": [],
+                "storage_type": state.get('storage_type', ''),
+                "user_rag_memory_id": state.get('user_rag_memory_id', ''),
+                "_intermediate": {
+                    "type": "verification",
+                    "title": "Data Verification",
+                    "result": "failed",
+                    "reason": f"验证过程出错: {str(e)}",
+                    "query": state.get('data', ''),
+                    "verified_count": 0,
+                    "storage_type": state.get('storage_type', ''),
+                    "user_rag_memory_id": state.get('user_rag_memory_id', '')
+                }
+            }
+        }

api/app/core/memory/agent/langgraph_graph/nodes/write_nodes.py

@@ -0,0 +1,55 @@
+from app.core.memory.agent.utils.llm_tools import WriteState
+from app.core.memory.agent.utils.write_tools import write
+from app.core.logging_config import get_agent_logger
+
+logger = get_agent_logger(__name__)
+
+
+async def write_node(state: WriteState) -> WriteState:
+    """
+        Write data to the database/file system.
+
+        Args:
+            state: WriteState containing messages, end_user_id, and memory_config
+
+        Returns:
+            dict: Contains 'write_result' with status and data fields
+        """
+    messages = state.get('messages', [])
+    end_user_id = state.get('end_user_id', '')
+    memory_config = state.get('memory_config', '')
+
+    # Convert LangChain messages to structured format expected by write()
+    structured_messages = []
+    for msg in messages:
+        if hasattr(msg, 'type') and hasattr(msg, 'content'):
+            # Map LangChain message types to role names
+            role = 'user' if msg.type == 'human' else 'assistant' if msg.type == 'ai' else msg.type
+            structured_messages.append({
+                "role": role,
+                "content": msg.content  # content is now guaranteed to be a string
+            })
+
+    try:
+        result = await write(
+            messages=structured_messages,
+            end_user_id=end_user_id,
+            memory_config=memory_config,
+        )
+        logger.info(f"Write completed successfully! Config: {memory_config.config_name}")
+
+        write_result = {
+            "status": "success",
+            "data": structured_messages,
+            "config_id": memory_config.config_id,
+            "config_name": memory_config.config_name,
+        }
+        return {"write_result": write_result}
+
+    except Exception as e:
+        logger.error(f"Data_write failed: {e}", exc_info=True)
+        write_result = {
+            "status": "error",
+            "message": str(e),
+        }
+        return {"write_result": write_result}

api/app/core/memory/agent/langgraph_graph/read_graph.py

@@ -1,469 +1,177 @@
-import json
-import os
-import re
-import time
-import warnings
+#!/usr/bin/env python3
 from contextlib import asynccontextmanager
-from typing import Literal
 
-from app.core.logging_config import get_agent_logger
-from app.core.memory.agent.langgraph_graph.nodes import (
-    ToolExecutionNode,
-    create_input_message,
-)
-from app.core.memory.agent.mcp_server.services.parameter_builder import ParameterBuilder
-from app.core.memory.agent.utils.llm_tools import COUNTState, ReadState
-from app.core.memory.agent.utils.multimodal import MultimodalProcessor
-from app.schemas.memory_config_schema import MemoryConfig
-from dotenv import load_dotenv
-from langchain_core.messages import AIMessage
-from langgraph.checkpoint.memory import InMemorySaver
-from langgraph.constants import END, START
+from langchain_core.messages import HumanMessage
+from langgraph.constants import START, END
 from langgraph.graph import StateGraph
-from langgraph.prebuilt import ToolNode
-
-logger = get_agent_logger(__name__)
-
-warnings.filterwarnings("ignore", category=RuntimeWarning)
-load_dotenv()
-redishost=os.getenv("REDISHOST")
-redisport=os.getenv('REDISPORT')
-redisdb=os.getenv('REDISDB')
-redispassword=os.getenv('REDISPASSWORD')
-counter = COUNTState(limit=3)
-
-# Update loop count in workflow
-async def update_loop_count(state):
-    """Update loop counter"""
-    current_count = state.get("loop_count", 0)
-    return {"loop_count": current_count + 1}
 
 
-def Verify_continue(state: ReadState) -> Literal["Summary", "Summary_fails", "content_input"]:
-    messages = state["messages"]
+from app.db import get_db
+from app.services.memory_config_service import MemoryConfigService
 
-    # Add boundary check
-    if not messages:
-        return END
-    counter.add(1)  # Increment by 1
+from app.core.memory.agent.utils.llm_tools import ReadState
+from app.core.memory.agent.langgraph_graph.nodes.data_nodes import content_input_node
+from app.core.memory.agent.langgraph_graph.nodes.problem_nodes import (
+    Split_The_Problem,
+    Problem_Extension,
+)
+from app.core.memory.agent.langgraph_graph.nodes.retrieve_nodes import (
+    retrieve,
+)
+from app.core.memory.agent.langgraph_graph.nodes.summary_nodes import (
+    Input_Summary,
+    Retrieve_Summary,
+    Summary_fails,
+    Summary,
+)
+from app.core.memory.agent.langgraph_graph.nodes.verification_nodes import Verify
+from app.core.memory.agent.langgraph_graph.routing.routers import (
+    Split_continue,
+    Retrieve_continue,
+    Verify_continue,
+)
 
-    loop_count = counter.get_total()
-    logger.debug(f"[should_continue] Current loop count: {loop_count}")
-
-    last_message = messages[-1]
-    last_message_str = str(last_message).replace('\\', '')
-    status_tools = re.findall(r'"split_result": "(.*?)"', last_message_str)
-    logger.debug(f"Status tools: {status_tools}")
-
-    if "success" in status_tools:
-        counter.reset()
-        return "Summary"
-    elif "failed" in status_tools:
-        if loop_count < 2:  # Maximum loop count is 3
-            return "content_input"
-        else:
-            counter.reset()
-            return "Summary_fails"
-    else:
-        # Add default return value to avoid returning None
-        counter.reset()
-        return "Summary"  # Default based on business requirements
-
-
-def Retrieve_continue(state) -> Literal["Verify", "Retrieve_Summary"]:
-    """
-    Determine routing based on search_switch value.
-
-    Args:
-        state: State dictionary containing search_switch
-
-    Returns:
-        Next node to execute
-    """
-    # Direct dictionary access instead of regex parsing
-    search_switch = state.get("search_switch")
-
-    # Handle case where search_switch might be in messages
-    if search_switch is None and "messages" in state:
-        messages = state.get("messages", [])
-        if messages:
-            last_message = messages[-1]
-            # Try to extract from tool_calls args
-            if hasattr(last_message, "tool_calls") and last_message.tool_calls:
-                for tool_call in last_message.tool_calls:
-                    if isinstance(tool_call, dict) and "args" in tool_call:
-                        search_switch = tool_call["args"].get("search_switch")
-                        break
-
-    # Convert to string for comparison if needed
-    if search_switch is not None:
-        search_switch = str(search_switch)
-        if search_switch == '0':
-            return 'Verify'
-        elif search_switch == '1':
-            return 'Retrieve_Summary'
-
-    # Add default return value to avoid returning None
-    return 'Retrieve_Summary'  # Default based on business logic
-
-
-def Split_continue(state) -> Literal["Split_The_Problem", "Input_Summary"]:
-    """
-    Determine routing based on search_switch value.
-
-    Args:
-        state: State dictionary containing search_switch
-
-    Returns:
-        Next node to execute
-    """
-    logger.debug(f"Split_continue state: {state}")
-
-    # Direct dictionary access instead of regex parsing
-    search_switch = state.get("search_switch")
-
-    # Handle case where search_switch might be in messages
-    if search_switch is None and "messages" in state:
-        messages = state.get("messages", [])
-        if messages:
-            last_message = messages[-1]
-            # Try to extract from tool_calls args
-            if hasattr(last_message, "tool_calls") and last_message.tool_calls:
-                for tool_call in last_message.tool_calls:
-                    if isinstance(tool_call, dict) and "args" in tool_call:
-                        search_switch = tool_call["args"].get("search_switch")
-                        break
-
-    # Convert to string for comparison if needed
-    if search_switch is not None:
-        search_switch = str(search_switch)
-        if search_switch == '2':
-            return 'Input_Summary'
-    return 'Split_The_Problem'  # Default case
-
-
-class ProblemExtensionNode:
-    def __init__(self, tool, id, namespace, search_switch, apply_id, group_id, storage_type="", user_rag_memory_id=""):
-        self.tool_node = ToolNode([tool])
-        self.id = id
-        self.tool_name = tool.name if hasattr(tool, 'name') else str(tool)
-        self.namespace = namespace
-        self.search_switch = search_switch
-        self.apply_id = apply_id
-        self.group_id = group_id
-        self.storage_type = storage_type
-        self.user_rag_memory_id = user_rag_memory_id
-
-    async def __call__(self, state):
-        messages = state["messages"]
-        last_message = messages[-1] if messages else ""
-        logger.debug(f"ProblemExtensionNode {self.id} - Current time: {time.time()} - Message: {last_message}")
-        if self.tool_name == 'Input_Summary':
-            tool_call = re.findall("'id': '(.*?)'", str(last_message))[0]
-        else:
-            tool_call = str(re.findall(r"tool_call_id=.*?'(.*?)'", str(last_message))[0]).replace('\\', '').split('_id')[1]
-        
-        # Try to extract actual content payload from previous tool result
-        raw_msg = last_message.content if hasattr(last_message, 'content') else str(last_message)
-        extracted_payload = None
-        # Capture ToolMessage content field (supports single/double quotes), avoid greedy matching
-        m = re.search(r"content=(?:\"|\')(.*?)(?:\"|\'),\s*name=", raw_msg, flags=re.S)
-        if m:
-            extracted_payload = m.group(1)
-        else:
-            # Fallback: use raw string directly
-            extracted_payload = raw_msg
-
-        # Try to parse content as JSON first
-        try:
-            content = json.loads(extracted_payload)
-        except Exception:
-            # Try to extract JSON fragment from text and parse
-            parsed = None
-            candidates = re.findall(r"[\[{].*[\]}]", extracted_payload, flags=re.S)
-            for cand in candidates:
-                try:
-                    parsed = json.loads(cand)
-                    break
-                except Exception:
-                    continue
-            # If still fails, use raw string as content
-            content = parsed if parsed is not None else extracted_payload
-
-        # Build correct parameters based on tool name
-        tool_args = {}
-
-        if self.tool_name == "Verify":
-            # Verify tool requires context and usermessages parameters
-            if isinstance(content, dict):
-                tool_args["context"] = content
-            else:
-                tool_args["context"] = {"content": content}
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-        elif self.tool_name == "Retrieve":
-            # Retrieve tool requires context and usermessages parameters
-            if isinstance(content, dict):
-                tool_args["context"] = content
-            else:
-                tool_args["context"] = {"content": content}
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["search_switch"] = str(self.search_switch)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-        elif self.tool_name == "Summary":
-            # Summary tool requires string type context parameter
-            if isinstance(content, dict):
-                # Convert dict to JSON string
-                tool_args["context"] = json.dumps(content, ensure_ascii=False)
-            else:
-                tool_args["context"] = str(content)
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-        elif self.tool_name == "Summary_fails":
-            # Summary_fails tool requires string type context parameter
-            if isinstance(content, dict):
-                # Convert dict to JSON string
-                tool_args["context"] = json.dumps(content, ensure_ascii=False)
-            else:
-                tool_args["context"] = str(content)
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-        elif self.tool_name == 'Input_Summary':
-            tool_args["context"] = str(last_message)
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["search_switch"] = str(self.search_switch)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-            tool_args["storage_type"] = getattr(self, 'storage_type', "")
-            tool_args["user_rag_memory_id"] = getattr(self, 'user_rag_memory_id', "")
-        elif self.tool_name == 'Retrieve_Summary':
-            # Retrieve_Summary expects dict directly, not JSON string
-            # content might be a JSON string, try to parse it
-            if isinstance(content, str):
-                try:
-                    parsed_content = json.loads(content)
-                    # Check if it has a "context" key
-                    if isinstance(parsed_content, dict) and "context" in parsed_content:
-                        tool_args["context"] = parsed_content["context"]
-                    else:
-                        tool_args["context"] = parsed_content
-                except json.JSONDecodeError:
-                    # If parsing fails, wrap the string
-                    tool_args["context"] = {"content": content}
-            elif isinstance(content, dict):
-                # Check if content has a "context" key that needs unwrapping
-                if "context" in content:
-                    tool_args["context"] = content["context"]
-                else:
-                    tool_args["context"] = content
-            else:
-                tool_args["context"] = {"content": str(content)}
-
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-        else:
-            # Other tools use context parameter
-            if isinstance(content, dict):
-                tool_args["context"] = content
-            else:
-                tool_args["context"] = {"content": content}
-            tool_args["usermessages"] = str(tool_call)
-            tool_args["apply_id"] = str(self.apply_id)
-            tool_args["group_id"] = str(self.group_id)
-
-
-        tool_input = {
-            "messages": [
-                AIMessage(
-                    content="",
-                    tool_calls=[{
-                        "name": self.tool_name,
-                        "args": tool_args,
-                        "id": self.id + f"{tool_call}",
-                    }]
-                )
-            ]
-        }
-        result = await self.tool_node.ainvoke(tool_input)
-        result_text = str(result)
-
-        return {"messages": [AIMessage(content=result_text)]}
 
 
 @asynccontextmanager
-async def make_read_graph(namespace, tools, search_switch, apply_id, group_id, memory_config: MemoryConfig, storage_type=None, user_rag_memory_id=None):
-    """
-    Create a read graph workflow for memory operations.
-    
-    Args:
-        namespace: Namespace identifier
-        tools: MCP tools loaded from session
-        search_switch: Search mode switch ("0", "1", or "2")
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type (optional)
-        user_rag_memory_id: User RAG memory ID (optional)
-    """
-    memory = InMemorySaver()
-    tool = [i.name for i in tools]
-    logger.info(f"Initializing read graph with tools: {tool}")
-    logger.info(f"Using memory_config: {memory_config.config_name} (id={memory_config.config_id})")
-    
-    # Extract tool functions
-    Split_The_Problem_ = next((t for t in tools if t.name == "Split_The_Problem"), None)
-    Problem_Extension_ = next((t for t in tools if t.name == "Problem_Extension"), None)
-    Retrieve_ = next((t for t in tools if t.name == "Retrieve"), None)
-    Verify_ = next((t for t in tools if t.name == "Verify"), None)
-    Summary_ = next((t for t in tools if t.name == "Summary"), None)
-    Summary_fails_ = next((t for t in tools if t.name == "Summary_fails"), None)
-    Retrieve_Summary_ = next((t for t in tools if t.name == "Retrieve_Summary"), None)
-    Input_Summary_ = next((t for t in tools if t.name == "Input_Summary"), None)
-    
-    # Instantiate services
-    parameter_builder = ParameterBuilder()
-    multimodal_processor = MultimodalProcessor()
-    
-    # Create nodes using new modular components
-    Split_The_Problem_node = ToolNode([Split_The_Problem_])
-    
-    Problem_Extension_node = ToolExecutionNode(
-        tool=Problem_Extension_,
-        node_id="Problem_Extension_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
+async def make_read_graph():
+    """创建并返回 LangGraph 工作流"""
+    try:
+        # Build workflow graph
+        workflow = StateGraph(ReadState)
+        workflow.add_node("content_input", content_input_node)
+        workflow.add_node("Split_The_Problem", Split_The_Problem)
+        workflow.add_node("Problem_Extension", Problem_Extension)
+        workflow.add_node("Input_Summary", Input_Summary)
+        # workflow.add_node("Retrieve", retrieve_nodes)
+        workflow.add_node("Retrieve", retrieve)
+        workflow.add_node("Verify", Verify)
+        workflow.add_node("Retrieve_Summary", Retrieve_Summary)
+        workflow.add_node("Summary", Summary)
+        workflow.add_node("Summary_fails", Summary_fails)
+        
+        # 添加边
+        workflow.add_edge(START, "content_input")
+        workflow.add_conditional_edges("content_input", Split_continue)
+        workflow.add_edge("Input_Summary", END)
+        workflow.add_edge("Split_The_Problem", "Problem_Extension")
+        workflow.add_edge("Problem_Extension", "Retrieve")
+        workflow.add_conditional_edges("Retrieve", Retrieve_continue)
+        workflow.add_edge("Retrieve_Summary", END)
+        workflow.add_conditional_edges("Verify", Verify_continue)
+        workflow.add_edge("Summary_fails", END)
+        workflow.add_edge("Summary", END)
+
+
+        '''-----'''
+        # workflow.add_edge("Retrieve", END)
+        
+        # 编译工作流
+        graph = workflow.compile()
+        yield graph
+        
+    except Exception as e:
+        print(f"创建工作流失败: {e}")
+        raise
+    finally:
+        print("工作流创建完成")
+
+async def main():
+    """主函数 - 运行工作流"""
+    message = "昨天有什么好看的电影"
+    end_user_id = '88a459f5_text09'  # 组ID
+    storage_type = 'neo4j'  # 存储类型
+    search_switch = '1'  # 搜索开关
+    user_rag_memory_id = 'wwwwwwww'  # 用户RAG记忆ID
+
+    # 获取数据库会话
+    db_session = next(get_db())
+    config_service = MemoryConfigService(db_session)
+    memory_config = config_service.load_memory_config(
+        config_id=17,  # 改为整数
+        service_name="MemoryAgentService"
     )
+    import time
+    start=time.time()
+    try:
+        async with make_read_graph() as graph:
+            config = {"configurable": {"thread_id": end_user_id}}
+            # 初始状态 - 包含所有必要字段
+            initial_state = {"messages": [HumanMessage(content=message)] ,"search_switch":search_switch,"end_user_id":end_user_id
+                             ,"storage_type":storage_type,"user_rag_memory_id":user_rag_memory_id,"memory_config":memory_config}
+            # 获取节点更新信息
+            _intermediate_outputs = []
+            summary = ''
+            
+            async for update_event in graph.astream(
+                    initial_state,
+                    stream_mode="updates",
+                    config=config
+            ):
+                for node_name, node_data in update_event.items():
+                    print(f"处理节点: {node_name}")
+                    
+                    # 处理不同Summary节点的返回结构
+                    if 'Summary' in node_name:
+                        if 'InputSummary' in node_data and 'summary_result' in node_data['InputSummary']:
+                            summary = node_data['InputSummary']['summary_result']
+                        elif 'RetrieveSummary' in node_data and 'summary_result' in node_data['RetrieveSummary']:
+                            summary = node_data['RetrieveSummary']['summary_result']
+                        elif 'summary' in node_data and 'summary_result' in node_data['summary']:
+                            summary = node_data['summary']['summary_result']
+                        elif 'SummaryFails' in node_data and 'summary_result' in node_data['SummaryFails']:
+                            summary = node_data['SummaryFails']['summary_result']
 
-    Retrieve_node = ToolExecutionNode(
-        tool=Retrieve_,
-        node_id="Retrieve_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
-    )
+                    spit_data = node_data.get('spit_data', {}).get('_intermediate', None)
+                    if spit_data and spit_data != [] and spit_data != {}:
+                        _intermediate_outputs.append(spit_data)
+                    
+                    # Problem_Extension 节点
+                    problem_extension = node_data.get('problem_extension', {}).get('_intermediate', None)
+                    if problem_extension and problem_extension != [] and problem_extension != {}:
+                        _intermediate_outputs.append(problem_extension)
+                    
+                    # Retrieve 节点
+                    retrieve_node = node_data.get('retrieve', {}).get('_intermediate_outputs', None)
+                    if retrieve_node and retrieve_node != [] and retrieve_node != {}:
+                        _intermediate_outputs.extend(retrieve_node)
+                    
+                    # Verify 节点
+                    verify_n = node_data.get('verify', {}).get('_intermediate', None)
+                    if verify_n and verify_n != [] and verify_n != {}:
+                        _intermediate_outputs.append(verify_n)
 
-    Verify_node = ToolExecutionNode(
-        tool=Verify_,
-        node_id="Verify_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
-    )
-    
-    Summary_node = ToolExecutionNode(
-        tool=Summary_,
-        node_id="Summary_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
-    )
+                    
+                    # Summary 节点
+                    summary_n = node_data.get('summary', {}).get('_intermediate', None)
+                    if summary_n and summary_n != [] and summary_n != {}:
+                        _intermediate_outputs.append(summary_n)
 
-    Summary_fails_node = ToolExecutionNode(
-        tool=Summary_fails_,
-        node_id="Summary_fails_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
-    )
+            # # 过滤掉空值
+            # _intermediate_outputs = [item for item in _intermediate_outputs if item and item != [] and item != {}]
+            #
+            # # 优化搜索结果
+            # print("=== 开始优化搜索结果 ===")
+            # optimized_outputs = merge_multiple_search_results(_intermediate_outputs)
+            # result=reorder_output_results(optimized_outputs)
+            # # 保存优化后的结果到文件
+            # with open('_intermediate_outputs_optimized.json', 'w', encoding='utf-8') as f:
+            #     import json
+            #     f.write(json.dumps(result, indent=4, ensure_ascii=False))
+            #
+            print(f"=== 最终摘要 ===")
+            print(summary)
+                
+    except Exception as e:
+        import traceback
+        traceback.print_exc()
 
-    Retrieve_Summary_node = ToolExecutionNode(
-        tool=Retrieve_Summary_,
-        node_id="Retrieve_Summary_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
-    )
+    end=time.time()
+    print(100*'y')
+    print(f"总耗时: {end-start}s")
+    print(100*'y')
 
-    Input_Summary_node = ToolExecutionNode(
-        tool=Input_Summary_,
-        node_id="Input_Summary_id",
-        namespace=namespace,
-        search_switch=search_switch,
-        apply_id=apply_id,
-        group_id=group_id,
-        parameter_builder=parameter_builder,
-        storage_type=storage_type,
-        user_rag_memory_id=user_rag_memory_id,
-        memory_config=memory_config,
-    )
 
-    async def content_input_node(state):
-        state_search_switch = state.get("search_switch", search_switch)
-
-        tool_name = "Input_Summary" if state_search_switch == '2' else "Split_The_Problem"
-        session_prefix = "input_summary_call_id" if state_search_switch == '2' else "split_call_id"
-
-        return await create_input_message(
-            state=state,
-            tool_name=tool_name,
-            session_id=f"{session_prefix}_{namespace}",
-            search_switch=search_switch,
-            apply_id=apply_id,
-            group_id=group_id,
-            multimodal_processor=multimodal_processor,
-            memory_config=memory_config,
-        )
-
-    
-    # Build workflow graph
-    workflow = StateGraph(ReadState)
-    workflow.add_node("content_input", content_input_node)
-    workflow.add_node("Split_The_Problem", Split_The_Problem_node)
-    workflow.add_node("Problem_Extension", Problem_Extension_node)
-    workflow.add_node("Retrieve", Retrieve_node)
-    workflow.add_node("Verify", Verify_node)
-    workflow.add_node("Summary", Summary_node)
-    workflow.add_node("Summary_fails", Summary_fails_node)
-    workflow.add_node("Retrieve_Summary", Retrieve_Summary_node)
-    workflow.add_node("Input_Summary", Input_Summary_node)
-
-    # Add edges using imported routers
-    workflow.add_edge(START, "content_input")
-    workflow.add_conditional_edges("content_input", Split_continue)
-    workflow.add_edge("Input_Summary", END)
-    workflow.add_edge("Split_The_Problem", "Problem_Extension")
-    workflow.add_edge("Problem_Extension", "Retrieve")
-    workflow.add_conditional_edges("Retrieve", Retrieve_continue)
-    workflow.add_edge("Retrieve_Summary", END)
-    workflow.add_conditional_edges("Verify", Verify_continue)
-    workflow.add_edge("Summary_fails", END)
-    workflow.add_edge("Summary", END)
-
-    graph = workflow.compile(checkpointer=memory)
-    yield graph
+if __name__ == "__main__":
+    import asyncio
+    asyncio.run(main())

api/app/core/memory/agent/langgraph_graph/routing/__init__.py

@@ -1,13 +0,0 @@
-"""LangGraph routing logic."""
-
-from app.core.memory.agent.langgraph_graph.routing.routers import (
-    Verify_continue,
-    Retrieve_continue,
-    Split_continue,
-)
-
-__all__ = [
-    "Verify_continue",
-    "Retrieve_continue",
-    "Split_continue",
-]

api/app/core/memory/agent/langgraph_graph/routing/routers.py

@@ -1,123 +1,61 @@
-"""
-Routing functions for LangGraph conditional edges.
 
-This module provides routing functions that determine the next node to execute
-based on state values. All functions return Literal types for type safety.
-"""
-
-import logging
-import re
 from typing import Literal
 
-from app.core.memory.agent.langgraph_graph.state.extractors import extract_search_switch
+from app.core.logging_config import get_agent_logger
 from app.core.memory.agent.utils.llm_tools import ReadState, COUNTState
 
-logger = logging.getLogger(__name__)
 
-# Global counter for Verify routing
+logger = get_agent_logger(__name__)
 counter = COUNTState(limit=3)
+def Split_continue(state:ReadState) -> Literal["Split_The_Problem", "Input_Summary"]:
+    """
+    Determine routing based on search_switch value.
 
+    Args:
+        state: State dictionary containing search_switch
 
+    Returns:
+        Next node to execute
+    """
+    logger.debug(f"Split_continue state: {state}")
+    search_switch = state.get('search_switch', '')
+    if search_switch is not None:
+        search_switch = str(search_switch)
+        if search_switch == '2':
+            return 'Input_Summary'
+    return 'Split_The_Problem'  # 默认情况
+
+def Retrieve_continue(state) -> Literal["Verify", "Retrieve_Summary"]:
+    """
+    Determine routing based on search_switch value.
+
+    Args:
+        state: State dictionary containing search_switch
+
+    Returns:
+        Next node to execute
+    """
+    search_switch = state.get('search_switch', '')
+    if search_switch is not None:
+        search_switch = str(search_switch)
+        if search_switch == '0':
+            return 'Verify'
+        elif search_switch == '1':
+            return 'Retrieve_Summary'
+    return 'Retrieve_Summary'  # Default based on business logic
 def Verify_continue(state: ReadState) -> Literal["Summary", "Summary_fails", "content_input"]:
-    """
-    Determine routing after Verify node based on verification result.
-    
-    This function checks the verification result in the last message and routes to:
-    - Summary: if verification succeeded
-    - content_input: if verification failed and retry limit not reached
-    - Summary_fails: if verification failed and retry limit reached
-    
-    Args:
-        state: LangGraph state containing messages
-        
-    Returns:
-        Next node name as Literal type
-    """
-    messages = state.get("messages", [])
-    
-    # Boundary check
-    if not messages:
-        logger.warning("[Verify_continue] No messages in state, defaulting to Summary")
-        counter.reset()
+    status=state.get('verify', '')['status']
+    # loop_count = counter.get_total()
+    if "success" in status:
+        # counter.reset()
         return "Summary"
-    
-    # Increment counter
-    counter.add(1)
-    loop_count = counter.get_total()
-    logger.debug(f"[Verify_continue] Current loop count: {loop_count}")
-    
-    # Extract verification result from last message
-    last_message = messages[-1]
-    last_message_str = str(last_message).replace('\\', '')
-    status_tools = re.findall(r'"split_result": "(.*?)"', last_message_str)
-    logger.debug(f"[Verify_continue] Status tools: {status_tools}")
-    
-    # Route based on verification result
-    if "success" in status_tools:
-        counter.reset()
-        return "Summary"
-    elif "failed" in status_tools:
-        if loop_count < 2:  # Max retry count is 2
-            return "content_input"
-        else:
-            counter.reset()
-            return "Summary_fails"
+    elif "failed" in status:
+        # if loop_count < 2:  # Maximum loop count is 3
+        #     return "content_input"
+        # else:
+            # counter.reset()
+        return "Summary_fails"
     else:
-        # Default to Summary if status is unclear
-        counter.reset()
-        return "Summary"
-
-
-def Retrieve_continue(state: dict) -> Literal["Verify", "Retrieve_Summary"]:
-    """
-    Determine routing after Retrieve node based on search_switch value.
-    
-    This function routes based on the search_switch parameter:
-    - search_switch == '0': Route to Verify (verification needed)
-    - search_switch == '1': Route to Retrieve_Summary (direct summary)
-    
-    Args:
-        state: LangGraph state dictionary
-        
-    Returns:
-        Next node name as Literal type
-    """
-    search_switch = extract_search_switch(state)
-    
-    logger.debug(f"[Retrieve_continue] search_switch: {search_switch}")
-    
-    if search_switch == '0':
-        return 'Verify'
-    elif search_switch == '1':
-        return 'Retrieve_Summary'
-    
-    # Default to Retrieve_Summary
-    logger.debug("[Retrieve_continue] No valid search_switch, defaulting to Retrieve_Summary")
-    return 'Retrieve_Summary'
-
-
-def Split_continue(state: dict) -> Literal["Split_The_Problem", "Input_Summary"]:
-    """
-    Determine routing after content_input node based on search_switch value.
-    
-    This function routes based on the search_switch parameter:
-    - search_switch == '2': Route to Input_Summary (direct input summary)
-    - Otherwise: Route to Split_The_Problem (problem decomposition)
-    
-    Args:
-        state: LangGraph state dictionary
-        
-    Returns:
-        Next node name as Literal type
-    """
-    logger.debug(f"[Split_continue] state keys: {state.keys()}")
-    
-    search_switch = extract_search_switch(state)
-    
-    logger.debug(f"[Split_continue] search_switch: {search_switch}")
-    
-    if search_switch == '2':
-        return 'Input_Summary'
-    
-    # Default to Split_The_Problem
-    return 'Split_The_Problem'
+        # Add default return value to avoid returning None
+        # counter.reset()
+        return "Summary"  # Default based on business requirements

api/app/core/memory/agent/langgraph_graph/routing/write_router.py

@@ -0,0 +1,238 @@
+import json
+import os
+
+from app.core.logging_config import get_agent_logger
+from app.core.memory.agent.langgraph_graph.tools.write_tool import  format_parsing,  messages_parse
+from app.core.memory.agent.langgraph_graph.write_graph import make_write_graph, long_term_storage
+
+from app.core.memory.agent.models.write_aggregate_model import WriteAggregateModel
+from app.core.memory.agent.utils.llm_tools import PROJECT_ROOT_
+from app.core.memory.agent.utils.redis_tool import write_store
+from app.core.memory.agent.utils.redis_tool import count_store
+from app.core.memory.agent.utils.template_tools import TemplateService
+from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+from app.db import get_db_context, get_db
+from app.repositories.memory_short_repository import LongTermMemoryRepository
+from app.schemas.memory_agent_schema import AgentMemory_Long_Term
+from app.services.memory_konwledges_server import write_rag
+from app.services.task_service import get_task_memory_write_result
+from app.tasks import write_message_task
+from app.utils.config_utils import resolve_config_id
+logger = get_agent_logger(__name__)
+template_root = os.path.join(PROJECT_ROOT_, 'memory', 'agent', 'utils', 'prompt')
+
+async def write_rag_agent(end_user_id, user_message, ai_message, user_rag_memory_id):
+    # RAG 模式：组合消息为字符串格式（保持原有逻辑）
+    combined_message = f"user: {user_message}\nassistant: {ai_message}"
+    await write_rag(end_user_id, combined_message, user_rag_memory_id)
+    logger.info(f'RAG_Agent:{end_user_id};{user_rag_memory_id}')
+async def write(storage_type, end_user_id, user_message, ai_message, user_rag_memory_id, actual_end_user_id,
+                actual_config_id, long_term_messages=[]):
+    """
+    写入记忆（支持结构化消息）
+
+    Args:
+        storage_type: 存储类型 (neo4j/rag)
+        end_user_id: 终端用户ID
+        user_message: 用户消息内容
+        ai_message: AI 回复内容
+        user_rag_memory_id: RAG 记忆ID
+        actual_end_user_id: 实际用户ID
+        actual_config_id: 配置ID
+
+    逻辑说明：
+    - RAG 模式：组合 user_message 和 ai_message 为字符串格式，保持原有逻辑不变
+    - Neo4j 模式：使用结构化消息列表
+      1. 如果 user_message 和 ai_message 都不为空：创建配对消息 [user, assistant]
+      2. 如果只有 user_message：创建单条用户消息 [user]（用于历史记忆场景）
+      3. 每条消息会被转换为独立的 Chunk，保留 speaker 字段
+    """
+
+    db = next(get_db())
+    try:
+        actual_config_id = resolve_config_id(actual_config_id, db)
+        # Neo4j 模式：使用结构化消息列表
+        structured_messages = []
+
+        # 始终添加用户消息（如果不为空）
+        if isinstance(user_message, str) and user_message.strip() != "":
+            structured_messages.append({"role": "user", "content": user_message})
+
+        # 只有当 AI 回复不为空时才添加 assistant 消息
+        if isinstance(ai_message, str) and ai_message.strip() != "":
+            structured_messages.append({"role": "assistant", "content": ai_message})
+
+        # 如果提供了 long_term_messages，使用它替代 structured_messages
+        if long_term_messages and isinstance(long_term_messages, list):
+            structured_messages = long_term_messages
+        elif long_term_messages and isinstance(long_term_messages, str):
+            # 如果是 JSON 字符串，先解析
+            try:
+                structured_messages = json.loads(long_term_messages)
+            except json.JSONDecodeError:
+                logger.error(f"Failed to parse long_term_messages as JSON: {long_term_messages}")
+
+        # 如果没有消息，直接返回
+        if not structured_messages:
+            logger.warning(f"No messages to write for user {actual_end_user_id}")
+            return
+
+        logger.info(
+            f"[WRITE] Submitting Celery task - user={actual_end_user_id}, messages={len(structured_messages)}, config={actual_config_id}")
+        write_id = write_message_task.delay(
+            actual_end_user_id,  # end_user_id: 用户ID
+            structured_messages,  # message: JSON 字符串格式的消息列表
+            str(actual_config_id),  # config_id: 配置ID字符串
+            storage_type,  # storage_type: "neo4j"
+            user_rag_memory_id or ""  # user_rag_memory_id: RAG记忆ID（Neo4j模式下不使用）
+        )
+        logger.info(f"[WRITE] Celery task submitted - task_id={write_id}")
+        write_status = get_task_memory_write_result(str(write_id))
+        logger.info(f'[WRITE] Task result - user={actual_end_user_id}, status={write_status}')
+    finally:
+        db.close()
+
+async def term_memory_save(long_term_messages,actual_config_id,end_user_id,type,scope):
+    with get_db_context() as db_session:
+        repo = LongTermMemoryRepository(db_session)
+
+
+        from app.core.memory.agent.utils.redis_tool import write_store
+        result = write_store.get_session_by_userid(end_user_id)
+        if type==AgentMemory_Long_Term.STRATEGY_CHUNK or AgentMemory_Long_Term.STRATEGY_AGGREGATE:
+            data = await format_parsing(result, "dict")
+            chunk_data = data[:scope]
+            if len(chunk_data)==scope:
+                repo.upsert(end_user_id, chunk_data)
+                logger.info(f'---------写入短长期-----------')
+        else:
+            long_time_data = write_store.find_user_recent_sessions(end_user_id, 5)
+            long_messages = await messages_parse(long_time_data)
+            repo.upsert(end_user_id, long_messages)
+            logger.info(f'写入短长期：')
+
+
+
+'''根据窗口'''
+async def window_dialogue(end_user_id,langchain_messages,memory_config,scope):
+    '''
+    根据窗口获取redis数据,写入neo4j：
+     Args:
+        end_user_id: 终端用户ID
+        memory_config: 内存配置对象
+        langchain_messages：原始数据LIST
+        scope：窗口大小
+    '''
+    scope=scope
+    is_end_user_id = count_store.get_sessions_count(end_user_id)
+    if is_end_user_id is not False:
+        is_end_user_id = count_store.get_sessions_count(end_user_id)[0]
+        redis_messages = count_store.get_sessions_count(end_user_id)[1]
+    if is_end_user_id and int(is_end_user_id) != int(scope):
+        is_end_user_id += 1
+        langchain_messages += redis_messages
+        count_store.update_sessions_count(end_user_id, is_end_user_id, langchain_messages)
+    elif int(is_end_user_id) == int(scope):
+        logger.info('写入长期记忆NEO4J')
+        formatted_messages = (redis_messages)
+        # 获取 config_id（如果 memory_config 是对象，提取 config_id；否则直接使用）
+        if hasattr(memory_config, 'config_id'):
+            config_id = memory_config.config_id
+        else:
+            config_id = memory_config
+        
+        await write(AgentMemory_Long_Term.STORAGE_NEO4J, end_user_id, "", "", None, end_user_id,
+                config_id, formatted_messages)
+        count_store.update_sessions_count(end_user_id, 1, langchain_messages)
+    else:
+        count_store.save_sessions_count(end_user_id, 1, langchain_messages)
+
+
+"""根据时间"""
+async def memory_long_term_storage(end_user_id,memory_config,time):
+    '''
+    根据时间获取redis数据,写入neo4j：
+     Args:
+        end_user_id: 终端用户ID
+        memory_config: 内存配置对象
+    '''
+    long_time_data = write_store.find_user_recent_sessions(end_user_id, time)
+    format_messages = (long_time_data)
+    messages=[]
+    memory_config=memory_config.config_id
+    for i in format_messages:
+        message=json.loads(i['Query'])
+        messages+= message
+    if format_messages!=[]:
+        await write(AgentMemory_Long_Term.STORAGE_NEO4J, end_user_id, "", "", None, end_user_id,
+                    memory_config, messages)
+'''聚合判断'''
+async def aggregate_judgment(end_user_id: str, ori_messages: list, memory_config) -> dict:
+    """
+    聚合判断函数：判断输入句子和历史消息是否描述同一事件
+    
+    Args:
+        end_user_id: 终端用户ID
+        ori_messages: 原始消息列表，格式如 [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]
+        memory_config: 内存配置对象
+    """
+
+    try:
+        # 1. 获取历史会话数据（使用新方法）
+        result = write_store.get_all_sessions_by_end_user_id(end_user_id)
+        history = await format_parsing(result)
+        if not result:
+            history = []
+        else:
+            history = await format_parsing(result)
+        json_schema = WriteAggregateModel.model_json_schema()
+        template_service = TemplateService(template_root)
+        system_prompt = await template_service.render_template(
+            template_name='write_aggregate_judgment.jinja2',
+            operation_name='aggregate_judgment',
+            history=history,
+            sentence=ori_messages,
+            json_schema=json_schema
+        )
+        with get_db_context() as db_session:
+            factory = MemoryClientFactory(db_session)
+            llm_client = factory.get_llm_client(memory_config.llm_model_id)
+            messages = [
+                {
+                    "role": "user",
+                    "content": system_prompt
+                }
+            ]
+            structured = await llm_client.response_structured(
+                messages=messages,
+                response_model=WriteAggregateModel
+            )
+        output_value = structured.output
+        if isinstance(output_value, list):
+            output_value = [
+                {"role": msg.role, "content": msg.content} 
+                for msg in output_value
+            ]
+
+        result_dict = {
+            "is_same_event": structured.is_same_event,
+            "output": output_value
+        }
+        if not structured.is_same_event:
+            logger.info(result_dict)
+            await write("neo4j", end_user_id, "", "", None, end_user_id,
+                        memory_config.config_id, output_value)
+        return result_dict
+        
+    except Exception as e:
+        print(f"[aggregate_judgment] 发生错误: {e}")
+        import traceback
+        traceback.print_exc()
+        
+        return {
+            "is_same_event": False,
+            "output": ori_messages,
+            "messages": ori_messages,
+            "history": history if 'history' in locals() else [],
+            "error": str(e)
+        }

api/app/core/memory/agent/langgraph_graph/state/__init__.py

@@ -1,13 +0,0 @@
-"""LangGraph state management utilities."""
-
-from app.core.memory.agent.langgraph_graph.state.extractors import (
-    extract_search_switch,
-    extract_tool_call_id,
-    extract_content_payload,
-)
-
-__all__ = [
-    "extract_search_switch",
-    "extract_tool_call_id",
-    "extract_content_payload",
-]

api/app/core/memory/agent/langgraph_graph/state/extractors.py

@@ -1,179 +0,0 @@
-"""
-State extraction utilities for type-safe access to LangGraph state values.
-
-This module provides utility functions for extracting values from LangGraph state
-dictionaries with proper error handling and sensible defaults.
-"""
-
-import json
-import logging
-from typing import Any, Optional
-
-logger = logging.getLogger(__name__)
-
-def extract_search_switch(state: dict) -> Optional[str]:
-    """
-    Extract search_switch from state or messages.
-    """
-
-    search_switch = state.get("search_switch")
-
-    if search_switch is not None:
-        return str(search_switch)
-
-    # Try to extract from messages
-    messages = state.get("messages", [])
-    if not messages:
-        return None
-
-    # 从最新的消息开始查找
-    for message in reversed(messages):
-        # 尝试从 tool_calls 中提取
-        if hasattr(message, "tool_calls") and message.tool_calls:
-            for tool_call in message.tool_calls:
-                if isinstance(tool_call, dict):
-                    # 从 tool_call 的 args 中提取
-                    if "args" in tool_call and isinstance(tool_call["args"], dict):
-                        search_switch = tool_call["args"].get("search_switch")
-                        if search_switch is not None:
-                            return str(search_switch)
-                    # 直接从 tool_call 中提取
-                    search_switch = tool_call.get("search_switch")
-                    if search_switch is not None:
-                        return str(search_switch)
-
-        # 尝试从 content 中提取（如果是 JSON 格式）
-        if hasattr(message, "content"):
-            try:
-                import json
-                if isinstance(message.content, str):
-                    content_data = json.loads(message.content)
-                    if isinstance(content_data, dict):
-                        search_switch = content_data.get("search_switch")
-                        if search_switch is not None:
-                            return str(search_switch)
-            except (json.JSONDecodeError, ValueError):
-                pass
-
-    return None
-
-
-def extract_tool_call_id(message: Any) -> str:
-    """
-    Extract tool call ID from message using structured attributes.
-    
-    This function extracts the tool call ID from a message object, handling both
-    direct attribute access and tool_calls list structures.
-    
-    Args:
-        message: Message object (typically ToolMessage or AIMessage)
-        
-    Returns:
-        Tool call ID as string
-        
-    Raises:
-        ValueError: If tool call ID cannot be extracted
-        
-    Examples:
-        >>> message = ToolMessage(content="...", tool_call_id="call_123")
-        >>> extract_tool_call_id(message)
-        'call_123'
-    """
-    # Try direct attribute access for ToolMessage
-    if hasattr(message, "tool_call_id"):
-        tool_call_id = message.tool_call_id
-        if tool_call_id:
-            return str(tool_call_id)
-    
-    # Try extracting from tool_calls list for AIMessage
-    if hasattr(message, "tool_calls") and message.tool_calls:
-        tool_call = message.tool_calls[0]
-        if isinstance(tool_call, dict) and "id" in tool_call:
-            return str(tool_call["id"])
-    
-    # Try extracting from id attribute
-    if hasattr(message, "id"):
-        message_id = message.id
-        if message_id:
-            return str(message_id)
-    
-    # If all else fails, raise an error
-    raise ValueError(f"Could not extract tool call ID from message: {type(message)}")
-
-
-def extract_content_payload(message: Any) -> Any:
-    """
-    Extract content payload from ToolMessage, parsing JSON if needed.
-    
-    This function extracts the content from a message and attempts to parse it as JSON
-    if it appears to be a JSON string. It handles various message formats and provides
-    sensible fallbacks.
-    
-    Args:
-        message: Message object (typically ToolMessage)
-        
-    Returns:
-        Parsed content (dict, list, or str)
-        
-    Examples:
-        >>> message = ToolMessage(content='{"key": "value"}')
-        >>> extract_content_payload(message)
-        {'key': 'value'}
-        
-        >>> message = ToolMessage(content='plain text')
-        >>> extract_content_payload(message)
-        'plain text'
-    """
-    # Extract raw content
-    # For ToolMessages (responses from tools), extract from content
-    if hasattr(message, "content"):
-        raw_content = message.content
-        logger.info(f"extract_content_payload: raw_content type={type(raw_content)}, value={str(raw_content)[:500]}")
-        
-        # Handle MCP content format: [{'type': 'text', 'text': '...'}]
-        if isinstance(raw_content, list):
-            for block in raw_content:
-                if isinstance(block, dict) and block.get('type') == 'text':
-                    raw_content = block.get('text', '')
-                    logger.info(f"extract_content_payload: extracted text from MCP format: {str(raw_content)[:300]}")
-                    break
-        
-        # If content is empty and this is an AIMessage with tool_calls,
-        # extract from args (this handles the initial tool call from content_input)
-        if not raw_content and hasattr(message, "tool_calls") and message.tool_calls:
-            tool_call = message.tool_calls[0]
-            if isinstance(tool_call, dict) and "args" in tool_call:
-                return tool_call["args"]
-    else:
-        raw_content = str(message)
-    
-    # If content is already a dict or list, return it directly
-    if isinstance(raw_content, (dict, list)):
-        logger.info(f"extract_content_payload: returning raw dict/list with keys={list(raw_content.keys()) if isinstance(raw_content, dict) else 'list'}")
-        return raw_content
-    
-    # Try to parse as JSON
-    if isinstance(raw_content, str):
-        # First, try direct JSON parsing
-        try:
-            parsed = json.loads(raw_content)
-            logger.info(f"extract_content_payload: parsed JSON, keys={list(parsed.keys()) if isinstance(parsed, dict) else 'list'}")
-            return parsed
-        except (json.JSONDecodeError, ValueError):
-            pass
-        
-        # If that fails, try to extract JSON from the string
-        # This handles cases where the content is embedded in a larger string
-        import re
-        json_candidates = re.findall(r'[\[{].*[\]}]', raw_content, flags=re.DOTALL)
-        for candidate in json_candidates:
-            try:
-                parsed = json.loads(candidate)
-                logger.info(f"extract_content_payload: parsed JSON from candidate, keys={list(parsed.keys()) if isinstance(parsed, dict) else 'list'}")
-                return parsed
-            except (json.JSONDecodeError, ValueError):
-                continue
-    
-    # If all parsing attempts fail, return the raw content
-    logger.info(f"extract_content_payload: returning raw content (parsing failed)")
-    return raw_content

api/app/core/memory/agent/langgraph_graph/tools/tool.py

@@ -0,0 +1,321 @@
+import asyncio
+import json
+from datetime import datetime, timedelta
+
+
+from langchain.tools import tool
+from pydantic import BaseModel, Field
+
+
+from app.core.memory.src.search import (
+    search_by_temporal,
+    search_by_keyword_temporal,
+)
+
+def extract_tool_message_content(response):
+    """从agent响应中提取ToolMessage内容和工具名称"""
+    messages = response.get('messages', [])
+
+    for message in messages:
+        if hasattr(message, 'tool_call_id') and hasattr(message, 'content'):
+            # 这是一个ToolMessage
+            tool_content = message.content
+            tool_name = None
+
+            # 尝试获取工具名称
+            if hasattr(message, 'name'):
+                tool_name = message.name
+            elif hasattr(message, 'tool_name'):
+                tool_name = message.tool_name
+
+            try:
+                # 解析JSON内容
+                parsed_content = json.loads(tool_content)
+                return {
+                    'tool_name': tool_name,
+                    'content': parsed_content
+                }
+            except json.JSONDecodeError:
+                # 如果不是JSON格式，直接返回内容
+                return {
+                    'tool_name': tool_name,
+                    'content': tool_content
+                }
+
+    return None
+
+
+class TimeRetrievalInput(BaseModel):
+    """时间检索工具的输入模式"""
+    context: str = Field(description="用户输入的查询内容")
+    end_user_id: str = Field(default="88a459f5_text09", description="组ID，用于过滤搜索结果")
+
+def create_time_retrieval_tool(end_user_id: str):
+    """
+    创建一个带有特定end_user_id的TimeRetrieval工具（同步版本），用于按时间范围搜索语句(Statements)
+    """
+    
+    def clean_temporal_result_fields(data):
+        """
+        清理时间搜索结果中不需要的字段，并修改结构
+        
+        Args:
+            data: 要清理的数据
+            
+        Returns:
+            清理后的数据
+        """
+        # 需要过滤的字段列表
+        fields_to_remove = {
+            'id', 'apply_id', 'user_id', 'chunk_id', 'created_at', 
+            'valid_at', 'invalid_at', 'statement_ids'
+        }
+        
+        if isinstance(data, dict):
+            cleaned = {}
+            for key, value in data.items():
+                if key == 'statements' and isinstance(value, dict) and 'statements' in value:
+                    # 将 statements: {"statements": [...]} 改为 time_search: {"statements": [...]}
+                    cleaned_value = clean_temporal_result_fields(value)
+                    # 进一步将内部的 statements 改为 time_search
+                    if 'statements' in cleaned_value:
+                        cleaned['results'] = {
+                            'time_search': cleaned_value['statements']
+                        }
+                    else:
+                        cleaned['results'] = cleaned_value
+                elif key not in fields_to_remove:
+                    cleaned[key] = clean_temporal_result_fields(value)
+            return cleaned
+        elif isinstance(data, list):
+            return [clean_temporal_result_fields(item) for item in data]
+        else:
+            return data
+    
+    @tool
+    def TimeRetrievalWithGroupId(context: str, start_date: str = None, end_date: str = None, end_user_id_param: str = None, clean_output: bool = True) -> str:
+        """
+        优化的时间检索工具，只结合时间范围搜索（同步版本），自动过滤不需要的元数据字段
+        显式接收参数：
+        - context: 查询上下文内容
+        - start_date: 开始时间（可选，格式：YYYY-MM-DD）
+        - end_date: 结束时间（可选，格式：YYYY-MM-DD）
+        - end_user_id_param: 组ID（可选，用于覆盖默认组ID）
+        - clean_output: 是否清理输出中的元数据字段
+        -end_date 需要根据用户的描述获取结束的时间，输出格式用strftime("%Y-%m-%d")
+        """
+        async def _async_search():
+            # 使用传入的参数或默认值
+            actual_end_user_id = end_user_id_param or end_user_id
+            actual_end_date = end_date or datetime.now().strftime("%Y-%m-%d")
+            actual_start_date = start_date or (datetime.now() - timedelta(days=7)).strftime("%Y-%m-%d")
+            
+            # 基本时间搜索
+            results = await search_by_temporal(
+                end_user_id=actual_end_user_id,
+                start_date=actual_start_date,
+                end_date=actual_end_date,
+                limit=10
+            )
+            
+            # 清理结果中不需要的字段
+            if clean_output:
+                cleaned_results = clean_temporal_result_fields(results)
+            else:
+                cleaned_results = results
+
+            return json.dumps(cleaned_results, ensure_ascii=False, indent=2)
+        
+        return asyncio.run(_async_search())
+
+    @tool
+    def KeywordTimeRetrieval(context: str, days_back: int = 7, start_date: str = None, end_date: str = None, clean_output: bool = True) -> str:
+        """
+        优化的关键词时间检索工具，结合关键词和时间范围搜索（同步版本），自动过滤不需要的元数据字段
+        显式接收参数：
+        - context: 查询内容
+        - days_back: 向前搜索的天数，默认7天
+        - start_date: 开始时间（可选，格式：YYYY-MM-DD）
+        - end_date: 结束时间（可选，格式：YYYY-MM-DD）
+        - clean_output: 是否清理输出中的元数据字段
+        - end_date 需要根据用户的描述获取结束的时间，输出格式用strftime("%Y-%m-%d")
+        """
+        async def _async_search():
+            actual_end_date = end_date or datetime.now().strftime("%Y-%m-%d")
+            actual_start_date = start_date or (datetime.now() - timedelta(days=days_back)).strftime("%Y-%m-%d")
+
+            # 关键词时间搜索
+            results = await search_by_keyword_temporal(
+                query_text=context,
+                end_user_id=end_user_id,
+                start_date=actual_start_date,
+                end_date=actual_end_date,
+                limit=15
+            )
+            
+            # 清理结果中不需要的字段
+            if clean_output:
+                cleaned_results = clean_temporal_result_fields(results)
+            else:
+                cleaned_results = results
+
+            return json.dumps(cleaned_results, ensure_ascii=False, indent=2)
+
+        return asyncio.run(_async_search())
+    
+    return TimeRetrievalWithGroupId
+
+
+def create_hybrid_retrieval_tool_async(memory_config, **search_params):
+    """
+    创建混合检索工具，使用run_hybrid_search进行混合检索，优化输出格式并过滤不需要的字段
+    
+    Args:
+        memory_config: 内存配置对象
+        **search_params: 搜索参数，包含end_user_id, limit, include等
+    """
+    
+    def clean_result_fields(data):
+        """
+        递归清理结果中不需要的字段
+        
+        Args:
+            data: 要清理的数据（可能是字典、列表或其他类型）
+            
+        Returns:
+            清理后的数据
+        """
+        # 需要过滤的字段列表
+        # TODO: fact_summary 功能暂时禁用，待后续开发完善后启用
+        fields_to_remove = {
+            'invalid_at', 'valid_at', 'chunk_id_from_rel', 'entity_ids', 
+            'expired_at', 'created_at', 'chunk_id', 'id', 'apply_id', 
+            'user_id', 'statement_ids', 'updated_at',"chunk_ids"  ,"fact_summary"
+        }
+        
+        if isinstance(data, dict):
+            # 对字典进行清理
+            cleaned = {}
+            for key, value in data.items():
+                if key not in fields_to_remove:
+                    cleaned[key] = clean_result_fields(value)  # 递归清理嵌套数据
+            return cleaned
+        elif isinstance(data, list):
+            # 对列表中的每个元素进行清理
+            return [clean_result_fields(item) for item in data]
+        else:
+            # 其他类型直接返回
+            return data
+    
+    @tool
+    async def HybridSearch(
+        context: str, 
+        search_type: str = "hybrid",
+        limit: int = 10,
+        end_user_id: str = None,
+        rerank_alpha: float = 0.6,
+        use_forgetting_rerank: bool = False,
+        use_llm_rerank: bool = False,
+        clean_output: bool = True  # 新增：是否清理输出字段
+    ) -> str:
+        """
+        优化的混合检索工具，支持关键词、向量和混合搜索，自动过滤不需要的元数据字段
+        
+        Args:
+            context: 查询内容
+            search_type: 搜索类型 ('keyword', 'embedding', 'hybrid')
+            limit: 结果数量限制
+            end_user_id: 组ID，用于过滤搜索结果
+            rerank_alpha: 重排序权重参数
+            use_forgetting_rerank: 是否使用遗忘重排序
+            use_llm_rerank: 是否使用LLM重排序
+            clean_output: 是否清理输出中的元数据字段
+        """
+        try:
+            # 导入run_hybrid_search函数
+            from app.core.memory.src.search import run_hybrid_search
+            
+            # 合并参数，优先使用传入的参数
+            final_params = {
+                "query_text": context,
+                "search_type": search_type,
+                "end_user_id": end_user_id or search_params.get("end_user_id"),
+                "limit": limit or search_params.get("limit", 10),
+                "include": search_params.get("include", ["summaries", "statements", "chunks", "entities"]),
+                "output_path": None,  # 不保存到文件
+                "memory_config": memory_config,
+                "rerank_alpha": rerank_alpha,
+                "use_forgetting_rerank": use_forgetting_rerank,
+                "use_llm_rerank": use_llm_rerank
+            }
+            
+            # 执行混合检索
+            raw_results = await run_hybrid_search(**final_params)
+            
+            # 清理结果中不需要的字段
+            if clean_output:
+                cleaned_results = clean_result_fields(raw_results)
+            else:
+                cleaned_results = raw_results
+            
+            # 格式化返回结果
+            formatted_results = {
+                "search_query": context,
+                "search_type": search_type,
+                "results": cleaned_results
+            }
+            
+            return json.dumps(formatted_results, ensure_ascii=False, indent=2, default=str)
+            
+        except Exception as e:
+            error_result = {
+                "error": f"混合检索失败: {str(e)}",
+                "search_query": context,
+                "search_type": search_type,
+                "timestamp": datetime.now().isoformat()
+            }
+            return json.dumps(error_result, ensure_ascii=False, indent=2)
+    
+    return HybridSearch
+
+
+def create_hybrid_retrieval_tool_sync(memory_config, **search_params):
+    """
+    创建同步版本的混合检索工具，优化输出格式并过滤不需要的字段
+    
+    Args:
+        memory_config: 内存配置对象
+        **search_params: 搜索参数
+    """
+    @tool
+    def HybridSearchSync(
+        context: str, 
+        search_type: str = "hybrid",
+        limit: int = 10,
+        end_user_id: str = None,
+        clean_output: bool = True
+    ) -> str:
+        """
+        优化的混合检索工具（同步版本），自动过滤不需要的元数据字段
+        
+        Args:
+            context: 查询内容
+            search_type: 搜索类型 ('keyword', 'embedding', 'hybrid')
+            limit: 结果数量限制
+            end_user_id: 组ID，用于过滤搜索结果
+            clean_output: 是否清理输出中的元数据字段
+        """
+        async def _async_search():
+            # 创建异步工具并执行
+            async_tool = create_hybrid_retrieval_tool_async(memory_config, **search_params)
+            return await async_tool.ainvoke({
+                "context": context,
+                "search_type": search_type,
+                "limit": limit,
+                "end_user_id": end_user_id,
+                "clean_output": clean_output
+            })
+        
+        return asyncio.run(_async_search())
+    
+    return HybridSearchSync

api/app/core/memory/agent/langgraph_graph/tools/write_tool.py

@@ -0,0 +1,72 @@
+import json
+
+from langchain_core.messages import HumanMessage, AIMessage
+async def format_parsing(messages: list,type:str='string'):
+    """
+    格式化解析消息列表
+    
+    Args:
+        messages: 消息列表
+        type: 返回类型 ('string' 或 'dict')
+        
+    Returns:
+        格式化后的消息列表
+    """
+    result = []
+    user=[]
+    ai=[]
+
+    for message in messages:
+        hstory_messages = message['messages']
+        for history_messag in hstory_messages.strip().splitlines():
+            history_messag = json.loads(history_messag)
+            for content in history_messag:
+                role = content['role']
+                content = content['content']
+                if type == "string":
+                    if role == 'human' or role=="user":
+                        content = '用户:' + content
+                    else:
+                        content = 'AI:' + content
+                    result.append(content)
+                if type == "dict" :
+                    if role == 'human'  or role=="user":
+                        user.append( content)
+                    else:
+                        ai.append(content)
+    if type == "dict":
+        for key,values in zip(user,ai):
+            result.append({key:values})
+    return result
+
+async def messages_parse(messages: list | dict):
+    user=[]
+    ai=[]
+    database=[]
+    for message in messages:
+        Query = message['Query']
+        Query = json.loads(Query)
+        for data in Query:
+            role = data['role']
+            if role == "human":
+                user.append(data['content'])
+            if role == "ai":
+                ai.append(data['content'])
+    for key, values in zip(user, ai):
+        database.append({key, values})
+    return  database
+
+
+async def agent_chat_messages(user_content,ai_content):
+    messages = [
+        {
+            "role": "user",
+            "content": f"{user_content}"
+        },
+        {
+            "role": "assistant",
+            "content": f"{ai_content}"
+        }
+
+    ]
+    return messages

api/app/core/memory/agent/langgraph_graph/write_graph.py

@@ -1,80 +1,103 @@
+
 import asyncio
 import json
 import sys
 import warnings
 from contextlib import asynccontextmanager
-
-from app.core.logging_config import get_agent_logger
-from app.core.memory.agent.utils.llm_tools import WriteState
-from app.schemas.memory_config_schema import MemoryConfig
-from langchain_core.messages import AIMessage
 from langgraph.constants import END, START
 from langgraph.graph import StateGraph
-from langgraph.prebuilt import ToolNode
+
+from app.db import get_db, get_db_context
+from app.core.logging_config import get_agent_logger
+from app.core.memory.agent.utils.llm_tools import WriteState
+from app.core.memory.agent.langgraph_graph.nodes.write_nodes import write_node
+from app.schemas.memory_agent_schema import AgentMemory_Long_Term
+from app.services.memory_config_service import MemoryConfigService
+
 
 warnings.filterwarnings("ignore", category=RuntimeWarning)
-
 logger = get_agent_logger(__name__)
 
 if sys.platform.startswith("win"):
     asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())
-
-
 @asynccontextmanager
-async def make_write_graph(user_id, tools, apply_id, group_id, memory_config: MemoryConfig):
+async def make_write_graph():
     """
     Create a write graph workflow for memory operations.
-    
+
     Args:
         user_id: User identifier
         tools: MCP tools loaded from session
         apply_id: Application identifier
-        group_id: Group identifier
+        end_user_id: Group identifier
         memory_config: MemoryConfig object containing all configuration
     """
-    logger.info("Loading MCP tools: %s", [t.name for t in tools])
-    logger.info(f"Using memory_config: {memory_config.config_name} (id={memory_config.config_id})")
-
-    data_write_tool = next((t for t in tools if t.name == "Data_write"), None)
-
-    if not data_write_tool:
-        logger.error("Data_write tool not found", exc_info=True)
-        raise ValueError("Data_write tool not found")
-
-    write_node = ToolNode([data_write_tool])
-
-    async def call_model(state):
-        messages = state["messages"]
-        last_message = messages[-1]
-        content = last_message[1] if isinstance(last_message, tuple) else last_message.content
-
-        # Call Data_write directly with memory_config
-        write_params = {
-            "content": content,
-            "apply_id": apply_id,
-            "group_id": group_id,
-            "user_id": user_id,
-            "memory_config": memory_config,
-        }
-        logger.debug(f"Passing memory_config to Data_write: {memory_config.config_id}")
-
-        write_result = await data_write_tool.ainvoke(write_params)
-
-        if isinstance(write_result, dict):
-            result_content = write_result.get("data", str(write_result))
-        else:
-            result_content = str(write_result)
-        logger.info("Write content: %s", result_content)
-        return {"messages": [AIMessage(content=result_content)]}
-
     workflow = StateGraph(WriteState)
-    workflow.add_node("content_input", call_model)
     workflow.add_node("save_neo4j", write_node)
-    workflow.add_edge(START, "content_input")
-    workflow.add_edge("content_input", "save_neo4j")
+    workflow.add_edge(START, "save_neo4j")
     workflow.add_edge("save_neo4j", END)
 
     graph = workflow.compile()
 
-
     yield graph
+
+async def long_term_storage(long_term_type:str="chunk",langchain_messages:list=[],memory_config:str='',end_user_id:str='',scope:int=6):
+    from app.core.memory.agent.langgraph_graph.routing.write_router import memory_long_term_storage, window_dialogue,aggregate_judgment
+    from app.core.memory.agent.utils.redis_tool import write_store
+    write_store.save_session_write(end_user_id,  (langchain_messages))
+    # 获取数据库会话
+    with get_db_context() as db_session:
+        config_service = MemoryConfigService(db_session)
+        memory_config = config_service.load_memory_config(
+            config_id=memory_config,  # 改为整数
+            service_name="MemoryAgentService"
+        )
+        if long_term_type=='chunk':
+            '''方案一:对话窗口6轮对话'''
+            await window_dialogue(end_user_id,langchain_messages,memory_config,scope)
+        if long_term_type=='time':
+            """时间"""
+            await memory_long_term_storage(end_user_id, memory_config,5)
+        if  long_term_type=='aggregate':
+            """方案三：聚合判断"""
+            await aggregate_judgment(end_user_id, langchain_messages, memory_config)
+
+
+
+async def write_long_term(storage_type,end_user_id,message_chat,aimessages,user_rag_memory_id,actual_config_id):
+    from app.core.memory.agent.langgraph_graph.routing.write_router import write_rag_agent
+    from app.core.memory.agent.langgraph_graph.routing.write_router import term_memory_save
+    from app.core.memory.agent.langgraph_graph.tools.write_tool import  agent_chat_messages
+    if storage_type == AgentMemory_Long_Term.STORAGE_RAG:
+        await write_rag_agent(end_user_id, message_chat, aimessages, user_rag_memory_id)
+    else:
+        # AI 回复写入（用户消息和 AI 回复配对，一次性写入完整对话）
+        CHUNK = AgentMemory_Long_Term.STRATEGY_CHUNK
+        SCOPE = AgentMemory_Long_Term.DEFAULT_SCOPE
+        long_term_messages = await agent_chat_messages(message_chat, aimessages)
+        await long_term_storage(long_term_type=CHUNK, langchain_messages=long_term_messages,
+                                memory_config=actual_config_id, end_user_id=end_user_id, scope=SCOPE)
+        await term_memory_save(long_term_messages, actual_config_id, end_user_id, CHUNK, scope=SCOPE)
+
+# async def main():
+#     """主函数 - 运行工作流"""
+#     langchain_messages = [
+#     {
+#       "role": "user",
+#       "content": "今天周五去爬山"
+#     },
+#     {
+#       "role": "assistant",
+#       "content": "好耶"
+#     }
+#
+#   ]
+#     end_user_id = '837fee1b-04a2-48ee-94d7-211488908940'  # 组ID
+#     memory_config="08ed205c-0f05-49c3-8e0c-a580d28f5fd4"
+#     await long_term_storage(long_term_type="chunk",langchain_messages=langchain_messages,memory_config=memory_config,end_user_id=end_user_id,scope=2)
+#
+#
+#
+# if __name__ == "__main__":
+#     import asyncio
+#     asyncio.run(main())

api/app/core/memory/agent/mcp_server/__init__.py

@@ -1,28 +0,0 @@
-"""
-MCP Server package for memory agent.
-
-This package provides the FastMCP server implementation with context-based
-dependency injection for tool functions.
-
-Package structure:
-- server: FastMCP server initialization and context setup
-- tools: MCP tool implementations
-- models: Pydantic response models
-- services: Business logic services
-"""
-# from app.core.memory.agent.mcp_server.server import (
-#     mcp,
-#     initialize_context,
-#     main,
-#     get_context_resource
-# )
-
-# # Import tools to register them (but don't export them)
-# from app.core.memory.agent.mcp_server import tools
-
-# __all__ = [
-#     'mcp',
-#     'initialize_context',
-#     'main',
-#     'get_context_resource',
-# ]

api/app/core/memory/agent/mcp_server/mcp_instance.py

@@ -1,11 +0,0 @@
-"""
-MCP Server Instance
-
-This module contains the FastMCP server instance that is shared across all modules.
-It's in a separate file to avoid circular import issues.
-"""
-from mcp.server.fastmcp import FastMCP
-
-# Initialize FastMCP server instance
-# This instance is shared across all tool modules
-mcp = FastMCP('data_flow')

api/app/core/memory/agent/mcp_server/models/verification_models.py

@@ -1,14 +0,0 @@
-"""Pydantic models for verification operations."""
-
-from typing import List, Optional, Dict, Any
-from pydantic import BaseModel, Field
-
-
-class VerificationResult(BaseModel):
-    """Result model for verification operation."""
-    
-    query: str
-    expansion_issue: List[Dict[str, Any]]
-    split_result: str
-    reason: Optional[str] = None
-    history: List[Dict[str, Any]] = Field(default_factory=list)

api/app/core/memory/agent/mcp_server/server.py

@@ -1,159 +0,0 @@
-"""
-MCP Server initialization with FastMCP context setup.
-
-This module initializes the FastMCP server and registers shared resources
-in the context for dependency injection into tool functions.
-"""
-import os
-import sys
-
-from app.core.config import settings
-from app.core.logging_config import get_agent_logger
-from app.core.memory.agent.mcp_server.mcp_instance import mcp
-from app.core.memory.agent.mcp_server.services.search_service import SearchService
-from app.core.memory.agent.mcp_server.services.session_service import SessionService
-from app.core.memory.agent.mcp_server.services.template_service import TemplateService
-from app.core.memory.agent.utils.llm_tools import PROJECT_ROOT_
-from app.core.memory.agent.utils.redis_tool import store
-
-logger = get_agent_logger(__name__)
-
-
-def get_context_resource(ctx, resource_name: str):
-    """
-    Helper function to retrieve a resource from the FastMCP context.
-    
-    Args:
-        ctx: FastMCP Context object (passed to tool functions)
-        resource_name: Name of the resource to retrieve
-        
-    Returns:
-        The requested resource
-        
-    Raises:
-        AttributeError: If the resource doesn't exist
-        
-    Example:
-        @mcp.tool()
-        async def my_tool(ctx: Context):
-            template_service = get_context_resource(ctx, 'template_service')
-            llm_client = get_context_resource(ctx, 'llm_client')
-    """
-    if not hasattr(ctx, 'fastmcp') or ctx.fastmcp is None:
-        raise RuntimeError("Context does not have fastmcp attribute")
-    
-    if not hasattr(ctx.fastmcp, resource_name):
-        raise AttributeError(
-            f"Resource '{resource_name}' not found in context. "
-            f"Available resources: {[k for k in dir(ctx.fastmcp) if not k.startswith('_')]}"
-        )
-    
-    return getattr(ctx.fastmcp, resource_name)
-
-
-def initialize_context():
-    """
-    Initialize and register shared resources in FastMCP context.
-    
-    This function sets up all shared resources that will be available
-    to tool functions via dependency injection through the context parameter.
-    
-    Resources are stored as attributes on the FastMCP instance and can be
-    accessed via ctx.fastmcp in tool functions.
-    
-    Resources registered:
-    - session_store: RedisSessionStore for session management
-    - llm_client: LLM client for structured API calls
-    - app_settings: Application settings (renamed to avoid conflict with FastMCP settings)
-    - template_service: Service for template rendering
-    - search_service: Service for hybrid search
-    - session_service: Service for session operations
-    """
-    try:
-        # Register Redis session store
-        logger.info("Registering session_store in context")
-        mcp.session_store = store
-        
-        # Note: LLM client is NOT loaded at server startup
-        # It should be loaded dynamically when needed, with config_id passed explicitly
-        # to make_write_graph or make_read_graph functions
-        logger.info("LLM client will be loaded dynamically with config_id when needed")
-        mcp.llm_client = None  # Placeholder - actual client loaded per-request with config_id
-        
-        # Register application settings (renamed to avoid conflict with FastMCP's settings)
-        logger.info("Registering app_settings in context")
-        mcp.app_settings = settings
-        
-        # Register template service
-        template_root = PROJECT_ROOT_ + '/agent/utils/prompt'
-        # logger.info(f"Registering template_service in context with root: {template_root}")
-        template_service = TemplateService(template_root)
-        mcp.template_service = template_service
-        
-        # Register search service
-        # logger.info("Registering search_service in context")
-        search_service = SearchService()
-        mcp.search_service = search_service
-        
-        # Register session service
-        # logger.info("Registering session_service in context")
-        session_service = SessionService(store)
-        mcp.session_service = session_service
-        
-        # logger.info("All context resources registered successfully")
-        
-    except Exception as e:
-        logger.error(f"Failed to initialize context: {e}", exc_info=True)
-        raise
-
-
-def main():
-    """
-    Main entry point for the MCP server.
-    
-    Initializes context and starts the server with SSE transport.
-    """
-    try:
-        logger.info("Starting MCP server initialization")
-        # Initialize context resources
-        initialize_context()
-        
-        # Import and register tools (imports trigger tool registration)
-        from app.core.memory.agent.mcp_server.tools import (  # noqa: F401
-            data_tools,
-            problem_tools,
-            retrieval_tools,
-            summary_tools,
-            verification_tools,
-        )
-        
-        # Tools are registered via imports above
-        
-        # Get MCP port from environment (default: 8081)
-        mcp_port = int(os.getenv("MCP_PORT", "8081"))
-        logger.info(f"Starting MCP server on {settings.SERVER_IP}:{mcp_port} with SSE transport")
-        
-        # Configure DNS rebinding protection for Docker container compatibility
-        from mcp.server.fastmcp.server import TransportSecuritySettings
-        
-        # Disable DNS rebinding protection to allow Docker container hostnames
-        # This allows containers to connect using service names like 'mcp-server'
-        mcp.settings.transport_security = TransportSecuritySettings(
-            enable_dns_rebinding_protection=False,
-        )
-        logger.info("DNS rebinding protection: disabled for Docker container compatibility")
-        
-        # logger.info(f"Starting MCP server on {settings.SERVER_IP}:{mcp_port} with SSE transport")
-        
-        # Run the server with SSE transport for HTTP connections
-        import uvicorn
-        app = mcp.sse_app()
-        uvicorn.run(app, host=settings.SERVER_IP, port=mcp_port, log_level="info")
-        
-    except Exception as e:
-        logger.error(f"Failed to start MCP server: {e}", exc_info=True)
-        sys.exit(1)
-
-
-if __name__ == "__main__":
-    main()

api/app/core/memory/agent/mcp_server/tools/__init__.py

@@ -1,27 +0,0 @@
-"""
-MCP Tools module.
-
-This module contains all MCP tool implementations organized by functionality.
-
-Tools are organized into the following modules:
-- problem_tools: Question segmentation and extension
-- retrieval_tools: Database and context retrieval
-- verification_tools: Data verification
-- summary_tools: Summarization and summary retrieval
-- data_tools: Data type differentiation and writing
-"""
-
-# Import all tool modules to register them with the MCP server
-from . import problem_tools
-from . import retrieval_tools
-from . import verification_tools
-from . import summary_tools
-from . import data_tools
-
-__all__ = [
-    'problem_tools',
-    'retrieval_tools',
-    'verification_tools',
-    'summary_tools',
-    'data_tools',
-]

api/app/core/memory/agent/mcp_server/tools/data_tools.py

@@ -1,155 +0,0 @@
-"""
-Data Tools for data type differentiation and writing.
-
-This module contains MCP tools for distinguishing data types and writing data.
-"""
-
-import os
-
-from app.core.logging_config import get_agent_logger
-from app.core.memory.agent.mcp_server.mcp_instance import mcp
-from app.core.memory.agent.mcp_server.models.retrieval_models import (
-    DistinguishTypeResponse,
-)
-from app.core.memory.agent.mcp_server.server import get_context_resource
-from app.core.memory.agent.utils.write_tools import write
-from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
-from app.db import get_db_context
-from app.schemas.memory_config_schema import MemoryConfig
-from mcp.server.fastmcp import Context
-
-logger = get_agent_logger(__name__)
-
-
-@mcp.tool()
-async def Data_type_differentiation(
-    ctx: Context,
-    context: str,
-    memory_config: MemoryConfig,
-) -> dict:
-    """
-    Distinguish the type of data (read or write).
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: Text to analyze for type differentiation
-        memory_config: MemoryConfig object containing LLM configuration
-        
-    Returns:
-        dict: Contains 'context' with the original text and 'type' field
-    """
-    try:
-        # Extract services from context
-        template_service = get_context_resource(ctx, 'template_service')
-        
-        # Get LLM client from memory_config using factory pattern
-        with get_db_context() as db:
-            factory = MemoryClientFactory(db)
-            llm_client = factory.get_llm_client_from_config(memory_config)
-        
-        # Render template
-        try:
-            system_prompt = await template_service.render_template(
-                template_name='distinguish_types_prompt.jinja2',
-                operation_name='status_typle',
-                user_query=context
-            )
-        except Exception as e:
-            logger.error(
-                f"Template rendering failed for Data_type_differentiation: {e}",
-                exc_info=True
-            )
-            return {
-                "type": "error",
-                "message": f"Prompt rendering failed: {str(e)}"
-            }
-
-        # Call LLM with structured response
-        try:
-            structured = await llm_client.response_structured(
-                messages=[{"role": "system", "content": system_prompt}],
-                response_model=DistinguishTypeResponse
-            )
-            
-            result = structured.model_dump()
-            
-            # Add context to result
-            result["context"] = context
-            
-            return result
-            
-        except Exception as e:
-            logger.error(
-                f"LLM call failed for Data_type_differentiation: {e}",
-                exc_info=True
-            )
-            return {
-                "context": context,
-                "type": "error",
-                "message": f"LLM call failed: {str(e)}"
-            }
-            
-    except Exception as e:
-        logger.error(
-            f"Data_type_differentiation failed: {e}",
-            exc_info=True
-        )
-        return {
-            "context": context,
-            "type": "error",
-            "message": str(e)
-        }
-
-
-@mcp.tool()
-async def Data_write(
-    ctx: Context,
-    content: str,
-    user_id: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-) -> dict:
-    """
-    Write data to the database/file system.
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        content: Data content to write
-        user_id: User identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        
-    Returns:
-        dict: Contains 'status', 'saved_to', and 'data' fields
-    """
-    try:
-        # Ensure output directory exists
-        os.makedirs("data_output", exist_ok=True)
-        file_path = os.path.join("data_output", "user_data.csv")
-
-        # Write data - clients are constructed inside write() from memory_config
-        await write(
-            content=content,
-            user_id=user_id,
-            apply_id=apply_id,
-            group_id=group_id,
-            memory_config=memory_config,
-        )
-        logger.info(f"Write completed successfully! Config: {memory_config.config_name}")
-
-        return {
-            "status": "success",
-            "saved_to": file_path,
-            "data": content,
-            "config_id": memory_config.config_id,
-            "config_name": memory_config.config_name,
-        }
-
-    except Exception as e:
-        logger.error(f"Data_write failed: {e}", exc_info=True)
-        return {
-            "status": "error",
-            "message": str(e),
-        }

api/app/core/memory/agent/mcp_server/tools/problem_tools.py

@@ -1,304 +0,0 @@
-"""
-Problem Tools for question segmentation and extension.
-
-This module contains MCP tools for breaking down and extending user questions.
-LLM clients are constructed from MemoryConfig when needed.
-"""
-
-import json
-import time
-
-from app.core.logging_config import get_agent_logger, log_time
-from app.core.memory.agent.mcp_server.mcp_instance import mcp
-from app.core.memory.agent.mcp_server.models.problem_models import (
-    ProblemBreakdownResponse,
-    ProblemExtensionResponse,
-)
-from app.core.memory.agent.mcp_server.server import get_context_resource
-from app.core.memory.agent.utils.messages_tool import Problem_Extension_messages_deal
-from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
-from app.db import get_db_context
-from app.schemas.memory_config_schema import MemoryConfig
-from mcp.server.fastmcp import Context
-
-logger = get_agent_logger(__name__)
-
-
-@mcp.tool()
-async def Split_The_Problem(
-    ctx: Context,
-    sentence: str,
-    sessionid: str,
-    messages_id: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-) -> dict:
-    """
-    Segment the dialogue or sentence into sub-problems.
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        sentence: Original sentence to split
-        sessionid: Session identifier
-        messages_id: Message identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        
-    Returns:
-        dict: Contains 'context' (JSON string of split results) and 'original' sentence
-    """
-    start = time.time()
-
-    try:
-        # Extract services from context
-        template_service = get_context_resource(ctx, "template_service")
-        session_service = get_context_resource(ctx, "session_service")
-
-        # Get LLM client from memory_config
-        with get_db_context() as db:
-            factory = MemoryClientFactory(db)
-            llm_client = factory.get_llm_client_from_config(memory_config)
-        
-        # Extract user ID from session
-        user_id = session_service.resolve_user_id(sessionid)
-        
-        # Get conversation history
-        history = await session_service.get_history(user_id, apply_id, group_id)
-        # Override with empty list for now (as in original)
-        history = []
-        
-        # Render template
-        try:
-            system_prompt = await template_service.render_template(
-                template_name='problem_breakdown_prompt.jinja2',
-                operation_name='split_the_problem',
-                history=history,
-                sentence=sentence
-            )
-        except Exception as e:
-            logger.error(
-                f"Template rendering failed for Split_The_Problem: {e}",
-                exc_info=True
-            )
-            return {
-                "context": json.dumps([], ensure_ascii=False),
-                "original": sentence,
-                "error": f"Prompt rendering failed: {str(e)}"
-            }
-        
-        # Call LLM with structured response
-        try:
-            structured = await llm_client.response_structured(
-                messages=[{"role": "system", "content": system_prompt}],
-                response_model=ProblemBreakdownResponse
-            )
-            
-            # Handle RootModel response with .root attribute access
-            if structured is None:
-                # LLM returned None, use empty list as fallback
-                split_result = json.dumps([], ensure_ascii=False)
-            elif hasattr(structured, 'root') and structured.root is not None:
-                split_result = json.dumps(
-                    [item.model_dump() for item in structured.root],
-                    ensure_ascii=False
-                )
-            elif isinstance(structured, list):
-                # Fallback: treat structured itself as the list
-                split_result = json.dumps(
-                    [item.model_dump() for item in structured],
-                    ensure_ascii=False
-                )
-            else:
-                # Last resort: use empty list
-                split_result = json.dumps([], ensure_ascii=False)
-                
-        except Exception as e:
-            logger.error(
-                f"LLM call failed for Split_The_Problem: {e}",
-                exc_info=True
-            )
-            split_result = json.dumps([], ensure_ascii=False)
-        
-        logger.info("Problem splitting")
-        logger.info(f"Problem split result: {split_result}")
-        
-        # Emit intermediate output for frontend
-        result = {
-            "context": split_result,
-            "original": sentence,
-            "_intermediate": {
-                "type": "problem_split",
-                "data": json.loads(split_result) if split_result else [],
-                "original_query": sentence
-            }
-        }
-        
-        return result
-        
-    except Exception as e:
-        logger.error(
-            f"Split_The_Problem failed: {e}",
-            exc_info=True
-        )
-        return {
-            "context": json.dumps([], ensure_ascii=False),
-            "original": sentence,
-            "error": str(e)
-        }
-        
-    finally:
-        # Log execution time
-        end = time.time()
-        try:
-            duration = end - start
-        except Exception:
-            duration = 0.0
-        log_time('Problem splitting', duration)
-
-
-@mcp.tool()
-async def Problem_Extension(
-    ctx: Context,
-    context: dict,
-    usermessages: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-    storage_type: str = "",
-    user_rag_memory_id: str = "",
-) -> dict:
-    """
-    Extend the problem with additional sub-questions.
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: Dictionary containing split problem results
-        usermessages: User messages identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type for the workspace (optional)
-        user_rag_memory_id: User RAG memory identifier (optional)
-        
-    Returns:
-        dict: Contains 'context' (aggregated questions) and 'original' question
-    """
-    start = time.time()
-
-    try:
-        # Extract services from context
-        template_service = get_context_resource(ctx, "template_service")
-        session_service = get_context_resource(ctx, "session_service")
-
-        # Get LLM client from memory_config
-        with get_db_context() as db:
-            factory = MemoryClientFactory(db)
-            llm_client = factory.get_llm_client_from_config(memory_config)
-        
-        # Resolve session ID from usermessages
-        from app.core.memory.agent.utils.messages_tool import Resolve_username
-        sessionid = Resolve_username(usermessages)
-        
-        # Get conversation history
-        history = await session_service.get_history(sessionid, apply_id, group_id)
-        # Override with empty list for now (as in original)
-        history = []
-        
-        # Process context to extract questions
-        extent_quest, original = await Problem_Extension_messages_deal(context)
-        
-        # Format questions for template rendering
-        questions_formatted = []
-        for msg in extent_quest:
-            if msg.get("role") == "user":
-                questions_formatted.append(msg.get("content", ""))
-        
-        # Render template
-        try:
-            system_prompt = await template_service.render_template(
-                template_name='Problem_Extension_prompt.jinja2',
-                operation_name='problem_extension',
-                history=history,
-                questions=questions_formatted
-            )
-        except Exception as e:
-            logger.error(
-                f"Template rendering failed for Problem_Extension: {e}",
-                exc_info=True
-            )
-            return {
-                "context": {},
-                "original": original,
-                "error": f"Prompt rendering failed: {str(e)}"
-            }
-        
-        # Call LLM with structured response
-        try:
-            response_content = await llm_client.response_structured(
-                messages=[{"role": "system", "content": system_prompt}],
-                response_model=ProblemExtensionResponse
-            )
-            
-            # Aggregate results by original question
-            aggregated_dict = {}
-            for item in response_content.root:
-                key = getattr(item, "original_question", None) or (
-                    item.get("original_question") if isinstance(item, dict) else None
-                )
-                value = getattr(item, "extended_question", None) or (
-                    item.get("extended_question") if isinstance(item, dict) else None
-                )
-                if not key or not value:
-                    continue
-                aggregated_dict.setdefault(key, []).append(value)
-                
-        except Exception as e:
-            logger.error(
-                f"LLM call failed for Problem_Extension: {e}",
-                exc_info=True
-            )
-            aggregated_dict = {}
-        
-        logger.info("Problem extension")
-        logger.info(f"Problem extension result: {aggregated_dict}")
-        
-        # Emit intermediate output for frontend
-        result = {
-            "context": aggregated_dict,
-            "original": original,
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "_intermediate": {
-                "type": "problem_extension",
-                "data": aggregated_dict,
-                "original_query": original,
-                "storage_type": storage_type,
-                "user_rag_memory_id": user_rag_memory_id
-            }
-        }
-        
-        return result
-        
-    except Exception as e:
-        logger.error(
-            f"Problem_Extension failed: {e}",
-            exc_info=True
-        )
-        return {
-            "context": {},
-            "original": context.get("original", ""),
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "error": str(e)
-        }
-        
-    finally:
-        # Log execution time
-        end = time.time()
-        try:
-            duration = end - start
-        except Exception:
-            duration = 0.0
-        log_time('Problem extension', duration)

api/app/core/memory/agent/mcp_server/tools/retrieval_tools.py

@@ -1,294 +0,0 @@
-"""
-Retrieval Tools for database and context retrieval.
-
-This module contains MCP tools for retrieving data using hybrid search.
-"""
-
-import os
-import time
-
-from app.core.logging_config import get_agent_logger, log_time
-from app.core.memory.agent.mcp_server.mcp_instance import mcp
-from app.core.memory.agent.mcp_server.server import get_context_resource
-from app.core.memory.agent.utils.llm_tools import (
-    deduplicate_entries,
-    merge_to_key_value_pairs,
-)
-from app.core.memory.agent.utils.messages_tool import Retriev_messages_deal
-from app.core.rag.nlp.search import knowledge_retrieval
-from app.schemas.memory_config_schema import MemoryConfig
-from dotenv import load_dotenv
-from mcp.server.fastmcp import Context
-
-load_dotenv()
-logger = get_agent_logger(__name__)
-
-
-@mcp.tool()
-async def Retrieve(
-    ctx: Context,
-    context,
-    usermessages: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-    storage_type: str = "",
-    user_rag_memory_id: str = "",
-) -> dict:
-    """
-    Retrieve data from the database using hybrid search.
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: Dictionary or string containing query information
-        usermessages: User messages identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type for the workspace (e.g., 'rag', 'vector')
-        user_rag_memory_id: User RAG memory identifier
-        
-    Returns:
-        dict: Contains 'context' with Query and Expansion_issue results
-    """
-    kb_config = {
-        "knowledge_bases": [
-            {
-                "kb_id": user_rag_memory_id,
-                "similarity_threshold": 0.7,
-                "vector_similarity_weight": 0.5,
-                "top_k": 10,
-                "retrieve_type": "participle"
-            }
-        ],
-        "merge_strategy": "weight",
-        "reranker_id": os.getenv('reranker_id'),
-        "reranker_top_k": 10
-    }
-    start = time.time()
-    logger.info(f"Retrieve: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
-    logger.info(f"Retrieve: context type={type(context)}, context={str(context)[:500]}")
-    
-    try:
-        # Extract services from context
-        search_service = get_context_resource(ctx, 'search_service')
-        
-        databases_anser = []
-        
-        # Handle both dict and string context
-        if isinstance(context, dict):
-            # Process dict context with extended questions
-            all_items = []
-            logger.info(f"Retrieve: context keys={list(context.keys())}")
-            content, original = await Retriev_messages_deal(context)
-            logger.info(f"Retrieve: after Retriev_messages_deal - content_type={type(content)}, content={str(content)[:300]}")
-            logger.info(f"Retrieve: original='{original[:100] if original else 'EMPTY'}'")
-            
-            if not original:
-                logger.warning(f"Retrieve: original query is empty! context={context}")
-            
-            # Extract all query items from content
-            # content is like {original_question: [extended_questions...], ...}
-            for key, values in content.items():
-                if isinstance(values, list):
-                    all_items.extend(values)
-                elif isinstance(values, str):
-                    all_items.append(values)
-                elif values is not None:
-                    # Fallback: convert non-empty non-list values to string
-                    all_items.append(str(values))
-            
-            # Execute search for each question
-            for idx, question in enumerate(all_items):
-                try:
-                    # Prepare search parameters based on storage type
-                    search_params = {
-                        "group_id": group_id,
-                        "question": question,
-                        "return_raw_results": True
-                    }
-
-                    # Add storage-specific parameters
-                    if storage_type == "rag" and user_rag_memory_id:
-                        retrieve_chunks_result = knowledge_retrieval(question, kb_config,[str(group_id)])
-                        try:
-                            retrieval_knowledge = [i.page_content for i in retrieve_chunks_result]
-                            clean_content = '\n\n'.join(retrieval_knowledge)
-                            cleaned_query=question
-                            raw_results=clean_content
-                            logger.info(f" Using RAG storage with memory_id={user_rag_memory_id}")
-                        except:
-                            clean_content = ''
-                            raw_results=''
-                            cleaned_query = question
-                            logger.info(f"No content retrieved from knowledge base: {user_rag_memory_id}")
-                    else:
-                        clean_content, cleaned_query, raw_results = await search_service.execute_hybrid_search(
-                            **search_params, memory_config=memory_config
-                        )
-
-                    databases_anser.append({
-                        "Query_small": cleaned_query,
-                        "Result_small": clean_content,
-                        "_intermediate": {
-                            "type": "search_result",
-                            "query": cleaned_query,
-                            "raw_results": raw_results,  
-                            "index": idx + 1,
-                            "total": len(all_items)
-                        }
-                    })
-                except Exception as e:
-                    logger.error(
-                        f"Retrieve: hybrid_search failed for question '{question}': {e}",
-                        exc_info=True
-                    )
-                    # Continue with empty result for this question
-                    databases_anser.append({
-                        "Query_small": question,
-                        "Result_small": ""
-                    })
-            
-            # Build initial database data structure
-            databases_data = {
-                "Query": original,
-                "Expansion_issue": databases_anser
-            }
-            
-            # Collect intermediate outputs before deduplication
-            intermediate_outputs = []
-            for item in databases_anser:
-                if '_intermediate' in item:
-                    intermediate_outputs.append(item['_intermediate'])
-            
-            # Deduplicate and merge results
-            deduplicated_data = deduplicate_entries(databases_data['Expansion_issue'])
-            deduplicated_data_merged = merge_to_key_value_pairs(
-                deduplicated_data,
-                'Query_small',
-                'Result_small'
-            )
-            
-            # Restructure for Verify/Retrieve_Summary compatibility
-            keys, val = [], []
-            for item in deduplicated_data_merged:
-                for items_key, items_value in item.items():
-                    keys.append(items_key)
-                    val.append(items_value)
-            
-            send_verify = []
-            for i, j in zip(keys, val, strict=False):
-                send_verify.append({
-                    "Query_small": i,
-                    "Answer_Small": j
-                })
-            
-            dup_databases = {
-                "Query": original,
-                "Expansion_issue": send_verify,
-                "_intermediate_outputs": intermediate_outputs  # Preserve intermediate outputs
-            }
-            
-            logger.info(f"Collected {len(intermediate_outputs)} intermediate outputs from search results")
-            
-        else:
-            # Handle string context (simple query)
-            query = str(context).strip()
-            
-            try:
-                # Prepare search parameters based on storage type
-                search_params = {
-                    "group_id": group_id,
-                    "question": query,
-                    "return_raw_results": True
-                }
-                
-                # Add storage-specific parameters
-                if storage_type == "rag" and user_rag_memory_id:
-                    retrieve_chunks_result = knowledge_retrieval(query, kb_config,[str(group_id)])
-                    try:
-                        retrieval_knowledge = [i.page_content for i in retrieve_chunks_result]
-                        clean_content = '\n\n'.join(retrieval_knowledge)
-                        cleaned_query = query
-                        raw_results = clean_content
-                        logger.info(f" Using RAG storage with memory_id={user_rag_memory_id}")
-                    except:
-                        clean_content = ''
-                        raw_results = ''
-                        cleaned_query = query
-                        logger.info(f"No content retrieved from knowledge base: {user_rag_memory_id}")
-                else:
-                    clean_content, cleaned_query, raw_results = await search_service.execute_hybrid_search(
-                        **search_params, memory_config=memory_config
-                    )
-                # Keep structure for Verify/Retrieve_Summary compatibility
-                dup_databases = {
-                    "Query": cleaned_query,
-                    "Expansion_issue": [{
-                        "Query_small": cleaned_query,
-                        "Answer_Small": clean_content,
-                        "_intermediate": {
-                            "type": "search_result",
-                            "query": cleaned_query,
-                            "raw_results": raw_results,
-                            "index": 1,
-                            "total": 1
-                        }
-                    }]
-                }
-            except Exception as e:
-                logger.error(
-                    f"Retrieve: hybrid_search failed for query '{query}': {e}",
-                    exc_info=True
-                )
-                # Return empty results on failure
-                dup_databases = {
-                    "Query": query,
-                    "Expansion_issue": []
-                }
-        
-        logger.info(
-            f"Retrieval: {storage_type}--{user_rag_memory_id}--Query={dup_databases.get('Query', '')}, "
-            f"Expansion_issue count={len(dup_databases.get('Expansion_issue', []))}"
-        )
-        
-        # Build result with intermediate outputs
-        result = {
-            "context": dup_databases,
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id
-        }
-        
-        # Add intermediate outputs list if they exist
-        intermediate_outputs = dup_databases.get('_intermediate_outputs', [])
-        if intermediate_outputs:
-            result['_intermediates'] = intermediate_outputs
-            logger.info(f"Adding {len(intermediate_outputs)} intermediate outputs to result")
-        else:
-            logger.warning("No intermediate outputs found in dup_databases")
-        
-        return result
-        
-    except Exception as e:
-        logger.error(
-            f"Retrieve failed: {e}",
-            exc_info=True
-        )
-        return {
-            "context": {
-                "Query": "",
-                "Expansion_issue": []
-            },
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "error": str(e)
-        }
-        
-    finally:
-        # Log execution time
-        end = time.time()
-        try:
-            duration = end - start
-        except Exception:
-            duration = 0.0
-        log_time('Retrieval', duration)

api/app/core/memory/agent/mcp_server/tools/summary_tools.py

@@ -1,640 +0,0 @@
-"""
-Summary Tools for data summarization.
-
-This module contains MCP tools for summarizing retrieved data and generating responses.
-LLM clients are constructed from MemoryConfig when needed.
-"""
-
-import json
-import os
-import re
-import time
-
-from app.core.logging_config import get_agent_logger, log_time
-from app.core.memory.agent.mcp_server.mcp_instance import mcp
-from app.core.memory.agent.mcp_server.models.summary_models import (
-    RetrieveSummaryResponse,
-    SummaryResponse,
-)
-from app.core.memory.agent.mcp_server.server import get_context_resource
-from app.core.memory.agent.utils.messages_tool import (
-    Resolve_username,
-    Summary_messages_deal,
-)
-from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
-from app.core.rag.nlp.search import knowledge_retrieval
-from app.db import get_db_context
-from app.schemas.memory_config_schema import MemoryConfig
-from dotenv import load_dotenv
-from mcp.server.fastmcp import Context
-
-load_dotenv()
-logger = get_agent_logger(__name__)
-
-
-@mcp.tool()
-async def Summary(
-    ctx: Context,
-    context: str,
-    usermessages: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-    storage_type: str = "",
-    user_rag_memory_id: str = "",
-) -> dict:
-    """
-    Summarize the verified data.
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: JSON string containing verified data
-        usermessages: User messages identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type for the workspace (optional)
-        user_rag_memory_id: User RAG memory identifier (optional)
-        
-    Returns:
-        dict: Contains 'status' and 'summary_result'
-    """
-    start = time.time()
-
-    try:
-        # Extract services from context
-        template_service = get_context_resource(ctx, "template_service")
-        session_service = get_context_resource(ctx, "session_service")
-
-        # Get LLM client from memory_config
-        with get_db_context() as db:
-            factory = MemoryClientFactory(db)
-            llm_client = factory.get_llm_client_from_config(memory_config)
-        
-        # Resolve session ID
-        sessionid = Resolve_username(usermessages)
-        
-        # Process context to extract answer and query
-        answer_small, query = await Summary_messages_deal(context)
-
-        
-        start_time= time.time()
-        history = await session_service.get_history(sessionid, apply_id, group_id)
-        end_time=time.time()
-        logger.info(f"Retrieve_Summary-REDIS搜索：{end_time - start_time}")
-        data = {
-            "query": query,
-            "history": history,
-            "retrieve_info": answer_small
-        }
-
-    except Exception as e:
-        logger.error(
-            f"Summary: initialization failed: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "error",
-            "summary_result": "信息不足，无法回答"
-        }
-
-    try:
-        # Render template
-        system_prompt = await template_service.render_template(
-            template_name='summary_prompt.jinja2',
-            operation_name='summary',
-            data=data,
-            query=query
-        )
-    except Exception as e:
-        logger.error(
-            f"Template rendering failed for Summary: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "error",
-            "message": f"Prompt rendering failed: {str(e)}"
-        }
-
-    try:
-        # Call LLM with structured response
-        structured = await llm_client.response_structured(
-            messages=[{"role": "system", "content": system_prompt}],
-            response_model=SummaryResponse
-        )
-
-        aimessages = structured.query_answer or ""
-
-    except Exception as e:
-        logger.error(
-            f"LLM call failed for Summary: {e}",
-            exc_info=True
-        )
-        aimessages = ""
-
-    try:
-        # Save session
-        if aimessages != "":
-            await session_service.save_session(
-            user_id=sessionid,
-            query=query,
-            apply_id=apply_id,
-            group_id=group_id,
-            ai_response=aimessages
-        )
-        logger.info(f"sessionid: {aimessages} 写入成功")
-    except Exception as e:
-        logger.error(
-            f"sessionid: {sessionid} 写入失败，错误信息：{str(e)}",
-            exc_info=True
-        )
-        return {
-            "status": "error",
-            "message": str(e)
-        }
-
-    # Cleanup duplicate sessions
-    await session_service.cleanup_duplicates()
-
-    # Use fallback if empty
-    if aimessages == '':
-        aimessages = '信息不足，无法回答'
-
-    logger.info(f"Summary after verification: {aimessages}")
-
-    # Log execution time
-    end = time.time()
-    try:
-        duration = end - start
-    except Exception:
-        duration = 0.0
-    log_time('Summary', duration)
-
-    return {
-        "status": "success",
-        "summary_result": aimessages,
-        "storage_type": storage_type,
-        "user_rag_memory_id": user_rag_memory_id
-    }
-
-
-@mcp.tool()
-async def Retrieve_Summary(
-    ctx: Context,
-    context: dict,
-    usermessages: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-    storage_type: str = "",
-    user_rag_memory_id: str = "",
-) -> dict:
-    """
-    Summarize data directly from retrieval results.
-
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: Dictionary containing Query and Expansion_issue from Retrieve
-        usermessages: User messages identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type for the workspace (optional)
-        user_rag_memory_id: User RAG memory identifier (optional)
-
-    Returns:
-        dict: Contains 'status' and 'summary_result'
-    """
-    start = time.time()
-
-    try:
-        # Extract services from context
-        template_service = get_context_resource(ctx, "template_service")
-        session_service = get_context_resource(ctx, "session_service")
-
-        # Get LLM client from memory_config
-        with get_db_context() as db:
-            factory = MemoryClientFactory(db)
-            llm_client = factory.get_llm_client_from_config(memory_config)
-
-        # Resolve session ID
-        sessionid = Resolve_username(usermessages)
-
-
-
-        # Handle both 'content' and 'context' keys (LangGraph uses 'content')
-        logger.debug(f"Retrieve_Summary: raw context type={type(context)}, keys={list(context.keys()) if isinstance(context, dict) else 'N/A'}")
-        
-        if isinstance(context, dict):
-            if "content" in context:
-                inner = context["content"]
-                # If it's a JSON string, parse it
-                if isinstance(inner, str):
-                    try:
-                        parsed = json.loads(inner)
-                        logger.info("Retrieve_Summary: successfully parsed JSON")
-                    except json.JSONDecodeError:
-                        # Try unescaping first
-                        try:
-                            unescaped = inner.encode('utf-8').decode('unicode_escape')
-                            parsed = json.loads(unescaped)
-                            logger.info("Retrieve_Summary: parsed after unescaping")
-                        except (json.JSONDecodeError, UnicodeDecodeError) as e:
-                            logger.error(
-                                f"Retrieve_Summary: parsing failed even after unescape: {e}"
-                            )
-                            context_dict = {"Query": "", "Expansion_issue": []}
-                            parsed = None
-
-                    if parsed:
-                        # Check if parsed has 'context' wrapper
-                        if isinstance(parsed, dict) and "context" in parsed:
-                            context_dict = parsed["context"]
-                        else:
-                            context_dict = parsed
-                elif isinstance(inner, dict):
-                    context_dict = inner
-                else:
-                    context_dict = {"Query": "", "Expansion_issue": []}
-            elif "context" in context:
-                context_dict = context["context"] if isinstance(context["context"], dict) else context
-            else:
-                context_dict = context
-        else:
-            context_dict = {"Query": "", "Expansion_issue": []}
-
-        query = context_dict.get("Query", "")
-        expansion_issue = context_dict.get("Expansion_issue", [])
-        
-        logger.debug(f"Retrieve_Summary: query='{query}', expansion_issue count={len(expansion_issue)}")
-        logger.debug(f"Retrieve_Summary: expansion_issue={expansion_issue[:2] if expansion_issue else 'empty'}")
-
-        # Extract retrieve_info from expansion_issue
-        retrieve_info = []
-        for item in expansion_issue:
-            # Check for both Answer_Small and Answer_Small (typo) for backward compatibility
-            answer = None
-            if isinstance(item, dict):
-                if "Answer_Small" in item:
-                    answer = item["Answer_Small"]
-
-
-                if answer is not None:
-                    # Handle both string and list formats
-                    if isinstance(answer, list):
-                        # Join list of characters/strings into a single string
-                        retrieve_info.append(''.join(str(x) for x in answer))
-                    elif isinstance(answer, str):
-                        retrieve_info.append(answer)
-                    else:
-                        retrieve_info.append(str(answer))
-
-        # Join all retrieve_info into a single string
-        retrieve_info_str = '\n\n'.join(retrieve_info) if retrieve_info else ""
-
-        start_time=time.time()
-        history = await session_service.get_history(sessionid, apply_id, group_id)
-        # Override with empty list for now (as in original)
-        end_time=time.time()
-        logger.info(f"Retrieve_Summary-REDIS搜索：{end_time - start_time}")
-    except Exception as e:
-        logger.error(
-            f"Retrieve_Summary: initialization failed: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "error",
-            "summary_result": "信息不足，无法回答"
-        }
-
-    try:
-        # Render template
-        system_prompt = await template_service.render_template(
-            template_name='Retrieve_Summary_prompt.jinja2',
-            operation_name='retrieve_summary',
-            query=query,
-            history=history,
-            retrieve_info=retrieve_info_str
-        )
-    except Exception as e:
-        logger.error(
-            f"Template rendering failed for Retrieve_Summary: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "error",
-            "message": f"Prompt rendering failed: {str(e)}"
-        }
-
-    try:
-        # Call LLM with structured response
-        structured = await llm_client.response_structured(
-            messages=[{"role": "system", "content": system_prompt}],
-            response_model=RetrieveSummaryResponse
-        )
-
-        # Handle case where structured response might be None or incomplete
-        if structured and hasattr(structured, 'data') and structured.data:
-            aimessages = structured.data.query_answer or ""
-        else:
-            logger.warning("Structured response is None or incomplete, using default message")
-            aimessages = "信息不足，无法回答"
-
-
-        # Check for insufficient information response
-        if '信息不足，无法回答' not  in str(aimessages) or str(aimessages)!="":
-            # Save session
-            await session_service.save_session(
-                user_id=sessionid,
-                query=query,
-                apply_id=apply_id,
-                group_id=group_id,
-                ai_response=aimessages
-            )
-            logger.info(f"sessionid: {aimessages} 写入成功")
-    except Exception as e:
-        logger.error(
-            f"Retrieve_Summary: LLM call failed: {e}",
-            exc_info=True
-        )
-        aimessages = ""
-    # Cleanup duplicate sessions
-    await session_service.cleanup_duplicates()
-
-    # Use fallback if empty
-    if aimessages == '':
-        aimessages = '信息不足，无法回答'
-
-    logger.info(f"Summary after retrieval: {aimessages}")
-
-    # Log execution time
-    end = time.time()
-    try:
-        duration = end - start
-    except Exception:
-        duration = 0.0
-    log_time('Retrieval summary', duration)
-
-    # Emit intermediate output for frontend
-    return {
-        "status": "success",
-        "summary_result": aimessages,
-        "storage_type": storage_type,
-        "user_rag_memory_id": user_rag_memory_id,
-        "_intermediate": {
-            "type": "retrieval_summary",
-            "summary": aimessages,
-            "query": query,
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id
-        }
-    }
-
-
-@mcp.tool()
-async def Input_Summary(
-    ctx: Context,
-    context: str,
-    usermessages: str,
-    search_switch: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-    storage_type: str = "",
-    user_rag_memory_id: str = "",
-) -> dict:
-    """
-    Generate a quick summary for direct input without verification.
-
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: String containing the input sentence
-        usermessages: User messages identifier
-        search_switch: Search switch value for routing ('2' for summaries only)
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type for the workspace (e.g., 'rag', 'vector')
-        user_rag_memory_id: User RAG memory identifier
-
-    Returns:
-        dict: Contains 'query_answer' with the summary result
-    """
-    start = time.time()
-    logger.info(f"Input_Summary: storage_type={storage_type}, user_rag_memory_id={user_rag_memory_id}")
-
-    try:
-        # Extract services from context
-        session_service = get_context_resource(ctx, "session_service")
-        search_service = get_context_resource(ctx, "search_service")
-
-        # Resolve session ID
-        sessionid = Resolve_username(usermessages) or ""
-        sessionid = sessionid.replace('call_id_', '')
-
-        start_time=time.time()
-        history = await session_service.get_history(
-            str(sessionid),
-            str(apply_id),
-            str(group_id)
-        )
-        end_time=time.time()
-        logger.info(f"Input_Summary-REDIS搜索：{end_time - start_time}")
-        # Override with empty list for now (as in original)
-
-        # Log the raw context for debugging
-        logger.info(f"Input_Summary: Received context type={type(context)}, value={context[:200] if isinstance(context, str) else context}")
-
-        # Extract sentence from context
-        # Context can be a string or might contain the sentence in various formats
-        try:
-            # Try to parse as JSON first
-            if isinstance(context, str) and (context.startswith('{') or context.startswith('[')):
-                try:
-                    import json
-                    context_dict = json.loads(context)
-                    if isinstance(context_dict, dict):
-                        query = context_dict.get('sentence', context_dict.get('content', context))
-                    else:
-                        query = context
-                except json.JSONDecodeError:
-                    # Not valid JSON, try regex
-                    match = re.search(r"'sentence':\s*['\"]?(.*?)['\"]?\s*,", context)
-                    query = match.group(1) if match else context
-            else:
-                query = context
-        except Exception as e:
-            logger.warning(f"Failed to extract query from context: {e}")
-            query = context
-
-        # Clean query
-        query = str(query).strip().strip("\"'")
-
-        logger.debug(f"Input_Summary: Extracted query='{query}' from context type={type(context)}")
-
-        # Execute search based on search_switch and storage_type
-        try:
-            logger.info(f"search_switch: {search_switch}, storage_type: {storage_type}")
-
-            # Prepare search parameters based on storage type
-            search_params = {
-                "group_id": group_id,
-                "question": query,
-                "return_raw_results": True
-            }
-
-            # Add storage-specific parameters
-
-            # Retrieval
-            if search_switch == '2':
-                search_params["include"] = ["summaries"]
-                if storage_type == "rag" and user_rag_memory_id:
-                    raw_results = []
-                    retrieve_info = ""
-                    kb_config={
-                        "knowledge_bases": [
-                            {
-                                "kb_id": user_rag_memory_id,
-                                "similarity_threshold": 0.7,
-                                "vector_similarity_weight": 0.5,
-                                "top_k": 10,
-                                "retrieve_type": "participle"
-                            }
-                        ],
-                        "merge_strategy": "weight",
-                        "reranker_id":os.getenv('reranker_id'),
-                        "reranker_top_k": 10
-                    }
-
-                    retrieve_chunks_result = knowledge_retrieval(query, kb_config,[str(group_id)])
-                    try:
-                        retrieval_knowledge = [i.page_content for i in retrieve_chunks_result]
-                        retrieve_info = '\n\n'.join(retrieval_knowledge)
-                        raw_results=[retrieve_info]
-                        logger.info(f"Input_Summary: Using RAG storage with memory_id={user_rag_memory_id}")
-                    except:
-                        retrieve_info=''
-                        raw_results=['']
-                        logger.info(f"No content retrieved from knowledge base: {user_rag_memory_id}")
-                else:
-                    retrieve_info, question, raw_results = await search_service.execute_hybrid_search(
-                        **search_params, memory_config=memory_config
-                    )
-                logger.info("Input_Summary: Using summary for retrieval")
-            else:
-                retrieve_info, question, raw_results = await search_service.execute_hybrid_search(
-                    **search_params, memory_config=memory_config
-                )
-
-        except Exception as e:
-            logger.error(
-                f"Input_Summary: hybrid_search failed, using empty results: {e}",
-                exc_info=True
-            )
-            retrieve_info, question, raw_results = "", query, []
-
-        # Return retrieved information directly without LLM processing
-        # Use the raw retrieved info as the answer
-        aimessages = retrieve_info if retrieve_info else "信息不足，无法回答"
-
-        logger.info(f"Quick answer (no LLM): {storage_type}--{user_rag_memory_id}--{aimessages[:500]}...")
-
-        # Emit intermediate output for frontend
-        return {
-            "status": "success",
-            "summary_result": aimessages,
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "_intermediate": {
-                "type": "input_summary",
-                "title": "快速答案",
-                "summary": aimessages,
-                "query": query,
-                "raw_results": raw_results,
-                "search_mode": "quick_search",
-                "storage_type": storage_type,
-                "user_rag_memory_id": user_rag_memory_id
-            }
-        }
-
-    except Exception as e:
-        logger.error(
-            f"Input_Summary failed: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "fail",
-            "summary_result": "信息不足，无法回答",
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "error": str(e)
-        }
-
-    finally:
-        # Log execution time
-        end = time.time()
-        try:
-            duration = end - start
-        except Exception:
-            duration = 0.0
-        log_time('Retrieval', duration)
-
-
-@mcp.tool()
-async def Summary_fails(
-    ctx: Context,
-    context: str,
-    usermessages: str,
-    apply_id: str,
-    group_id: str,
-    storage_type: str = "",
-    user_rag_memory_id: str = ""
-) -> dict:
-    """
-    Handle workflow failure when summary cannot be generated.
-
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: Failure context string
-        usermessages: User messages identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        storage_type: Storage type for the workspace (optional)
-        user_rag_memory_id: User RAG memory identifier (optional)
-
-    Returns:
-        dict: Contains 'query_answer' with failure message
-    """
-    try:
-        # Extract services from context
-        session_service = get_context_resource(ctx, 'session_service')
-
-        # Parse session ID from usermessages
-        usermessages_parts = usermessages.split('_')[1:]
-        sessionid = '_'.join(usermessages_parts[:-1])
-
-        # Cleanup duplicate sessions
-        await session_service.cleanup_duplicates()
-
-        logger.info("没有相关数据")
-        logger.debug(f"Summary_fails called with apply_id: {apply_id}, group_id: {group_id}")
-        
-        return {
-            "status": "success",
-            "summary_result": "没有相关数据",
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id
-        }
-        
-    except Exception as e:
-        logger.error(
-            f"Summary_fails failed: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "fail",
-            "summary_result": "没有相关数据",
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "error": str(e)
-        }

api/app/core/memory/agent/mcp_server/tools/verification_tools.py

@@ -1,174 +0,0 @@
-"""
-Verification Tools for data verification.
-
-This module contains MCP tools for verifying retrieved data.
-"""
-import time
-
-from app.core.logging_config import get_agent_logger, log_time
-from app.core.memory.agent.mcp_server.mcp_instance import mcp
-from app.core.memory.agent.mcp_server.server import get_context_resource
-from app.core.memory.agent.utils.llm_tools import PROJECT_ROOT_
-from app.core.memory.agent.utils.messages_tool import (
-    Resolve_username,
-    Retrieve_verify_tool_messages_deal,
-    Verify_messages_deal,
-)
-from app.core.memory.agent.utils.verify_tool import VerifyTool
-from app.schemas.memory_config_schema import MemoryConfig
-from jinja2 import Template
-from mcp.server.fastmcp import Context
-
-logger = get_agent_logger(__name__)
-
-
-@mcp.tool()
-async def Verify(
-    ctx: Context,
-    context: dict,
-    usermessages: str,
-    apply_id: str,
-    group_id: str,
-    memory_config: MemoryConfig,
-    storage_type: str = "",
-    user_rag_memory_id: str = ""
-) -> dict:
-    """
-    Verify the retrieved data.
-    
-    Args:
-        ctx: FastMCP context for dependency injection
-        context: Dictionary containing query and expansion issues
-        usermessages: User messages identifier
-        apply_id: Application identifier
-        group_id: Group identifier
-        memory_config: MemoryConfig object containing all configuration
-        storage_type: Storage type for the workspace (optional)
-        user_rag_memory_id: User RAG memory identifier (optional)
-        
-    Returns:
-        dict: Contains 'status' and 'verified_data' with verification results
-    """
-    start = time.time()
-
-
-    try:
-        # Extract services from context
-        session_service = get_context_resource(ctx, 'session_service')
-        
-        # Load verification prompt template
-        file_path = PROJECT_ROOT_ + '/agent/utils/prompt/split_verify_prompt.jinja2'
-        
-        # Read template file directly (VerifyTool expects raw template content)
-        from app.core.memory.agent.utils.messages_tool import read_template_file
-        system_prompt = await read_template_file(file_path)
-
-
-        
-        # Resolve session ID
-        sessionid = Resolve_username(usermessages)
-        
-        # Get conversation history
-        history = await session_service.get_history(sessionid, apply_id, group_id)
-
-        template = Template(system_prompt)
-        system_prompt = template.render(history=history, sentence=context)
-        
-        # Process context to extract query and results
-        Query_small, Result_small, query = await Verify_messages_deal(context)
-        
-        # Build query list for verification
-        query_list = []
-        for query_small, anser in zip(Query_small, Result_small, strict=False):
-            query_list.append({
-                'Query_small': query_small,
-                'Answer_Small': anser
-            })
-        
-        messages = {
-            "Query": query,
-            "Expansion_issue": query_list
-        }
-
-
-        
-        # Call verification workflow with LLM model ID from memory_config
-        verify_tool = VerifyTool(
-            system_prompt=system_prompt,
-            verify_data=messages,
-            llm_model_id=str(memory_config.llm_model_id)
-        )
-        verify_result = await verify_tool.verify()
-        
-        # Parse LLM verification result with error handling
-        try:
-            messages_deal = await Retrieve_verify_tool_messages_deal(
-                verify_result,
-                history,
-                query
-            )
-        except Exception as e:
-            logger.error(
-                f"Retrieve_verify_tool_messages_deal parsing failed: {e}",
-                exc_info=True
-            )
-            # Fallback to avoid 500 errors
-            messages_deal = {
-                "data": {
-                    "query": query,
-                    "expansion_issue": []
-                },
-                "split_result": "failed",
-                "reason": str(e),
-                "history": history,
-            }
-        
-        logger.info(f"Verification result: {messages_deal}")
-        
-        # Emit intermediate output for frontend
-        return {
-            "status": "success",
-            "verified_data": messages_deal,
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "_intermediate": {
-                "type": "verification",
-                "title": "Data Verification",
-                "result": messages_deal.get("split_result", "unknown"),
-                "reason": messages_deal.get("reason", ""),
-                "query": query,
-                "verified_count": len(query_list),
-                "storage_type": storage_type,
-                "user_rag_memory_id": user_rag_memory_id
-            }
-        }
-        
-    except Exception as e:
-        logger.error(
-            f"Verify failed: {e}",
-            exc_info=True
-        )
-        return {
-            "status": "error",
-            "message": str(e),
-            "storage_type": storage_type,
-            "user_rag_memory_id": user_rag_memory_id,
-            "verified_data": {
-                "data": {
-                    "query": "",
-                    "expansion_issue": []
-                },
-                "split_result": "failed",
-                "reason": str(e),
-                "history": [],
-            }
-        }
-        
-    finally:
-        # Log execution time
-        end = time.time()
-        try:
-            duration = end - start
-        except Exception:
-            duration = 0.0
-        log_time('Verification', duration)

api/app/core/memory/agent/models/verification_models.py

@@ -0,0 +1,32 @@
+"""Pydantic models for verification operations."""
+
+from typing import List, Optional, Dict, Any
+from pydantic import BaseModel, Field
+
+
+class VerificationItem(BaseModel):
+    """Individual verification item for a query-answer pair."""
+    
+    query_small: str = Field(..., description="子问题")
+    answer_small: str = Field(..., description="子问题的回答")
+    status: str = Field(..., description="验证状态：True 或 False")
+    query_answer: str = Field(..., description="问题的答案（与 answer_small 相同）")
+
+
+class VerificationResult(BaseModel):
+    """Result model for verification operation."""
+    
+    query: str = Field(..., description="原始查询问题")
+    history: List[Dict[str, Any]] = Field(default_factory=list, description="历史对话记录")
+    expansion_issue: List[VerificationItem] = Field(
+        default_factory=list, 
+        description="验证后的数据列表，包含所有通过验证的问答对"
+    )
+    split_result: str = Field(
+        ..., 
+        description="验证结果状态：success（expansion_issue 非空）或 failed（expansion_issue 为空）"
+    )
+    reason: Optional[str] = Field(
+        None, 
+        description="验证结果的说明和分析"
+    )

api/app/core/memory/agent/models/write_aggregate_model.py

@@ -0,0 +1,28 @@
+"""Pydantic models for write aggregate judgment operations."""
+
+from typing import List, Union
+from pydantic import BaseModel, Field
+
+
+class MessageItem(BaseModel):
+    """Individual message item in conversation."""
+    
+    role: str = Field(..., description="角色：user 或 assistant")
+    content: str = Field(..., description="消息内容")
+
+
+class WriteAggregateResponse(BaseModel):
+    """Response model for aggregate judgment containing judgment result and output."""
+    
+    is_same_event: bool = Field(
+        ..., 
+        description="是否是同一事件。True表示是同一事件，False表示不同事件"
+    )
+    output: Union[List[MessageItem], bool] = Field(
+        ..., 
+        description="如果is_same_event为True，返回False；如果is_same_event为False，返回消息列表"
+    )
+
+
+# 为了保持向后兼容，保留旧的类名作为别名
+WriteAggregateModel = WriteAggregateResponse

api/app/core/memory/agent/multimodal/oss_picture.py

@@ -1,114 +0,0 @@
-import os
-import sys
-import traceback
-
-import requests
-
-# from qcloud_cos import CosConfig, CosS3Client
-# from qcloud_cos.cos_exception import CosClientError, CosServiceError
-
-# from config.paths import BASE_DIR
-BASE_DIR = os.path.dirname(os.path.realpath(sys.argv[0]))
-
-class OSSUploader:
-    """对象存储文件上传工具类"""
-
-    def __init__(self, env):
-        api = {
-            "test": "https://testlingqi.redbearai.com/api/user/file/common/upload/v2/anon",
-            "prod": "https://lingqi.redbearai.com/api/user/file/common/upload/v2/anon"
-        }
-        self.api = api.get(env, "https://testlingqi.redbearai.com/api/user/file/common/upload/v2/anon")
-        self.privacy = "false"
-        self.headers = {
-            "User-Agent": 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) '
-                          'AppleWebKit/537.36 (KHTML, like Gecko)'
-                          ' Chrome/133.0.6833.84 Safari/537.36'
-        }
-
-    @staticmethod
-    def _generate_object_key(file_path, prefix='xhs_'):
-        """
-        生成对象存储的Key
-
-        :param file_path: 本地文件路径
-        :param prefix: 存储前缀，用于分类存储
-        :return: 生成的对象Key
-        """
-        # 文件md5值.后缀名
-        filename = os.path.basename(file_path)
-        filename = f"{filename}"
-
-        # 组合成完整的对象Key
-        return f"{prefix}{filename}"
-
-    def upload_image(self, file_name, prefix='jd_'):
-        """
-        上传文件到COS并返回可访问的URL
-
-        :param file_url: 文件路径
-        :param file_name: 文件名称
-        :param media_type: 文件类型
-        :param prefix: 存储前缀，用于分类存储
-        :return: 文件访问URL
-        """
-        # 检查文件是否存在
-
-
-
-        file_path = os.path.join(BASE_DIR, file_name)
-
-        # response = requests.get(url, headers=self.headers, stream=True)
-
-        # if response.status_code == 200:
-        #     with open(file_path, "wb") as f:
-        #         for chunk in response.iter_content(1024):  # 分块写入，避免内存占用过大
-        #             f.write(chunk)
-        # else:
-        #     raise Exception(f"文件下载失败,{file_name}")
-
-        # 生成对象Key
-        object_key = self._generate_object_key(file_path, prefix +file_name.split('.')[-1])
-
-        try:
-            upload_response = requests.post(
-                self.api,
-                data={
-                    "privacy": self.privacy,
-                    "fileName": object_key,
-                }
-            )
-
-            if upload_response.status_code != 200:
-                raise Exception('上传接口请求失败')
-            resp = upload_response.json()
-            name = resp["data"]["name"]
-            file_url = resp["data"]["path"]
-            policy = resp["data"]["policy"]
-            with open(file_path, 'rb') as f:
-                oss_push_resp = requests.post(
-                    policy["host"],
-                    files={
-                        "key": policy["dir"],
-                        "OSSAccessKeyId": policy["accessid"],
-                        "name": name,
-                        "policy": policy["policy"],
-                        "success_action_status": 200,
-                        "signature": policy["signature"],
-                        "file": f,
-                    }
-                )
-                if oss_push_resp.status_code == 200:
-                    return file_url
-            raise Exception("OSS上传失败")
-        except Exception:
-            raise Exception(f"上传失败: \n{traceback.format_exc()}")
-        finally:
-            print('success')
-            # os.remove(file_path)
-
-
-if __name__ == '__main__':
-    cos_uploader = OSSUploader("prod")
-    url =cos_uploader.upload_image('./example01.jpg')
-    print(url)

api/app/core/memory/agent/multimodal/speech_model.py

@@ -1,121 +0,0 @@
-import asyncio
-import re
-
-from app.core.memory.agent.utils.llm_tools import PROJECT_ROOT_, picture_model_requests,Picture_recognize, Voice_recognize
-from app.core.memory.agent.utils.messages_tool import read_template_file
-
-import requests
-import json
-import os
-import time
-# file_urls = [
-#     "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav",
-#     "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_male2.wav",
-# ]
-class Vico_recognition:
-    def __init__(self,file_urls):
-        self.api_key=''
-        self.backend_model_name=''
-        self.api_base=''
-        self.file_urls=file_urls
-
-    # 提交文件转写任务，包含待转写文件url列表
-    async  def submit_task(self) -> str:
-        self.api_key, self.backend_model_name, self.api_base =await Voice_recognize()
-
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json",
-            "X-DashScope-Async": "enable",
-        }
-        data = {
-            "model": self.backend_model_name,
-            "input": {"file_urls": self.file_urls},
-            "parameters": {
-                "channel_id": [0],
-                "vocabulary_id": "vocab-Xxxx",
-            },
-        }
-        # 录音文件转写服务url
-        service_url = (
-            "https://dashscope.aliyuncs.com/api/v1/services/audio/asr/transcription"
-        )
-        response = requests.post(
-            service_url, headers=headers, data=json.dumps(data)
-        )
-
-        # 打印响应内容
-        if response.status_code == 200:
-            return response.json()["output"]["task_id"]
-        else:
-            print("task failed!")
-            print(response.json())
-            return None
-
-    async def download_transcription_result(self, transcription_url):
-        """
-        Args:
-            transcription_url (str): 转写结果文件URL
-        Returns:
-            dict: 转写结果内容
-        """
-        try:
-            response = requests.get(transcription_url)
-            response.raise_for_status()
-            return response.json()
-        except Exception as e:
-            print(f"下载转写结果失败: {e}")
-            return None
-
-    # 循环查询任务状态直到成功
-    async def wait_for_complete(self,task_id):
-        self.api_key, self.backend_model_name, self.api_base = await Voice_recognize()
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json",
-            "X-DashScope-Async": "enable",
-        }
-
-        pending = True
-        while pending:
-            # 查询任务状态服务url
-            service_url = f"https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}"
-            response = requests.post(
-                service_url, headers=headers
-            )
-            if response.status_code == 200:
-                status = response.json()['output']['task_status']
-                if status == 'SUCCEEDED':
-                    print("task succeeded!")
-                    pending = False
-                    return response.json()['output']['results']
-                elif status == 'RUNNING' or status == 'PENDING':
-                    pass
-                else:
-                    print("task failed!")
-                    pending = False
-            else:
-                print("query failed!")
-                pending = False
-            time.sleep(0.1)
-    async def run(self):
-        self.api_key, self.backend_model_name, self.api_base = await Voice_recognize()
-        task_id=await self.submit_task()
-        result=await self.wait_for_complete(task_id)
-        result_context=[]
-        for  i in result:
-            transcription_url=i['transcription_url']
-            print(f"转写URL: {transcription_url}")
-
-            # 下载并打印转写内容
-            content = await self.download_transcription_result(transcription_url)
-            if content:
-                content=json.dumps(content, indent=2, ensure_ascii=False)
-                context=re.findall(r'"text": "(.*?)"', content)
-                result_context.append(context[0])
-        result=''.join(result_context)
-        return (result)
-
-
-
-

api/app/core/memory/agent/services/optimized_llm_service.py

@@ -0,0 +1,277 @@
+"""
+优化的LLM服务类，用于压缩和统一LLM调用
+"""
+
+import asyncio
+from typing import Any, Dict, List, Optional, Type, TypeVar, Union
+from pydantic import BaseModel
+from sqlalchemy.orm import Session
+
+from app.core.logging_config import get_agent_logger
+from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
+from app.core.memory.llm_tools.openai_client import OpenAIClient
+
+T = TypeVar('T', bound=BaseModel)
+
+logger = get_agent_logger(__name__)
+
+
+class OptimizedLLMService:
+    """
+    优化的LLM服务类，提供统一的LLM调用接口
+    
+    特性：
+    1. 客户端复用 - 避免重复创建LLM客户端
+    2. 批量处理 - 支持并发处理多个请求
+    3. 错误处理 - 统一的错误处理和降级策略
+    4. 性能优化 - 缓存和连接池优化
+    """
+    
+    def __init__(self, db_session: Session):
+        self.db_session = db_session
+        self.client_factory = MemoryClientFactory(db_session)
+        self._client_cache: Dict[str, OpenAIClient] = {}
+        
+    def _get_cached_client(self, llm_model_id: str) -> OpenAIClient:
+        """获取缓存的LLM客户端，避免重复创建"""
+        if llm_model_id not in self._client_cache:
+            self._client_cache[llm_model_id] = self.client_factory.get_llm_client(llm_model_id)
+        return self._client_cache[llm_model_id]
+    
+    async def structured_response(
+        self,
+        llm_model_id: str,
+        system_prompt: str,
+        response_model: Type[T],
+        user_message: Optional[str] = None,
+        fallback_value: Optional[Any] = None
+    ) -> T:
+        """
+        统一的结构化响应接口
+        
+        Args:
+            llm_model_id: LLM模型ID
+            system_prompt: 系统提示词
+            response_model: 响应模型类
+            user_message: 用户消息（可选）
+            fallback_value: 失败时的降级值
+            
+        Returns:
+            结构化响应对象
+        """
+        try:
+            llm_client = self._get_cached_client(llm_model_id)
+            
+            messages = [{"role": "system", "content": system_prompt}]
+            if user_message:
+                messages.append({"role": "user", "content": user_message})
+            
+            logger.debug(f"LLM调用: model={llm_model_id}, prompt_length={len(system_prompt)}")
+            
+            structured = await llm_client.response_structured(
+                messages=messages,
+                response_model=response_model
+            )
+            
+            if structured is None:
+                logger.warning(f"LLM返回None，使用降级值")
+                return self._create_fallback_response(response_model, fallback_value)
+            
+            return structured
+            
+        except Exception as e:
+            logger.error(f"结构化响应失败: {e}", exc_info=True)
+            return self._create_fallback_response(response_model, fallback_value)
+    
+    async def batch_structured_response(
+        self,
+        llm_model_id: str,
+        requests: List[Dict[str, Any]],
+        response_model: Type[T],
+        max_concurrent: int = 5
+    ) -> List[T]:
+        """
+        批量处理结构化响应
+        
+        Args:
+            llm_model_id: LLM模型ID
+            requests: 请求列表，每个请求包含system_prompt等参数
+            response_model: 响应模型类
+            max_concurrent: 最大并发数
+            
+        Returns:
+            结构化响应列表
+        """
+        semaphore = asyncio.Semaphore(max_concurrent)
+        
+        async def process_single_request(request: Dict[str, Any]) -> T:
+            async with semaphore:
+                return await self.structured_response(
+                    llm_model_id=llm_model_id,
+                    system_prompt=request.get('system_prompt', ''),
+                    response_model=response_model,
+                    user_message=request.get('user_message'),
+                    fallback_value=request.get('fallback_value')
+                )
+        
+        tasks = [process_single_request(req) for req in requests]
+        return await asyncio.gather(*tasks)
+    
+    async def simple_response(
+        self,
+        llm_model_id: str,
+        system_prompt: str,
+        user_message: Optional[str] = None,
+        fallback_message: str = "信息不足，无法回答"
+    ) -> str:
+        """
+        简单的文本响应接口
+        
+        Args:
+            llm_model_id: LLM模型ID
+            system_prompt: 系统提示词
+            user_message: 用户消息（可选）
+            fallback_message: 失败时的降级消息
+            
+        Returns:
+            响应文本
+        """
+        try:
+            llm_client = self._get_cached_client(llm_model_id)
+            
+            messages = [{"role": "system", "content": system_prompt}]
+            if user_message:
+                messages.append({"role": "user", "content": user_message})
+            
+            response = await llm_client.response(messages=messages)
+            
+            if not response or not response.strip():
+                return fallback_message
+            
+            return response.strip()
+            
+        except Exception as e:
+            logger.error(f"简单响应失败: {e}", exc_info=True)
+            return fallback_message
+    
+    def _create_fallback_response(self, response_model: Type[T], fallback_value: Optional[Any]) -> T:
+        """创建降级响应"""
+        try:
+            if fallback_value is not None:
+                if isinstance(fallback_value, response_model):
+                    return fallback_value
+                elif isinstance(fallback_value, dict):
+                    return response_model(**fallback_value)
+            
+            # 尝试创建空的响应模型
+            if hasattr(response_model, 'root'):
+                # RootModel类型
+                return response_model([])
+            else:
+                # 普通BaseModel类型
+                return response_model()
+                
+        except Exception as e:
+            logger.error(f"创建降级响应失败: {e}")
+            # 最后的降级策略
+            if hasattr(response_model, 'root'):
+                return response_model([])
+            else:
+                return response_model()
+    
+    def clear_cache(self):
+        """清理客户端缓存"""
+        self._client_cache.clear()
+
+
+class LLMServiceMixin:
+    """
+    LLM服务混入类，为节点提供便捷的LLM调用方法
+    """
+    
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self._llm_service: Optional[OptimizedLLMService] = None
+    
+    def get_llm_service(self, db_session: Session) -> OptimizedLLMService:
+        """获取LLM服务实例"""
+        if self._llm_service is None:
+            self._llm_service = OptimizedLLMService(db_session)
+        return self._llm_service
+    
+    async def call_llm_structured(
+        self,
+        state: Dict[str, Any],
+        db_session: Session,
+        system_prompt: str,
+        response_model: Type[T],
+        user_message: Optional[str] = None,
+        fallback_value: Optional[Any] = None
+    ) -> T:
+        """
+        便捷的结构化LLM调用方法
+        
+        Args:
+            state: 状态字典，包含memory_config
+            db_session: 数据库会话
+            system_prompt: 系统提示词
+            response_model: 响应模型类
+            user_message: 用户消息（可选）
+            fallback_value: 失败时的降级值
+            
+        Returns:
+            结构化响应对象
+        """
+        memory_config = state.get('memory_config')
+        if not memory_config:
+            raise ValueError("State中缺少memory_config")
+        
+        llm_model_id = memory_config.llm_model_id
+        if not llm_model_id:
+            raise ValueError("Memory config中缺少llm_model_id")
+        
+        llm_service = self.get_llm_service(db_session)
+        return await llm_service.structured_response(
+            llm_model_id=llm_model_id,
+            system_prompt=system_prompt,
+            response_model=response_model,
+            user_message=user_message,
+            fallback_value=fallback_value
+        )
+    
+    async def call_llm_simple(
+        self,
+        state: Dict[str, Any],
+        db_session: Session,
+        system_prompt: str,
+        user_message: Optional[str] = None,
+        fallback_message: str = "信息不足，无法回答"
+    ) -> str:
+        """
+        便捷的简单LLM调用方法
+        
+        Args:
+            state: 状态字典，包含memory_config
+            db_session: 数据库会话
+            system_prompt: 系统提示词
+            user_message: 用户消息（可选）
+            fallback_message: 失败时的降级消息
+            
+        Returns:
+            响应文本
+        """
+        memory_config = state.get('memory_config')
+        if not memory_config:
+            raise ValueError("State中缺少memory_config")
+        
+        llm_model_id = memory_config.llm_model_id
+        if not llm_model_id:
+            raise ValueError("Memory config中缺少llm_model_id")
+        
+        llm_service = self.get_llm_service(db_session)
+        return await llm_service.simple_response(
+            llm_model_id=llm_model_id,
+            system_prompt=system_prompt,
+            user_message=user_message,
+            fallback_message=fallback_message
+        )

api/app/core/memory/agent/services/parameter_builder.py

@@ -4,22 +4,19 @@ Parameter Builder for constructing tool call arguments.
 This service provides tool-specific parameter transformation logic
 to build correct arguments for each tool type.
 """
-
 from typing import Any, Dict, Optional
-
 from app.core.logging_config import get_agent_logger
-from app.schemas.memory_config_schema import MemoryConfig
 
 logger = get_agent_logger(__name__)
 
 
 class ParameterBuilder:
     """Service for building tool call arguments based on tool type."""
-
+    
     def __init__(self):
         """Initialize the parameter builder."""
         logger.info("ParameterBuilder initialized")
-
+    
     def build_tool_args(
         self,
         tool_name: str,
@@ -27,10 +24,9 @@ class ParameterBuilder:
         tool_call_id: str,
         search_switch: str,
         apply_id: str,
-        group_id: str,
-        memory_config: MemoryConfig,
+        end_user_id: str,
         storage_type: Optional[str] = None,
-        user_rag_memory_id: Optional[str] = None,
+        user_rag_memory_id: Optional[str] = None
     ) -> Dict[str, Any]:
         """
         Build tool arguments based on tool type.
@@ -48,8 +44,7 @@ class ParameterBuilder:
             tool_call_id: Extracted tool call identifier
             search_switch: Search routing parameter
             apply_id: Application identifier
-            group_id: Group identifier
-            memory_config: MemoryConfig object containing all configuration
+            end_user_id: Group identifier
             storage_type: Storage type for the workspace (optional)
             user_rag_memory_id: User RAG memory ID for knowledge base retrieval (optional)
             
@@ -60,19 +55,18 @@ class ParameterBuilder:
         base_args = {
             "usermessages": tool_call_id,
             "apply_id": apply_id,
-            "group_id": group_id,
-            "memory_config": memory_config,
+            "end_user_id": end_user_id
         }
-
+        
         # Always add storage_type and user_rag_memory_id (with defaults if None)
         base_args["storage_type"] = storage_type if storage_type is not None else ""
         base_args["user_rag_memory_id"] = user_rag_memory_id if user_rag_memory_id is not None else ""
         
         # Tool-specific argument construction
-        if tool_name in ["Verify", "Summary", "Summary_fails", "Retrieve_Summary", "Problem_Extension"]:
-            # These tools expect dict context
+        if tool_name in ["Verify","Summary", "Summary_fails",'Retrieve_Summary']:
+            # Verify expects dict context
             return {
-                "context": content if isinstance(content, dict) else {"content": content},
+                "context": content if isinstance(content, dict) else {},
                 **base_args
             }
 

api/app/core/memory/agent/services/search_service.py

@@ -4,31 +4,21 @@ Search Service for executing hybrid search and processing results.
 This service provides clean search result processing with content extraction
 and deduplication.
 """
-
-from typing import TYPE_CHECKING, List, Optional, Tuple
+from typing import List, Tuple, Optional
 
 from app.core.logging_config import get_agent_logger
 from app.core.memory.src.search import run_hybrid_search
 from app.core.memory.utils.data.text_utils import escape_lucene_query
 
-if TYPE_CHECKING:
-    from app.schemas.memory_config_schema import MemoryConfig
 
 logger = get_agent_logger(__name__)
 
 
 class SearchService:
     """Service for executing hybrid search and processing results."""
-
-    def __init__(self, memory_config: "MemoryConfig" = None):
-        """
-        Initialize the search service.
-        
-        Args:
-            memory_config: Optional MemoryConfig for embedding model configuration.
-                          If not provided, must be passed to execute_hybrid_search.
-        """
-        self.memory_config = memory_config
+    
+    def __init__(self):
+        """Initialize the search service."""
         logger.info("SearchService initialized")
     
     def extract_content_from_result(self, result: dict) -> str:
@@ -101,61 +91,51 @@ class SearchService:
     
     async def execute_hybrid_search(
         self,
-        group_id: str,
+        end_user_id: str,
         question: str,
-        limit: int = 15,
+        limit: int = 5,
         search_type: str = "hybrid",
         include: Optional[List[str]] = None,
-        rerank_alpha: float = 0.6,
-        activation_boost_factor: float = 0.8,
+        rerank_alpha: float = 0.4,
         output_path: str = "search_results.json",
         return_raw_results: bool = False,
-        memory_config: "MemoryConfig" = None,
+        memory_config = None
     ) -> Tuple[str, str, Optional[dict]]:
         """
-        Execute hybrid search with two-stage ranking.
-        
-        Stage 1: Filter by content relevance (BM25 + Embedding)
-        Stage 2: Rerank by activation values (ACTR)
+        Execute hybrid search and return clean content.
         
         Args:
-            group_id: Group identifier for filtering
+            end_user_id: Group identifier for filtering results
             question: Search query text
-            limit: Max results per category (default: 15)
-            search_type: "hybrid", "keyword", or "embedding" (default: "hybrid")
-            include: Result types (default: ["statements", "chunks", "entities", "summaries"])
-            rerank_alpha: BM25 weight (default: 0.6)
-            activation_boost_factor: Activation impact on memory strength (default: 0.8)
-            output_path: JSON output path (default: "search_results.json")
-            return_raw_results: Return full metadata (default: False)
-            memory_config: MemoryConfig for embedding model
+            limit: Maximum number of results to return (default: 5)
+            search_type: Type of search - "hybrid", "keyword", or "embedding" (default: "hybrid")
+            include: List of result types to include (default: ["statements", "chunks", "entities", "summaries"])
+            rerank_alpha: Weight for BM25 scores in reranking (default: 0.4)
+            output_path: Path to save search results (default: "search_results.json")
+            return_raw_results: If True, also return the raw search results as third element (default: False)
+            memory_config: Memory configuration object (required)
         
         Returns:
-            Tuple[str, str, Optional[dict]]: (clean_content, cleaned_query, raw_results)
+            Tuple of (clean_content, cleaned_query, raw_results)
+            raw_results is None if return_raw_results=False
         """
         if include is None:
             include = ["statements", "chunks", "entities", "summaries"]
-
-        # Use provided memory_config or fall back to instance config
-        config = memory_config or self.memory_config
-        if not config:
-            raise ValueError("memory_config is required for search - either pass it to __init__ or execute_hybrid_search")
-
+        
         # Clean query
         cleaned_query = self.clean_query(question)
-
+        
         try:
-            # Execute search using memory_config
+            # Execute search
             answer = await run_hybrid_search(
                 query_text=cleaned_query,
                 search_type=search_type,
-                group_id=group_id,
+                end_user_id=end_user_id,
                 limit=limit,
                 include=include,
                 output_path=output_path,
-                memory_config=config,
-                rerank_alpha=rerank_alpha,
-                activation_boost_factor=activation_boost_factor,
+                memory_config=memory_config,
+                rerank_alpha=rerank_alpha
             )
             
             # Extract results based on search type and include parameter
@@ -206,7 +186,7 @@ class SearchService:
             
         except Exception as e:
             logger.error(
-                f"Search failed for query '{question}' in group '{group_id}': {e}",
+                f"Search failed for query '{question}' in group '{end_user_id}': {e}",
                 exc_info=True
             )
             # Return empty results on failure

api/app/core/memory/agent/services/session_service.py

@@ -59,7 +59,7 @@ class SessionService:
         self,
         user_id: str,
         apply_id: str,
-        group_id: str
+        end_user_id: str
     ) -> List[dict]:
         """
         Retrieve conversation history from Redis.
@@ -67,20 +67,20 @@ class SessionService:
         Args:
             user_id: User identifier
             apply_id: Application identifier
-            group_id: Group identifier
+            end_user_id: Group identifier
             
         Returns:
             List of conversation history items with Query and Answer keys
             Returns empty list if no history found or on error
         """
         try:
-            history = self.store.find_user_apply_group(user_id, apply_id, group_id)
+            history = self.store.find_user_apply_group(user_id, apply_id, end_user_id)
             
             # Validate history structure
             if not isinstance(history, list):
                 logger.warning(
                     f"Invalid history format for user {user_id}, "
-                    f"apply {apply_id}, group {group_id}: expected list, got {type(history)}"
+                    f"apply {apply_id}, group {end_user_id}: expected list, got {type(history)}"
                 )
                 return []
             
@@ -89,7 +89,7 @@ class SessionService:
         except Exception as e:
             logger.error(
                 f"Failed to retrieve history for user {user_id}, "
-                f"apply {apply_id}, group {group_id}: {e}",
+                f"apply {apply_id}, group {end_user_id}: {e}",
                 exc_info=True
             )
             # Return empty list on error to allow execution to continue
@@ -100,7 +100,7 @@ class SessionService:
         user_id: str,
         query: str,
         apply_id: str,
-        group_id: str,
+        end_user_id: str,
         ai_response: str
     ) -> Optional[str]:
         """
@@ -110,7 +110,7 @@ class SessionService:
             user_id: User identifier
             query: User query/message
             apply_id: Application identifier
-            group_id: Group identifier
+            end_user_id: Group identifier
             ai_response: AI response/answer
             
         Returns:
@@ -131,7 +131,7 @@ class SessionService:
                 userid=user_id,
                 messages=query,
                 apply_id=apply_id,
-                group_id=group_id,
+                end_user_id=end_user_id,
                 aimessages=ai_response
             )
             
@@ -152,7 +152,7 @@ class SessionService:
         Duplicates are identified by matching:
         - sessionid
         - user_id (id field)
-        - group_id
+        - end_user_id
         - messages
         - aimessages
         

api/app/core/memory/agent/services/template_service.py

@@ -3,12 +3,22 @@ Template Service for loading and rendering Jinja2 templates.
 
 This service provides centralized template management with caching and error handling.
 """
+
 import os
 from functools import lru_cache
-from typing import Optional
-from jinja2 import Environment, FileSystemLoader, Template, TemplateNotFound
 
-from app.core.logging_config import get_agent_logger, log_prompt_rendering
+from jinja2 import (
+    Environment,
+    FileSystemLoader,
+    Template,
+    TemplateNotFound,
+)
+
+from app.core.logging_config import (
+    get_agent_logger,
+    log_prompt_rendering,
+)
+
 
 
 logger = get_agent_logger(__name__)

api/app/core/memory/agent/utils/__init__.py

@@ -1,7 +0,0 @@
-"""Agent utilities."""
-
-from app.core.memory.agent.utils.multimodal import MultimodalProcessor
-
-__all__ = [
-    "MultimodalProcessor",
-]

api/app/core/memory/agent/utils/get_dialogs.py

@@ -9,62 +9,59 @@ from app.core.memory.models.message_models import DialogData, ConversationContex
 
 async def get_chunked_dialogs(
         chunker_strategy: str = "RecursiveChunker",
-        group_id: str = "group_1",
-        user_id: str = "user1",
-        apply_id: str = "applyid",
-        content: str = "这是用户的输入",
+        end_user_id: str = "group_1",
+        messages: list = None,
         ref_id: str = "wyl_20251027",
         config_id: str = None
 ) -> List[DialogData]:
-    """Generate chunks from all test data entries using the specified chunker strategy.
+    """Generate chunks from structured messages using the specified chunker strategy.
 
     Args:
         chunker_strategy: The chunking strategy to use (default: RecursiveChunker)
-        group_id: Group identifier
-        user_id: User identifier
-        apply_id: Application identifier
-        content: Dialog content
+        end_user_id: Group identifier
+        messages: Structured message list [{"role": "user", "content": "..."}, ...]
         ref_id: Reference identifier
         config_id: Configuration ID for processing
 
     Returns:
-        List of DialogData objects with generated chunks for each test entry
+        List of DialogData objects with generated chunks
     """
-    dialog_data_list = []
-    messages = []
+    from app.core.logging_config import get_agent_logger
+    logger = get_agent_logger(__name__)
 
-    messages.append(ConversationMessage(role="用户", msg=content))
+    if not messages or not isinstance(messages, list) or len(messages) == 0:
+        raise ValueError("messages parameter must be a non-empty list")
 
-    # Create DialogData
-    conversation_context = ConversationContext(msgs=messages)
-    # Create DialogData with group_id based on the entry's id for uniqueness
+    conversation_messages = []
+
+    for idx, msg in enumerate(messages):
+        if not isinstance(msg, dict) or 'role' not in msg or 'content' not in msg:
+            raise ValueError(f"Message {idx} format error: must contain 'role' and 'content' fields")
+
+        role = msg['role']
+        content = msg['content']
+
+        if role not in ['user', 'assistant']:
+            raise ValueError(f"Message {idx} role must be 'user' or 'assistant', got: {role}")
+
+        if content.strip():
+            conversation_messages.append(ConversationMessage(role=role, msg=content.strip()))
+
+    if not conversation_messages:
+        raise ValueError("Message list cannot be empty after filtering")
+
+    conversation_context = ConversationContext(msgs=conversation_messages)
     dialog_data = DialogData(
         context=conversation_context,
         ref_id=ref_id,
-        group_id=group_id,
-        user_id=user_id,
-        apply_id=apply_id,
+        end_user_id=end_user_id,
         config_id=config_id
     )
-    # Create DialogueChunker and process the dialogue
+
     chunker = DialogueChunker(chunker_strategy)
     extracted_chunks = await chunker.process_dialogue(dialog_data)
     dialog_data.chunks = extracted_chunks
 
-    dialog_data_list.append(dialog_data)
+    logger.info(f"DialogData created with {len(extracted_chunks)} chunks")
 
-    # Convert to dict with datetime serialized
-    def serialize_datetime(obj):
-        if isinstance(obj, datetime):
-            return obj.isoformat()
-        raise TypeError(f"Object of type {obj.__class__.__name__} is not JSON serializable")
-
-    combined_output = [dd.model_dump() for dd in dialog_data_list]
-
-    print(dialog_data_list)
-
-    # with open(os.path.join(os.path.dirname(__file__), "chunker_test_output.txt"), "w", encoding="utf-8") as f:
-    #     json.dump(combined_output, f, ensure_ascii=False, indent=4, default=serialize_datetime)
-
-
-    return dialog_data_list
+    return [dialog_data]

api/app/core/memory/agent/utils/llm_client_pool.py

@@ -0,0 +1,56 @@
+
+import asyncio
+from typing import Dict, Optional
+from app.core.memory.utils.llm.llm_utils import get_llm_client_fast
+from app.db import get_db
+from app.core.logging_config import get_agent_logger
+
+logger = get_agent_logger(__name__)
+
+class LLMClientPool:
+    """LLM客户端连接池"""
+    
+    def __init__(self, max_size: int = 5):
+        self.max_size = max_size
+        self.pools: Dict[str, asyncio.Queue] = {}
+        self.active_clients: Dict[str, int] = {}
+        
+    async def get_client(self, llm_model_id: str):
+        """获取LLM客户端"""
+        if llm_model_id not in self.pools:
+            self.pools[llm_model_id] = asyncio.Queue(maxsize=self.max_size)
+            self.active_clients[llm_model_id] = 0
+            
+        pool = self.pools[llm_model_id]
+        
+        try:
+            # 尝试从池中获取客户端
+            client = pool.get_nowait()
+            logger.debug(f"从池中获取LLM客户端: {llm_model_id}")
+            return client
+        except asyncio.QueueEmpty:
+            # 池为空，创建新客户端
+            if self.active_clients[llm_model_id] < self.max_size:
+                db_session = next(get_db())
+                client = get_llm_client_fast(llm_model_id, db_session)
+                self.active_clients[llm_model_id] += 1
+                logger.debug(f"创建新LLM客户端: {llm_model_id}")
+                return client
+            else:
+                # 等待可用客户端
+                logger.debug(f"等待LLM客户端可用: {llm_model_id}")
+                return await pool.get()
+                
+    async def return_client(self, llm_model_id: str, client):
+        """归还LLM客户端到池中"""
+        if llm_model_id in self.pools:
+            try:
+                self.pools[llm_model_id].put_nowait(client)
+                logger.debug(f"归还LLM客户端到池: {llm_model_id}")
+            except asyncio.QueueFull:
+                # 池已满，丢弃客户端
+                self.active_clients[llm_model_id] -= 1
+                logger.debug(f"池已满，丢弃LLM客户端: {llm_model_id}")
+
+# 全局客户端池
+llm_client_pool = LLMClientPool()

api/app/core/memory/agent/utils/llm_tools.py

@@ -1,82 +1,83 @@
-import asyncio
-import json
-import logging
 import os
 from collections import defaultdict
+from pathlib import Path
 from typing import Annotated, TypedDict
 
-from app.core.memory.agent.utils.messages_tool import read_template_file
-from app.core.memory.utils.config.config_utils import (
-    get_picture_config,
-    get_voice_config,
-)
-
-# Removed global variable imports - use dependency injection instead
-from dotenv import load_dotenv
 from langchain_core.messages import AnyMessage
 from langgraph.graph import add_messages
-from openai import OpenAI
 
-PROJECT_ROOT_ = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-logger = logging.getLogger(__name__)
+PROJECT_ROOT_ = str(Path(__file__).resolve().parents[3])
 
-load_dotenv()
-
-
-async def picture_model_requests(image_url):
-    '''
-
-    Args:
-        image_url:
-    Returns:
-
-    '''
-    file_path = PROJECT_ROOT_ + '/agent/utils/prompt/Template_for_image_recognition_prompt.jinja2 '
-    system_prompt = await read_template_file(file_path)
-    result = await Picture_recognize(image_url,system_prompt)
-    return (result)
 class WriteState(TypedDict):
     '''
     Langgrapg Writing TypedDict
     '''
     messages: Annotated[list[AnyMessage], add_messages]
-    user_id:str
-    apply_id:str
-    group_id:str
+    end_user_id: str
     errors: list[dict]  # Track errors: [{"tool": "tool_name", "error": "message"}]
+    memory_config: object
+    write_result: dict
+    data: str
 
 class ReadState(TypedDict):
-    '''
-       Langgrapg READING TypedDict
-       name:
-       id:user id
-       loop_count:Traverse times
-       search_switch：type
-       config_id: configuration id for filtering results
-       errors: list of errors that occurred during workflow execution
-       '''
-    messages: Annotated[list[AnyMessage], add_messages] #消息追加的模式增加消息
-    name: str
-    id: str
-    loop_count:int
+    """
+    LangGraph 工作流状态定义
+
+    Attributes:
+        messages: 消息列表，支持自动追加
+        loop_count: 遍历次数
+        search_switch: 搜索类型开关
+        end_user_id: 组标识
+        config_id: 配置ID，用于过滤结果
+        data: 从content_input_node传递的内容数据
+        spit_data: 从Split_The_Problem传递的分解结果
+        tool_calls: 工具调用请求列表
+        tool_results: 工具执行结果列表
+        memory_config: 内存配置对象
+    """
+    messages: Annotated[list[AnyMessage], add_messages]  # 消息追加模式
+    loop_count: int
     search_switch: str
-    user_id: str
-    apply_id: str
-    group_id: str
+    end_user_id: str
     config_id: str
-    errors: list[dict]  # Track errors: [{"tool": "tool_name", "error": "message"}]
-
-
+    data: str  # 新增字段用于传递内容
+    spit_data: dict  # 新增字段用于传递问题分解结果
+    problem_extension:dict
+    storage_type: str
+    user_rag_memory_id: str
+    llm_id: str
+    embedding_id: str
+    memory_config: object  # 新增字段用于传递内存配置对象
+    retrieve:dict
+    RetrieveSummary: dict
+    InputSummary: dict
+    verify: dict
+    SummaryFails: dict
+    summary: dict
 class COUNTState:
-    '''
-    The number of times the workflow dialogue retrieval content has no correct message recall traversal
-    '''
+    """
+    工作流对话检索内容计数器
+
+    用于记录工作流对话检索内容没有正确消息召回遍历的次数。
+    """
+
     def __init__(self, limit: int = 5):
+        """
+        初始化计数器
+
+        Args:
+            limit: 最大计数限制，默认为5
+        """
         self.total: int = 0  # 当前累加值
         self.limit: int = limit  # 最大上限
 
-    def add(self, value: int = 1):
-        """累加数字，如果达到上限就保持最大值"""
+    def add(self, value: int = 1) -> None:
+        """
+        累加数字，如果达到上限就保持最大值
+
+        Args:
+            value: 要累加的值，默认为1
+        """
         self.total += value
         print(f"[COUNTState] 当前值: {self.total}")
         if self.total >= self.limit:
@@ -84,21 +85,19 @@ class COUNTState:
             self.total = self.limit  # 达到上限不再增加
 
     def get_total(self) -> int:
-        """获取当前累加值"""
+        """
+        获取当前累加值
+
+        Returns:
+            当前累加值
+        """
         return self.total
 
-    def reset(self):
+    def reset(self) -> None:
         """手动重置累加值"""
         self.total = 0
         print("[COUNTState] 已重置为 0")
 
-
-def merge_to_key_value_pairs(data, query_key, result_key):
-    grouped = defaultdict(list)
-    for item in data:
-        grouped[item[query_key]].append(item[result_key])
-    return [{key: values} for key, values in grouped.items()]
-
 def deduplicate_entries(entries):
     seen = set()
     deduped = []
@@ -109,70 +108,37 @@ def deduplicate_entries(entries):
             deduped.append(entry)
     return deduped
 
+def merge_to_key_value_pairs(data, query_key, result_key):
+    grouped = defaultdict(list)
+    for item in data:
+        grouped[item[query_key]].append(item[result_key])
+    return [{key: values} for key, values in grouped.items()]
 
 
-async def Picture_recognize(image_path, PROMPT_TICKET_EXTRACTION, picture_model_name: str) -> str:
+def convert_extended_question_to_question(data):
     """
-    Updated to eliminate global variables in favor of explicit parameters.
-    
+    递归地将数据中的 extended_question 字段转换为 question 字段
+
     Args:
-        image_path: Path to image file
-        PROMPT_TICKET_EXTRACTION: Extraction prompt
-        picture_model_name: Picture model name (required, no longer from global variables)
+        data: 要转换的数据（可能是字典、列表或其他类型）
+
+    Returns:
+        转换后的数据
     """
-    try:
-        model_config = get_picture_config(picture_model_name)
-    except Exception as e:
-            err = f"LLM配置不可用：{str(e)}。请检查 config.json 和 runtime.json。"
-            logger.error(err)
-            return err
-    api_key = os.getenv(model_config["api_key"])  # 从环境变量读取对应后端的 API key
-    backend_model_name = model_config["llm_name"].split("/")[-1]
-    api_base=model_config['api_base']
-
-    logger.info(f"model_name: {backend_model_name}")
-    logger.info(f"api_key set: {'yes' if api_key else 'no'}")
-    logger.info(f"base_url: {model_config['api_base']}")
-
-    client = OpenAI(
-        api_key=api_key, base_url=api_base,
-    )
-    completion = client.chat.completions.create(
-        model=backend_model_name,
-        messages=[
-                {
-                    "role": "user",
-                    "content": [
-                        {
-                            "type": "image_url",
-                            "image_url":image_path,
-                        },
-                        {"type": "text",
-                         "text": PROMPT_TICKET_EXTRACTION}
-                    ]
-                }
-            ])
-    picture_text = completion.choices[0].message.content
-    picture_text = picture_text.replace('```json', '').replace('```', '')
-    picture_text = json.loads(picture_text)
-    return (picture_text['statement'])
-
-async def Voice_recognize(voice_model_name: str):
-    """
-    Updated to eliminate global variables in favor of explicit parameters.
-    
-    Args:
-        voice_model_name: Voice model name (required, no longer from global variables)
-    """
-    try:
-        model_config = get_voice_config(voice_model_name)
-    except Exception as e:
-            err = f"LLM配置不可用：{str(e)}。请检查 config.json 和 runtime.json。"
-            logger.error(err)
-            return err
-    api_key = os.getenv(model_config["api_key"])  # 从环境变量读取对应后端的 API key
-    backend_model_name = model_config["llm_name"].split("/")[-1]
-    api_base = model_config['api_base']
-    return api_key,backend_model_name,api_base
-
-
+    if isinstance(data, dict):
+        # 创建新字典来存储转换后的数据
+        converted = {}
+        for key, value in data.items():
+            if key == 'extended_question':
+                # 将 extended_question 转换为 question
+                converted['question'] = convert_extended_question_to_question(value)
+            else:
+                # 递归处理其他字段
+                converted[key] = convert_extended_question_to_question(value)
+        return converted
+    elif isinstance(data, list):
+        # 递归处理列表中的每个元素
+        return [convert_extended_question_to_question(item) for item in data]
+    else:
+        # 其他类型直接返回
+        return data

api/app/core/memory/agent/utils/mcp_tools.py

@@ -1,33 +0,0 @@
-import os
-from app.core.config import settings
-
-def get_mcp_server_config():
-    """
-    Get the MCP server configuration.
-    
-    Uses MCP_SERVER_URL environment variable if set (for Docker),
-    otherwise falls back to SERVER_IP and MCP_PORT (for local development).
-    """
-    # Get MCP port from environment (default: 8081)
-    mcp_port = os.getenv("MCP_PORT", "8081")
-    
-    # In Docker: MCP_SERVER_URL=http://mcp-server:8081
-    # In local dev: uses SERVER_IP (127.0.0.1 or localhost)
-    mcp_server_url = os.getenv("MCP_SERVER_URL")
-    
-    if mcp_server_url:
-        # Docker environment: use full URL from environment
-        base_url = mcp_server_url
-    else:
-        # Local development: build URL from SERVER_IP and MCP_PORT
-        base_url = f"http://{settings.SERVER_IP}:{mcp_port}"
-    
-    mcp_server_config = {
-        "data_flow": {
-            "url": f"{base_url}/sse",
-            "transport": "sse",
-            "timeout": 15000,
-            "sse_read_timeout": 15000,
-        }
-    }
-    return mcp_server_config

api/app/core/memory/agent/utils/messages_tool.py

@@ -1,260 +0,0 @@
-import json
-import logging
-import re
-from typing import Any, List
-
-from app.core.logging_config import get_agent_logger
-from langchain_core.messages import AnyMessage
-
-logger = get_agent_logger(__name__)
-
-
-def _to_openai_messages(msgs: List[AnyMessage]) -> List[dict]:
-    out = []
-    for m in msgs:
-        if hasattr(m, "content"):
-            out.append({"role": "user", "content": getattr(m, "content", "")})
-        elif isinstance(m, dict) and "role" in m and "content" in m:
-            out.append(m)
-        else:
-            out.append({"role": "user", "content": str(m)})
-    return out
-
-
-def _extract_content(resp: Any) -> str:
-    """Extract LLM content and sanitize to raw JSON/text.
-
-    - Supports both object and dict response shapes.
-    - Removes leading role labels (e.g., "Assistant:").
-    - Strips Markdown code fences like ```json ... ```.
-    - Attempts to isolate the first valid JSON array/object block when extra text is present.
-    """
-
-    def _to_text(r: Any) -> str:
-        try:
-            # 对象形式: resp.choices[0].message.content
-            if hasattr(r, "choices") and getattr(r, "choices", None):
-                msg = r.choices[0].message
-                if hasattr(msg, "content"):
-                    return msg.content
-                if isinstance(msg, dict) and "content" in msg:
-                    return msg["content"]
-            # 字典形式: resp["choices"][0]["message"]["content"]
-            if isinstance(r, dict):
-                return r.get("choices", [{}])[0].get("message", {}).get("content", "")
-        except Exception:
-            pass
-        return str(r)
-
-    def _clean_text(text: str) -> str:
-        s = str(text).strip()
-        # 移除可能的角色前缀
-        s = re.sub(r"^\s*(Assistant|assistant)\s*:\s*", "", s)
-        # 提取 ```json ... ``` 代码块
-        m = re.search(r"```json\s*(.*?)\s*```", s, flags=re.S | re.I)
-        if m:
-            s = m.group(1).strip()
-        # 如果仍然包含多余文本，尝试截取第一个 JSON 数组/对象片段
-        if not (s.startswith("{") or s.startswith("[")):
-            left = s.find("[")
-            right = s.rfind("]")
-            if left != -1 and right != -1 and right > left:
-                s = s[left:right + 1].strip()
-            else:
-                left = s.find("{")
-                right = s.rfind("}")
-                if left != -1 and right != -1 and right > left:
-                    s = s[left:right + 1].strip()
-        return s
-
-    raw = _to_text(resp)
-    return _clean_text(raw)
-
-def Resolve_username(usermessages):
-    '''
-    Extract username
-    Args:
-        usermessages: user name
-
-    Returns:
-
-    '''
-    usermessages = usermessages.split('_')[1:]
-    sessionid = '_'.join(usermessages[:-1])
-    return sessionid
-
-
-# TODO: USE app.core.memory.src.utils.render_template instead
-async def read_template_file(template_path: str) -> str:
-    """
-    读取模板文件
-
-    Args:
-        template_path: 模板文件路径
-
-    Returns:
-        模板内容字符串
-
-    Note:
-        建议使用 app.core.memory.utils.template_render 中的统一模板渲染功能
-    """
-    try:
-        with open(template_path, "r", encoding="utf-8") as f:
-            return f.read()
-    except FileNotFoundError:
-        logger.error(f"模板文件未找到: {template_path}")
-        raise
-    except IOError as e:
-        logger.error(f"读取模板文件失败: {template_path}, 错误: {str(e)}", exc_info=True)
-        raise
-
-
-async def Problem_Extension_messages_deal(context):
-    '''
-    Extract data
-    Args:
-        context:
-    Returns:
-    '''
-    extent_quest = []
-    original = context.get('original', '')
-    messages = context.get('context', '')
-    
-    # Handle empty or non-string messages
-    if not messages:
-        return extent_quest, original
-    
-    if isinstance(messages, str):
-        try:
-            messages = json.loads(messages)
-        except json.JSONDecodeError:
-            # If JSON parsing fails, return empty list
-            return extent_quest, original
-    
-    if isinstance(messages, list):
-        for message in messages:
-            question = message.get('question', '')
-            type = message.get('type', '')
-            extent_quest.append({"role": "user", "content": f"问题:{question}；问题类型：{type}"})
-
-    return extent_quest, original
-
-
-async def Retriev_messages_deal(context):
-    '''
-    Extract data
-    Args:
-        context:
-    Returns:
-    '''
-    logger.info(f"Retriev_messages_deal input: type={type(context)}, value={str(context)[:500]}")
-    
-    if isinstance(context, dict):
-        logger.info(f"Retriev_messages_deal: context is dict with keys={list(context.keys())}")
-        if 'context' in context or 'original' in context:
-            content = context.get('context', {})
-            original = context.get('original', '')
-            logger.info(f"Retriev_messages_deal output: content_type={type(content)}, content={str(content)[:300]}, original='{original[:50] if original else ''}'")
-            return content, original
-    
-    # Return empty defaults if context is not a dict or doesn't have expected keys
-    logger.warning(f"Retriev_messages_deal: context missing expected keys, returning empty defaults")
-    return {}, ''
-
-async  def Verify_messages_deal(context):
-    '''
-    Extract data
-    Args:
-        context:
-    Returns:
-    '''
-
-    query = context['context']['Query']
-    Query_small_list = context['context']['Expansion_issue']
-    Result_small = []
-    Query_small = []
-    for i in Query_small_list:
-        Result_small.append(i['Answer_Small'][0])
-        Query_small.append(i['Query_small'])
-    return Query_small, Result_small, query
-
-
-async def Summary_messages_deal(context):
-    '''
-    Extract data
-    Args:
-        context:
-    Returns:
-    '''
-    messages = str(context).replace('\\n', '').replace('\n', '').replace('\\', '')
-    query = re.findall(r'"query": (.*?),', messages)[0]
-    query = query.replace('[', '').replace(']', '').strip()
-    matches = re.findall(r'"answer_small"\s*:\s*"(\[.*?\])"', messages)
-    answer_small_texts = []
-    for m in matches:
-        try:
-            parsed = json.loads(m)
-            for item in parsed:
-                answer_small_texts.append(item.strip().replace('\\', '').replace('[', '').replace(']', ''))
-        except Exception:
-            answer_small_texts.append(m.strip().replace('\\', '').replace('[', '').replace(']', ''))
-
-    return answer_small_texts, query
-
-
-async def VerifyTool_messages_deal(context):
-    '''
-    Extract data
-    Args:
-        context:
-    Returns:
-    '''
-    messages = str(context).replace('\\n', '').replace('\n', '').replace('\\', '')
-    content_messages = messages.split('"context":')[1].replace('""', '"')
-    messages = str(content_messages).split("name='Retrieve'")[0]
-    query = re.findall('"Query": "(.*?)"', messages)[0]
-    Query_small = re.findall('"Query_small": "(.*?)"', messages)
-    Result_small = re.findall('"Result_small": "(.*?)"', messages)
-    return Query_small, Result_small, query
-
-
-async def Retrieve_Summary_messages_deal(context):
-    pass
-
-
-async def Retrieve_verify_tool_messages_deal(context, history, query):
-    '''
-    Extract data
-    Args:
-        context:
-    Returns:
-    '''
-    results = []
-    # 统一转为字符串，避免 None 或非字符串导致正则报错
-    text = str(context)
-    blocks = re.findall(r'\{(.*?)\}', text, flags=re.S)
-    for block in blocks:
-        query_small = re.search(r'"Query_small"\s*:\s*"([^"]*)"', block)
-        answer_small = re.search(r'"Answer_Small"\s*:\s*(\[[^\]]*\])', block)
-        status = re.search(r'"status"\s*:\s*"([^"]*)"', block)
-        query_answer = re.search(r'"Query_answer"\s*:\s*"([^"]*)"', block)
-
-        results.append({
-            "query_small": query_small.group(1) if query_small else None,
-            "answer_small": answer_small.group(1) if answer_small else None,
-            # 将缺失的 status 统一为空字符串，后续用字符串判定，避免 NoneType 错误
-            "status": status.group(1) if status else "",
-            "query_answer": query_answer.group(1) if query_answer else None
-        })
-    result = []
-    for r in results:
-        # 统一按字符串判定状态，兼容大小写和缺失情况
-        status_str = str(r.get('status', '')).strip().lower()
-        if status_str == 'false':
-            continue
-        else:
-            result.append(r)
-    split_result = 'failed' if not result else 'success'
-    result = {"data": {"query": query, "expansion_issue": result}, "split_result": split_result, "reason": "",
-              "history": history}
-    return result

api/app/core/memory/agent/utils/messages_tools.py

@@ -0,0 +1,194 @@
+from typing import List, Dict, Any
+from app.core.logging_config import get_agent_logger
+
+logger = get_agent_logger(__name__)
+async def read_template_file(template_path: str) -> str:
+    """
+    读取模板文件
+
+    Args:
+        template_path: 模板文件路径
+
+    Returns:
+        模板内容字符串
+
+    Note:
+        建议使用 app.core.memory.utils.template_render 中的统一模板渲染功能
+    """
+    try:
+        with open(template_path, "r", encoding="utf-8") as f:
+            return f.read()
+    except FileNotFoundError:
+        logger.error(f"模板文件未找到: {template_path}")
+        raise
+    except IOError as e:
+        logger.error(f"读取模板文件失败: {template_path}, 错误: {str(e)}", exc_info=True)
+        raise
+
+def reorder_output_results(results: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+    """
+    重新排序输出结果，将 retrieval_summary 类型的数据放到最后面
+
+    Args:
+        results: 原始输出结果列表
+
+    Returns:
+        重新排序后的结果列表
+    """
+    retrieval_summaries = []
+    other_results = []
+
+    # 分离 retrieval_summary 和其他类型的结果
+    for result in results:
+        if 'summary' in result.get('type'):
+            retrieval_summaries.append(result)
+        else:
+            other_results.append(result)
+
+    # 将 retrieval_summary 放到最后
+    return other_results + retrieval_summaries
+
+def optimize_search_results(intermediate_outputs):
+    """
+    优化检索结果，合并多个搜索结果，过滤空结果，统一格式
+
+    Args:
+        intermediate_outputs: 原始的中间输出列表
+
+    Returns:
+        优化后的检索结果列表
+    """
+    optimized_results = []
+
+    for item in intermediate_outputs:
+        if not item or item == [] or item == {}:
+            continue
+
+        # 检查是否是搜索结果类型
+        if isinstance(item, dict) and item.get('type') == 'search_result':
+            raw_results = item.get('raw_results', {})
+
+            # 如果 raw_results 为空，跳过
+            if not raw_results or raw_results == [] or raw_results == {}:
+                continue
+
+            # 创建优化后的结果结构
+            optimized_item = {
+                "type": "search_result",
+                "title": f"检索结果 ({item.get('index', 1)}/{item.get('total', 1)})",
+                "query": item.get('query', ''),
+                "raw_results": {},
+                "index": item.get('index', 1),
+                "total": item.get('total', 1)
+            }
+
+            # 合并所有搜索结果类型到一个 raw_results 中
+            merged_raw_results = {}
+
+            # 处理 time_search
+            if 'time_search' in raw_results and raw_results['time_search']:
+                merged_raw_results['time_search'] = raw_results['time_search']
+
+            # 处理 keyword_search
+            if 'keyword_search' in raw_results and raw_results['keyword_search']:
+                merged_raw_results['keyword_search'] = raw_results['keyword_search']
+
+            # 处理 embedding_search
+            if 'embedding_search' in raw_results and raw_results['embedding_search']:
+                merged_raw_results['embedding_search'] = raw_results['embedding_search']
+
+            # 处理 combined_summary
+            if 'combined_summary' in raw_results and raw_results['combined_summary']:
+                merged_raw_results['combined_summary'] = raw_results['combined_summary']
+
+            # 处理 reranked_results
+            if 'reranked_results' in raw_results and raw_results['reranked_results']:
+                merged_raw_results['reranked_results'] = raw_results['reranked_results']
+
+            # 如果合并后的结果不为空，添加到优化结果中
+            if merged_raw_results:
+                optimized_item['raw_results'] = merged_raw_results
+                optimized_results.append(optimized_item)
+        else:
+            # 非搜索结果类型，直接添加
+            optimized_results.append(item)
+
+    return optimized_results
+
+
+def merge_multiple_search_results(intermediate_outputs):
+    """
+    将多个搜索结果合并为一个统一的搜索结果
+
+    Args:
+        intermediate_outputs: 原始的中间输出列表
+
+    Returns:
+        合并后的结果列表
+    """
+    search_results = []
+    other_results = []
+
+    # 分离搜索结果和其他结果
+    for item in intermediate_outputs:
+        if isinstance(item, dict) and item.get('type') == 'search_result':
+            raw_results = item.get('raw_results', {})
+            # 只保留有内容的搜索结果
+            if raw_results and raw_results != [] and raw_results != {}:
+                search_results.append(item)
+        else:
+            other_results.append(item)
+
+    # 如果没有搜索结果，返回原始结果
+    if not search_results:
+        return intermediate_outputs
+
+    # 如果只有一个搜索结果，优化格式后返回
+    if len(search_results) == 1:
+        optimized = optimize_search_results(search_results)
+        return other_results + optimized
+
+    # 合并多个搜索结果
+    merged_raw_results = {}
+    all_queries = []
+
+    for result in search_results:
+        query = result.get('query', '')
+        if query:
+            all_queries.append(query)
+
+        raw_results = result.get('raw_results', {})
+
+        # 合并各种搜索类型的结果
+        for search_type in ['time_search', 'keyword_search', 'embedding_search', 'combined_summary',
+                            'reranked_results']:
+            if search_type in raw_results and raw_results[search_type]:
+                if search_type not in merged_raw_results:
+                    merged_raw_results[search_type] = raw_results[search_type]
+                else:
+                    # 如果是字典类型，需要合并
+                    if isinstance(raw_results[search_type], dict) and isinstance(merged_raw_results[search_type], dict):
+                        for key, value in raw_results[search_type].items():
+                            if key not in merged_raw_results[search_type]:
+                                merged_raw_results[search_type][key] = value
+                            elif isinstance(value, list) and isinstance(merged_raw_results[search_type][key], list):
+                                merged_raw_results[search_type][key].extend(value)
+                    elif isinstance(raw_results[search_type], list):
+                        if isinstance(merged_raw_results[search_type], list):
+                            merged_raw_results[search_type].extend(raw_results[search_type])
+                        else:
+                            merged_raw_results[search_type] = raw_results[search_type]
+
+    # 创建合并后的结果
+    if merged_raw_results:
+        merged_result = {
+            "type": "search_result",
+            "title": f"合并检索结果 (共{len(search_results)}个查询)",
+            "query": " | ".join(all_queries),
+            "raw_results": merged_raw_results,
+            "index": 1,
+            "total": 1
+        }
+        return other_results + [merged_result]
+
+    return other_results

api/app/core/memory/agent/utils/model_tool.py

@@ -1,38 +0,0 @@
-
-
-# project_root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-# sys.path.insert(0, project_root)
-
-# load_dotenv()
-
-# async def llm_client_chat(messages: List[dict]) -> str:
-#     """使用 OpenAI 兼容接口进行对话，返回内容字符串。"""
-#     try:
-#         cfg = get_model_config(SELECTED_LLM_ID)
-#         rb_config = RedBearModelConfig(
-#         model_name=cfg["model_name"],
-#         provider=cfg["provider"],
-#         api_key=cfg["api_key"],
-#         base_url=cfg["base_url"],
-#     )
-#         client = OpenAIClient(model_config=rb_config, type_="chat")
-
-#     except Exception as e:
-#         logger.error(f"获取模型配置失败：{e}")
-#         err = f"获取模型配置失败：{str(e)}。请检查!!!"
-#         return err
-#     try:
-#         response = await client.chat(messages)
-#         print(f"model_tool's llm_client_chat response ======>:\n {response}")
-#         return _extract_content(response)
-#         # return _extract_content(result)
-#     except Exception as e:
-#         logger.error(f"LLM调用失败：{str(e)}。请检查 model_name、api_key、api_base 是否正确。")
-#     return f"LLM调用失败：{str(e)}。请检查 model_name、api_key、api_base 是否正确。"
-
-# async def main(image_url):
-#     await llm_client_chat(image_url)
-#
-# # 运行主函数
-# asyncio.run(main(['https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_male2.wav']))
-#

api/app/core/memory/agent/utils/multimodal.py

@@ -1,131 +0,0 @@
-"""
-Multimodal input processor for handling image and audio content.
-
-This module provides utilities for detecting and processing multimodal inputs
-(images and audio files) by converting them to text using appropriate models.
-"""
-
-import logging
-from typing import List
-
-from app.core.memory.agent.multimodal.speech_model import Vico_recognition
-from app.core.memory.agent.utils.llm_tools import picture_model_requests
-
-logger = logging.getLogger(__name__)
-
-
-class MultimodalProcessor:
-    """
-    Processor for handling multimodal inputs (images and audio).
-    
-    This class detects image and audio file paths in input content and converts
-    them to text using appropriate recognition models.
-    """
-    
-    # Supported file extensions
-    IMAGE_EXTENSIONS = ['.jpg', '.png']
-    AUDIO_EXTENSIONS = [
-        'aac', 'amr', 'avi', 'flac', 'flv', 'm4a', 'mkv', 'mov',
-        'mp3', 'mp4', 'mpeg', 'ogg', 'opus', 'wav', 'webm', 'wma', 'wmv'
-    ]
-    
-    def __init__(self):
-        """Initialize the multimodal processor."""
-        pass
-    
-    def is_image(self, content: str) -> bool:
-        """
-        Check if content is an image file path.
-        
-        Args:
-            content: Input string to check
-            
-        Returns:
-            True if content ends with a supported image extension
-            
-        Examples:
-            >>> processor = MultimodalProcessor()
-            >>> processor.is_image("photo.jpg")
-            True
-            >>> processor.is_image("document.pdf")
-            False
-        """
-        if not isinstance(content, str):
-            return False
-        
-        content_lower = content.lower()
-        return any(content_lower.endswith(ext) for ext in self.IMAGE_EXTENSIONS)
-    
-    def is_audio(self, content: str) -> bool:
-        """
-        Check if content is an audio file path.
-        
-        Args:
-            content: Input string to check
-            
-        Returns:
-            True if content ends with a supported audio extension
-            
-        Examples:
-            >>> processor = MultimodalProcessor()
-            >>> processor.is_audio("recording.mp3")
-            True
-            >>> processor.is_audio("video.mp4")
-            True
-            >>> processor.is_audio("document.txt")
-            False
-        """
-        if not isinstance(content, str):
-            return False
-        
-        content_lower = content.lower()
-        return any(content_lower.endswith(f'.{ext}') for ext in self.AUDIO_EXTENSIONS)
-    
-    async def process_input(self, content: str) -> str:
-        """
-        Process input content, converting images/audio to text if needed.
-        
-        This method detects if the input is an image or audio file and converts
-        it to text using the appropriate recognition model. If processing fails
-        or the content is not multimodal, it returns the original content.
-        
-        Args:
-            content: Input string (may be file path or regular text)
-            
-        Returns:
-            Text content (original or converted from image/audio)
-            
-        Examples:
-            >>> processor = MultimodalProcessor()
-            >>> await processor.process_input("photo.jpg")
-            "Recognized text from image..."
-            
-            >>> await processor.process_input("Hello world")
-            "Hello world"
-        """
-        if not isinstance(content, str):
-            logger.warning(f"[MultimodalProcessor] Content is not a string: {type(content)}")
-            return str(content)
-        
-        try:
-            # Check for image input
-            if self.is_image(content):
-                logger.info(f"[MultimodalProcessor] Detected image input: {content}")
-                result = await picture_model_requests(content)
-                logger.info(f"[MultimodalProcessor] Image recognition result: {result[:100]}...")
-                return result
-            
-            # Check for audio input
-            if self.is_audio(content):
-                logger.info(f"[MultimodalProcessor] Detected audio input: {content}")
-                result = await Vico_recognition([content]).run()
-                logger.info(f"[MultimodalProcessor] Audio recognition result: {result[:100]}...")
-                return result
-        
-        except Exception as e:
-            logger.error(f"[MultimodalProcessor] Error processing multimodal input: {e}", exc_info=True)
-            logger.info("[MultimodalProcessor] Falling back to original content")
-            return content
-        
-        # Return original content if not multimodal
-        return content

api/app/core/memory/agent/utils/performance_monitor.py

@@ -0,0 +1,56 @@
+
+import time
+import json
+from collections import defaultdict
+from typing import Dict, List
+from app.core.logging_config import get_agent_logger
+
+logger = get_agent_logger(__name__)
+
+class ProblemExtensionMonitor:
+    """Problem_Extension性能监控器"""
+    
+    def __init__(self):
+        self.metrics = defaultdict(list)
+        self.slow_queries = []
+        self.error_count = 0
+        
+    def record_execution(self, duration: float, question_count: int, success: bool):
+        """记录执行指标"""
+        self.metrics['durations'].append(duration)
+        self.metrics['question_counts'].append(question_count)
+        
+        if not success:
+            self.error_count += 1
+            
+        # 记录慢查询（超过10秒）
+        if duration > 10.0:
+            self.slow_queries.append({
+                'duration': duration,
+                'question_count': question_count,
+                'timestamp': time.time()
+            })
+            
+    def get_stats(self) -> Dict:
+        """获取统计信息"""
+        durations = self.metrics['durations']
+        if not durations:
+            return {"message": "暂无数据"}
+            
+        return {
+            "total_executions": len(durations),
+            "avg_duration": sum(durations) / len(durations),
+            "max_duration": max(durations),
+            "min_duration": min(durations),
+            "slow_queries_count": len(self.slow_queries),
+            "error_rate": self.error_count / len(durations) if durations else 0,
+            "recent_slow_queries": self.slow_queries[-5:]  # 最近5个慢查询
+        }
+        
+    def log_stats(self):
+        """记录统计信息到日志"""
+        stats = self.get_stats()
+        logger.info(f"Problem_Extension性能统计: {json.dumps(stats, indent=2)}")
+
+# 全局监控器实例
+performance_monitor = ProblemExtensionMonitor()

api/app/core/memory/agent/utils/prompt/Problem_Extension_prompt_simplified.jinja2

@@ -0,0 +1,81 @@
+
+你是一个高效的问题拆分助手，任务是根据用户提供的原始问题和问题类型，生成可操作的扩展问题，用于精确回答原问题。请严格遵循以下规则：
+
+角色：
+- 你是“问题拆分专家”，专注于逻辑、信息完整性和可操作性。
+- 你能够结合【历史信息】、【上下文】、【背景知识】进行分析，以保持问题拆分的连贯性和相关性。
+- 如果历史信息或上下文与当前问题无关，可忽略。
+
+---
+
+### 历史信息参考
+在生成扩展问题时，你可以参考以下历史数据（如果提供）：
+- 历史对话或任务的主题；
+- 历史中出现的关键实体（时间、人物、地点、研究主题等）；
+- 历史中已解答的问题（避免重复）；
+- 历史推理链（保持逻辑一致性）。
+
+> 如果没有提供历史信息，则仅根据当前输入问题进行分析。
+输入历史信息内容:{{history}}
+
+## User Input
+{% if questions is string %}
+{{ questions }}
+{% else %}
+{% for question in questions %}
+- {{ question }}
+{% endfor %}
+{% endif %}
+
+需求：
+- 如果问题是单跳问题（单步可答），直接保留原问题提取重要提问部分作为拆分/扩展问题。
+- 如果问题是多跳问题（需多个信息点才能回答），对问题进行扩展拆分。
+- 扩展问题必须完整覆盖原问题的所有关键要素，包括时间、主体、动作、目标等，不得遗漏。
+- 扩展问题不得冗余：避免重复询问相同信息或过度拆分同一主题。
+- 扩展问题必须高度相关：每个子问题直接服务于原问题，不引入未提及的新概念、人物或细节。
+- 扩展问题必须可操作：每个子问题能在有限资源下独立解答。
+- 子问题数量不超过4个。
+- 拆分问题的时候可以考虑输入的历史内容，以保持逻辑连贯。
+    比如：输入历史信息内容:[{'Query': '4月27日，我和你推荐过一本书，书名是什么？', 'ANswer': '张曼玉推荐了《小王子》'}]
+    拆分问题：4月27日，我和你推荐过一本书，书名是什么？，可以拆分为：4月27日，张曼玉推荐过一本书，书名是什么？
+
+
+
+输出要求：
+- 仅输出 JSON 数组，不要包含任何解释或代码块。
+- 每个元素包含：
+  - `original_question`: 原始问题
+  - `extended_question`: 扩展后的问题
+  - `type`: 类型（事实检索/澄清/定义/比较/行动建议）
+  - `reason`: 生成该扩展问题的简短理由
+- 使用标准 ASCII 双引号，无换行；确保字符串正确关闭并以逗号分隔。
+
+示例：
+输入：
+[
+  "问题：今年诺贝尔物理学奖的获奖者是谁，他们因为什么贡献获奖？；问题类型：多跳",
+]
+
+输出：
+[
+  {
+    "original_question": "今年诺贝尔物理学奖的获奖者是谁，他们因为什么贡献获奖？",
+    "extended_question": "今年诺贝尔物理学奖的获奖者有哪些人？",
+    "type": "多跳",
+    "reason": "输出原问题的关键要素"
+  },
+  {
+    "original_question": "今年诺贝尔物理学奖的获奖者是谁，他们因为什么贡献获奖？",
+    "extended_question": "今年诺贝尔物理学奖的获奖者是因哪些具体贡献获奖的？",
+    "type": "多跳",
+    "reason": "输出原问题的关键要素"
+  }
+]
+**Output format**
+**CRITICAL JSON FORMATTING REQUIREMENTS:**
+1. Use only standard ASCII double quotes (") for JSON structure - never use Chinese quotation marks ("") or other Unicode quotes
+2. If the extracted statement text contains quotation marks, escape them properly using backslashes (\")
+3. Ensure all JSON strings are properly closed and comma-separated
+4. Do not include line breaks within JSON string values
+
+The output language should always be the same as the input language.{{ json_schema }}

api/app/core/memory/agent/utils/prompt/Retrieve_Summary_prompt.jinja2

@@ -1,13 +1,10 @@
 # 角色
 你是一个专业的问答助手，擅长基于检索信息和历史对话回答用户问题。
-
 # 任务
 根据提供的上下文信息回答用户的问题。
-
 # 输入信息
 - 历史对话：{{history}}
 - 检索信息：{{retrieve_info}}
-
 ## User Query
 {{query}}
 

api/app/core/memory/agent/utils/prompt/direct_summary_prompt.jinja2

@@ -0,0 +1,61 @@
+# 角色
+你是一个智能问答助手，基于检索信息和历史对话回答用户问题。
+# 任务
+根据提供的上下文信息回答用户的问题。
+# 输入信息
+- 历史对话：{{history}}
+- 检索信息：{{retrieve_info}}
+# 用户问题
+{{query}}
+# 回答指南
+## 1. 仔细阅读检索信息
+- 答案可能直接或间接地出现在检索信息中
+- 如果检索信息中提到"小曼会使用Python"，说明用户名是"小曼"
+- 第三人称描述的偏好、行为通常指用户本人
+
+## 2. 判断信息相关性
+**情况A：信息匹配问题**
+- 直接回答，像自然对话一样
+- 例：检索到"小曼会使用Python" → 问"我叫什么" → 答"你叫小曼"
+
+**情况B：信息部分相关**
+- 先回答已知部分，再自然地询问更多信息
+- 例：检索到"用户去过上海的面包店" → 问"我吃过哪家面包" → 答"我记得你去过上海的面包店，但具体是哪家我不太清楚，是哪家呢？"
+
+**情况C：信息完全不相关**
+- 自然地表达不知道，但可以提及检索到的相关信息，让对话更连贯
+- 使用友好的表达：
+  - "你好像没和我说过...，但是我知道你[检索到的相关信息]"
+  - "关于这个我不太清楚，不过我记得你[检索到的相关信息]，能告诉我更多吗？"
+  - "我不记得你提到过...，但你[检索到的相关信息]"
+- 即使检索信息不直接回答问题，也可以自然地融入对话中
+- 避免僵硬的"信息不足，无法回答"
+## 3. 回答要求
+- 像人类对话一样自然流畅
+- 不要提及"检索信息"、"搜索结果"、"根据资料"等技术术语
+- 不要解释推理过程或引用信息来源
+- 保持友好、乐于助人的语气
+- 使用与问题相同的语言回答
+# 关键示例
+**示例1 - 直接匹配：**
+- 检索信息："小曼会使用Python..."
+- 问题："我叫什么"
+- ✓ 正确："你叫小曼"
+- ✗ 错误："你没有告诉我你的名字"
+**示例2 - 间接匹配：**
+- 检索信息："用户很喜欢吃星巴克的甜品"
+- 问题："我喜欢什么"
+- ✓ 正确："你很喜欢吃星巴克的甜品"
+- ✗ 错误："信息不足"
+**示例3 - 信息不匹配（推荐做法）：**
+- 检索信息："用户只喝拿铁咖啡，认为美式咖啡太苦"
+- 问题："我吃过哪家面包"
+- ✓ 最佳："你好像没和我说过吃过哪家面包，但是我知道你喜欢喝拿铁，能跟我分享一下吗？"
+- ✓ 可以："你好像没和我说过吃过哪家面包，能跟我分享一下吗？"
+- ✗ 错误："用户只喝拿铁咖啡，认为美式咖啡太苦。"（答非所问）
+- ✗ 错误："信息不足，无法回答。"（太僵硬）
+# 重要提醒
+- 检索信息中描述用户行为/偏好时提到的名字，就是用户的名字
+- 信息不匹配时，不要强行回答无关内容，但可以自然地提及检索到的信息，让对话更有温度
+- 用对话式语言表达"不知道"，而非机械模板
+- 检索信息代表你对用户的了解，即使不直接回答问题，也能体现你对用户的记忆

api/app/core/memory/agent/utils/prompt/fail_summary_prompt.jinja2

@@ -0,0 +1,43 @@
+{# 角色定义 #}
+你是专业的问题解答专家+引导学者
+
+{# 输入数据展示 #}
+{% if data %}
+## 输入数据
+上下文信息:
+{% for item in data.history %}
+- {{ item }}
+{% endfor %}
+检索到的所有信息:
+{% for item in data.retrieve_info %}
+- {{ item }}
+{% endfor %}
+{% endif %}
+
+## User Query
+{{ query }}
+
+{# 问题回答标准 #}
+## 问题回答核心标准
+根据上下文信息(history)和检索到的所有信息(retrieve_info)准确回答用户的问题(query)。
+注意，仔细阅读检索信息，答案可能直接或间接地出现在检索信息中或者历史上下文消息中，同时需要 判断信息相关性
+**情况A：信息匹配问题**
+- 直接回答，像自然对话一样
+- 例：检索到"小曼会使用Python" → 问"我叫什么" → 答"你叫小曼"
+
+**情况B：信息部分相关**
+- 先回答已知部分，再自然地询问更多信息
+- 例：检索到"用户去过上海的面包店" → 问"我吃过哪家面包" → 答"我记得你去过上海的面包店，但具体是哪家我不太清楚，是哪家呢？"
+
+**情况C：信息完全不相关**
+- 自然地表达不知道，但可以提及检索到的相关信息，让对话更连贯
+- 使用友好的表达：
+  - "你好像没和我说过...，但是我知道你[检索到的相关信息]"
+  - "关于这个我不太清楚，不过我记得你[检索到的相关信息]，能告诉我更多吗？"
+  - "我不记得你提到过...，但你[检索到的相关信息]"
+- 即使检索信息不直接回答问题，也可以自然地融入对话中
+- 避免僵硬的"信息不足，无法回答"
+
+{# 重要提醒 #}
+当检索以及上下文的历史信息都无法回答的时候，可引导对方进行提问/回答，或者进行其他引导
+当检索或者上下文中出现了，相似的问题，可以委婉，提醒对方，我记得刚刚提过这个问题，但是我自己不记得了，能在描述一次吗～以此为例

api/app/core/memory/agent/utils/prompt/split_verify_prompt.jinja2

@@ -9,8 +9,8 @@
 3. 判断Answer_Small和Query_Small之间分析出来的关系状态
 4. 如果是True保留，否则不要相对应的问题和回答
 5. 输出，需要严格按照模版
-输入：{{history}}
-历史消息：{"history":{{sentence}}}
+输入：{{sentence}}
+历史消息：{"history":{{history}}}
 ### 第一步 获取用户的输入
 获取用户的输入提取对应的Query_Small和Answer_Small
 ### 第二步 分析验证
@@ -42,19 +42,33 @@
 如果状态是TRUE保留这条数据，否则需不需要这条数据
 ### 第五步 输出格式
 按照json的形式输出
-{"data":"Query":原来Query的字段，"history":原来的history字段，
-"expansion_issue":以为列表的形式存储验证之后的数据比如[
-{"query_small": query_small,
-                "answer_small": answer_small,,
-                "status": 回答的结果是否符合query_small，填写状态,
-                "query_answer": answer_small},
+{"query":"原来Query的字段",
+"history":"原来的history字段",
+"expansion_issue":以列表的形式存储验证之后的数据比如[
 {
-                "query_small": "张曼婷生日是什么时候？",
-                "answer_small": "张曼婷喜欢绘画。",
-                "status": "True",
-                "query_answer": "张曼 婷喜欢绘画。"
-            },{}......]
-,
- "split_result":如果expansion_issue是空的列表返回failed，不是空列表返回success,
-  "reason": 为以上分析完之后的结果给一个说明
-            }
+    "query_small": "子问题",
+    "answer_small": "子问题的回答",
+    "status": "True或False，表示回答是否符合query_small",
+    "query_answer": "问题的答案（与answer_small相同）"
+},
+{
+    "query_small": "张曼婷生日是什么时候？",
+    "answer_small": "张曼婷喜欢绘画。",
+    "status": "False",
+    "query_answer": "张曼婷喜欢绘画。"
+}
+],
+"split_result":"如果expansion_issue是空的列表返回failed，不是空列表返回success",
+"reason": "为以上分析完之后的结果给一个说明"
+}
+
+**输出格式要求**
+**CRITICAL JSON FORMATTING REQUIREMENTS:**
+1. Use only standard ASCII double quotes (") for JSON structure - never use Chinese quotation marks ("") or other Unicode quotes
+2. If the extracted statement text contains quotation marks, escape them properly using backslashes (\")
+3. Ensure all JSON strings are properly closed and comma-separated
+4. Do not include line breaks within JSON string values
+5. The output language should always be the same as the input language
+
+**JSON Schema:**
+{{ json_schema }}

api/app/core/memory/agent/utils/prompt/write_aggregate_judgment.jinja2

@@ -0,0 +1,57 @@
+输入句子：{{sentence}}
+历史消息：{{history}}
+
+# 你的角色
+你是一个擅长事件聚合与语义判断的专家。
+
+# 你的任务
+结合历史消息和输入句子，判断它们是否在描述**同一件事件或同一事件链**。
+
+以下情况视为"同一事件"（需要返回 is_same_event=True, output=False）：
+- 描述的是同一个具体事件或事实
+- 存在明显的因果关系、前后发展关系
+- 是对同一事件的补充、解释、追问或延展
+- 逻辑上属于同一语境下的连续讨论
+
+以下情况视为"不同事件"（需要返回 is_same_event=False, output=消息列表）：
+- 话题不同，事件主体不同
+- 时间、地点、对象明显不同
+- 只是语义相似，但并非同一具体事件
+- 无直接事件、因果或逻辑关联
+
+# 输出规则（非常重要）
+你必须按照以下JSON格式输出：
+
+**如果是同一事件：**
+```json
+{
+  "is_same_event": true,
+  "output": false
+}
+```
+
+**如果不是同一事件：**
+```json
+{
+  "is_same_event": false,
+  "output": [
+    {
+      "role": "user",
+      "content": "输入句子的内容"
+    },
+    {
+      "role": "assistant",
+      "content": "对应的回复内容"
+    }
+  ]
+}
+```
+
+# JSON Schema
+{{json_schema}}
+
+# 注意事项
+- 必须严格按照上述格式输出
+- output 字段：如果是同一事件返回 false，如果不是同一事件返回完整的消息列表
+- 消息列表必须包含 role 和 content 字段
+- 不要输出任何解释、分析或多余内容

api/app/core/memory/agent/utils/redis_base.py

@@ -0,0 +1,186 @@
+import json
+from typing import Any, List, Dict, Optional
+from datetime import datetime, timedelta
+
+
+def serialize_messages(messages: Any) -> str:
+    """
+    将消息序列化为 JSON 字符串，支持 LangChain 消息对象
+
+    Args:
+        messages: 可以是 list、dict、string 或 LangChain 消息对象列表
+
+    Returns:
+        str: JSON 字符串
+    """
+    if isinstance(messages, str):
+        return messages
+
+    if isinstance(messages, (list, tuple)):
+        # 检查是否是 LangChain 消息对象列表
+        serialized_list = []
+        for msg in messages:
+            if hasattr(msg, 'type') and hasattr(msg, 'content'):
+                # LangChain 消息对象
+                serialized_list.append({
+                    'type': msg.type,
+                    'content': msg.content,
+                    'role': getattr(msg, 'role', msg.type)
+                })
+            elif isinstance(msg, dict):
+                serialized_list.append(msg)
+            else:
+                serialized_list.append(str(msg))
+        return json.dumps(serialized_list, ensure_ascii=False)
+
+    if isinstance(messages, dict):
+        return json.dumps(messages, ensure_ascii=False)
+
+    # 其他类型转为字符串
+    return str(messages)
+
+
+def deserialize_messages(messages_str: str) -> Any:
+    """
+    将 JSON 字符串反序列化为原始格式
+
+    Args:
+        messages_str: JSON 字符串
+
+    Returns:
+        反序列化后的对象（list、dict 或 string）
+    """
+    if not messages_str:
+        return []
+
+    try:
+        return json.loads(messages_str)
+    except (json.JSONDecodeError, TypeError):
+        return messages_str
+
+
+def fix_encoding(text: str) -> str:
+    """
+    修复错误编码的文本
+    
+    Args:
+        text: 需要修复的文本
+        
+    Returns:
+        str: 修复后的文本
+    """
+    if not text or not isinstance(text, str):
+        return text
+    try:
+        # 尝试修复 Latin-1 误编码为 UTF-8 的情况
+        return text.encode('latin-1').decode('utf-8')
+    except (UnicodeDecodeError, UnicodeEncodeError):
+        # 如果修复失败，返回原文本
+        return text
+
+
+def format_session_data(data: Dict[str, Any], include_time: bool = False) -> Dict[str, Any]:
+    """
+    格式化会话数据为统一的输出格式
+    
+    Args:
+        data: 原始会话数据
+        include_time: 是否包含时间字段
+        
+    Returns:
+        Dict: 格式化后的数据 {"Query": "...", "Answer": "...", "starttime": "..."}
+    """
+    result = {
+        "Query": fix_encoding(data.get('messages', '')),
+        "Answer": fix_encoding(data.get('aimessages', ''))
+    }
+    
+    if include_time:
+        result["starttime"] = data.get('starttime', '')
+    
+    return result
+
+
+def filter_by_time_range(items: List[Dict], minutes: int) -> List[Dict]:
+    """
+    根据时间范围过滤数据
+    
+    Args:
+        items: 包含 starttime 字段的数据列表
+        minutes: 时间范围（分钟）
+        
+    Returns:
+        List[Dict]: 过滤后的数据列表
+    """
+    time_threshold = datetime.now() - timedelta(minutes=minutes)
+    time_threshold_str = time_threshold.strftime("%Y-%m-%d %H:%M:%S")
+    
+    filtered_items = []
+    for item in items:
+        starttime = item.get('starttime', '')
+        if starttime and starttime >= time_threshold_str:
+            filtered_items.append(item)
+    
+    return filtered_items
+
+
+def sort_and_limit_results(items: List[Dict], limit: int = 6, 
+                           remove_time: bool = True) -> List[Dict]:
+    """
+    对结果进行排序、限制数量并移除时间字段
+    
+    Args:
+        items: 数据列表
+        limit: 最大返回数量
+        remove_time: 是否移除 starttime 字段
+        
+    Returns:
+        List[Dict]: 处理后的数据列表
+    """
+    # 按时间降序排序（最新的在前）
+    items.sort(key=lambda x: x.get('starttime', ''), reverse=True)
+    
+    # 限制数量
+    result_items = items[:limit]
+    
+    # 移除 starttime 字段
+    if remove_time:
+        for item in result_items:
+            item.pop('starttime', None)
+    
+    # 如果结果少于1条，返回空列表
+    if len(result_items) < 1:
+        return []
+    
+    return result_items
+
+
+def generate_session_key(session_id: str, key_type: str = "session") -> str:
+    """
+    生成 Redis key
+    
+    Args:
+        session_id: 会话ID
+        key_type: key 类型 ("session", "read", "write", "count")
+        
+    Returns:
+        str: Redis key
+    """
+    if key_type == "count":
+        return f"session:count:{session_id}"
+    elif key_type == "write":
+        return f"session:write:{session_id}"
+    elif key_type == "session" or key_type == "read":
+        return f"session:{session_id}"
+    else:
+        return f"session:{session_id}"
+
+
+def get_current_timestamp() -> str:
+    """
+    获取当前时间戳字符串
+    
+    Returns:
+        str: 格式化的时间字符串 "YYYY-MM-DD HH:MM:SS"
+    """
+    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")

api/app/core/memory/agent/utils/redis_tool.py

@@ -1,11 +1,36 @@
 import redis
 import uuid
-from datetime import datetime
 from app.core.config import settings
+from typing import List, Dict, Any, Optional, Union
+
+from app.core.memory.agent.utils.redis_base import (
+    serialize_messages, 
+    deserialize_messages,
+    fix_encoding,
+    format_session_data,
+    filter_by_time_range,
+    sort_and_limit_results,
+    generate_session_key,
+    get_current_timestamp
+)
 
 
-class RedisSessionStore:
+
+
+class RedisWriteStore:
+    """Redis Write 类型存储类，用于管理 save_session_write 相关的数据"""
+    
     def __init__(self, host='localhost', port=6379, db=0, password=None, session_id=''):
+        """
+        初始化 Redis 连接
+        
+        Args:
+            host: Redis 主机地址
+            port: Redis 端口
+            db: Redis 数据库编号
+            password: Redis 密码
+            session_id: 会话ID
+        """
         self.r = redis.Redis(
             host=host,
             port=port,
@@ -16,210 +41,633 @@ class RedisSessionStore:
         )
         self.uudi = session_id
 
-    def _fix_encoding(self, text):
-        """修复错误编码的文本"""
-        if not text or not isinstance(text, str):
-            return text
-        try:
-            # 尝试修复 Latin-1 误编码为 UTF-8 的情况
-            return text.encode('latin-1').decode('utf-8')
-        except (UnicodeDecodeError, UnicodeEncodeError):
-            # 如果修复失败，返回原文本
-            return text
-
-    # 修改后的 save_session 方法
-    def save_session(self, userid, messages, aimessages, apply_id, group_id):
+    def save_session_write(self, userid: str, messages: str) -> str:
         """
         写入一条会话数据，返回 session_id
-        优化版本：确保写入时间不超过1秒
+
+        Args:
+            userid: 用户ID
+            messages: 用户消息
+
+        Returns:
+            str: 新生成的 session_id
         """
         try:
-            session_id = str(uuid.uuid4())  # 为每次会话生成新的 ID
-            starttime = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-            key = f"session:{session_id}"  # 使用新生成的 session_id 作为 key
+            messages = serialize_messages(messages)
+            session_id = str(uuid.uuid4())
+            key = generate_session_key(session_id, key_type="write")
 
-            # 使用 pipeline 批量写入，减少网络往返
             pipe = self.r.pipeline()
-
-            # 直接写入数据，decode_responses=True 已经处理了编码
             pipe.hset(key, mapping={
                 "id": self.uudi,
                 "sessionid": userid,
-                "apply_id": apply_id,
-                "group_id": group_id,
                 "messages": messages,
-                "aimessages": aimessages,
-                "starttime": starttime
+                "starttime": get_current_timestamp()
             })
-
-            # 可选：设置过期时间（例如30天），避免数据无限增长
-            # pipe.expire(key, 30 * 24 * 60 * 60)
-
-            # 执行批量操作
             result = pipe.execute()
 
-            print(f"保存结果: {result[0]}, session_id: {session_id}")
-            return session_id  # 返回新生成的 session_id
+            print(f"[save_session_write] 保存结果: {result[0]}, session_id: {session_id}")
+            return session_id
         except Exception as e:
-            print(f"保存会话失败: {e}")
+            print(f"[save_session_write] 保存会话失败: {e}")
             raise e
 
-    def save_sessions_batch(self, sessions_data):
+    def get_session_by_userid(self, userid: str) -> Union[List[Dict[str, str]], bool]:
         """
-        批量写入多条会话数据，返回 session_id 列表
-        sessions_data: list of dict, 每个 dict 包含 userid, messages, aimessages, apply_id, group_id
-        优化版本：批量操作，大幅提升性能
+        通过 save_session_write 的 userid 获取 sessionid 和 messages
+        
+        Args:
+            userid: 用户ID (对应 sessionid 字段)
+            
+        Returns:
+            List[Dict] 或 False: 如果找到数据返回 [{"sessionid": "...", "messages": "..."}, ...]，否则返回 False
         """
         try:
-            session_ids = []
+            # 只查询 write 类型的 key
+            keys = self.r.keys('session:write:*')
+            if not keys:
+                return False
+
+            # 批量获取数据
             pipe = self.r.pipeline()
+            for key in keys:
+                pipe.hgetall(key)
+            all_data = pipe.execute()
 
-            for session in sessions_data:
-                session_id = str(uuid.uuid4())
-                starttime = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-                key = f"session:{session_id}"
-
-                pipe.hset(key, mapping={
-                    "id": self.uudi,
-                    "sessionid": session.get('userid'),
-                    "apply_id": session.get('apply_id'),
-                    "group_id": session.get('group_id'),
-                    "messages": session.get('messages'),
-                    "aimessages": session.get('aimessages'),
-                    "starttime": starttime
-                })
-
-                session_ids.append(session_id)
-
-            # 一次性执行所有写入操作
-            results = pipe.execute()
-            print(f"批量保存完成: {len(session_ids)} 条记录")
-            return session_ids
+            # 筛选符合 userid 的数据
+            results = []
+            for key, data in zip(keys, all_data):
+                if not data:
+                    continue
+                
+                # 从 write 类型读取，匹配 sessionid 字段
+                if data.get('sessionid') == userid:
+                    # 从 key 中提取 session_id: session:write:{session_id}
+                    session_id = key.split(':')[-1]
+                    results.append({
+                        "sessionid": session_id,
+                        "messages": fix_encoding(data.get('messages', ''))
+                    })
+            
+            if not results:
+                return False
+            
+            print(f"[get_session_by_userid] userid={userid}, 找到 {len(results)} 条数据")
+            return results
         except Exception as e:
-            print(f"批量保存会话失败: {e}")
-            raise e
+            print(f"[get_session_by_userid] 查询失败: {e}")
+            return False
+    
+    def get_all_sessions_by_end_user_id(self, end_user_id: str) -> Union[List[Dict[str, Any]], bool]:
+        """
+        通过 end_user_id 获取所有 write 类型的会话数据
+        
+        Args:
+            end_user_id: 终端用户ID (对应 sessionid 字段)
+            
+        Returns:
+            List[Dict] 或 False: 如果找到数据返回完整的会话信息列表，否则返回 False
+            
+        返回格式:
+        [
+            {
+                "session_id": "uuid",
+                "id": "...",
+                "sessionid": "end_user_id",
+                "messages": "...",
+                "starttime": "timestamp"
+            },
+            ...
+        ]
+        """
+        try:
+            # 只查询 write 类型的 key
+            keys = self.r.keys('session:write:*')
+            if not keys:
+                print(f"[get_all_sessions_by_end_user_id] 没有找到任何 write 类型的会话")
+                return False
 
-    # ---------------- 读取 ----------------
-    def get_session(self, session_id):
-        """
-        读取一条会话数据
-        """
-        key = f"session:{session_id}"
-        data = self.r.hgetall(key)
-        return data if data else None
+            # 批量获取数据
+            pipe = self.r.pipeline()
+            for key in keys:
+                pipe.hgetall(key)
+            all_data = pipe.execute()
 
-    def get_session_apply_group(self, sessionid, apply_id, group_id):
-        """
-        根据 sessionid、apply_id 和 group_id 三个条件查询会话数据
-        """
-        result_items = []
+            # 筛选符合 end_user_id 的数据
+            results = []
+            for key, data in zip(keys, all_data):
+                if not data:
+                    continue
+                
+                # 从 write 类型读取，匹配 sessionid 字段
+                if data.get('sessionid') == end_user_id:
+                    # 从 key 中提取 session_id: session:write:{session_id}
+                    session_id = key.split(':')[-1]
+                    
+                    # 构建完整的会话信息
+                    session_info = {
+                        "session_id": session_id,
+                        "id": data.get('id', ''),
+                        "sessionid": data.get('sessionid', ''),
+                        "messages": fix_encoding(data.get('messages', '')),
+                        "starttime": data.get('starttime', '')
+                    }
+                    results.append(session_info)
+            
+            if not results:
+                print(f"[get_all_sessions_by_end_user_id] end_user_id={end_user_id}, 没有找到数据")
+                return False
+            
+            # 按时间排序（最新的在前）
+            results.sort(key=lambda x: x.get('starttime', ''), reverse=True)
+            
+            print(f"[get_all_sessions_by_end_user_id] end_user_id={end_user_id}, 找到 {len(results)} 条数据")
+            return results
+        except Exception as e:
+            print(f"[get_all_sessions_by_end_user_id] 查询失败: {e}")
+            import traceback
+            traceback.print_exc()
+            return False
 
-        # 遍历所有会话数据
-        for key in self.r.keys('session:*'):
-            data = self.r.hgetall(key)
-
-            if not data:
-                continue
-
-            # 检查三个条件是否都匹配
-            if (data.get('sessionid') == sessionid and
-                    data.get('apply_id') == apply_id and
-                    data.get('group_id') == group_id):
-                result_items.append(data)
-
-        return result_items
-
-    def get_all_sessions(self):
+    def find_user_recent_sessions(self, userid: str, 
+                                  minutes: int = 5) -> List[Dict[str, str]]:
         """
-        获取所有会话数据
-        """
-        sessions = {}
-        for key in self.r.keys('session:*'):
-            sid = key.split(':')[1]
-            sessions[sid] = self.get_session(sid)
-        return sessions
-
-    # ---------------- 更新 ----------------
-    def update_session(self, session_id, field, value):
-        """
-        更新单个字段
-        优化版本：使用 pipeline 减少网络往返
-        """
-        key = f"session:{session_id}"
-        pipe = self.r.pipeline()
-        pipe.exists(key)
-        pipe.hset(key, field, value)
-        results = pipe.execute()
-        return bool(results[0])  # 返回 key 是否存在
-
-    # ---------------- 删除 ----------------
-    def delete_session(self, session_id):
-        """
-        删除单条会话
-        """
-        key = f"session:{session_id}"
-        return self.r.delete(key)
-
-    def delete_all_sessions(self):
-        """
-        删除所有会话
-        """
-        keys = self.r.keys('session:*')
-        if keys:
-            return self.r.delete(*keys)
-        return 0
-
-    def delete_duplicate_sessions(self):
-        """
-        删除重复会话数据，条件：
-        "sessionid"、"user_id"、"group_id"、"messages"、"aimessages" 五个字段都相同的只保留一个，其他删除
-        优化版本：使用 pipeline 批量操作，确保在1秒内完成
+        根据 userid 从 save_session_write 写入的数据中查询最近 N 分钟内的会话数据
+        
+        Args:
+            userid: 用户ID (对应 sessionid 字段)
+            minutes: 查询最近几分钟的数据，默认5分钟
+            
+        Returns:
+            List[Dict]: 会话列表 [{"Query": "...", "Answer": "..."}, ...]
         """
         import time
         start_time = time.time()
-
-        # 第一步：使用 pipeline 批量获取所有 key
-        keys = self.r.keys('session:*')
-
+        
+        # 只查询 write 类型的 key
+        keys = self.r.keys('session:write:*')
         if not keys:
-            print("[delete_duplicate_sessions] 没有会话数据")
-            return 0
+            print(f"[find_user_recent_sessions] 查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
+            return []
 
-        # 第二步：使用 pipeline 批量获取所有数据
+        # 批量获取数据
         pipe = self.r.pipeline()
         for key in keys:
             pipe.hgetall(key)
         all_data = pipe.execute()
 
-        # 第三步：在内存中识别重复数据
-        seen = {}  # 用字典记录：identifier -> key（保留第一个出现的 key）
-        keys_to_delete = []  # 需要删除的 key 列表
+        # 筛选符合 userid 的数据
+        matched_items = []
+        for data in all_data:
+            if not data:
+                continue
+            
+            # 从 write 类型读取，匹配 sessionid 字段
+            if data.get('sessionid') == userid and data.get('starttime'):
+                # write 类型没有 aimessages，所以 Answer 为空
+                matched_items.append({
+                    "Query": fix_encoding(data.get('messages', '')),
+                    "Answer": "",
+                    "starttime": data.get('starttime', '')
+                })
+        
+        # 根据时间范围过滤
+        filtered_items = filter_by_time_range(matched_items, minutes)
+        # 排序并移除时间字段
+        result_items = sort_and_limit_results(filtered_items, limit=None)
+        print(result_items)
 
-        for key, data in zip(keys, all_data, strict=False):
+        elapsed_time = time.time() - start_time
+        print(f"[find_user_recent_sessions] userid={userid}, minutes={minutes}, "
+              f"查询耗时: {elapsed_time:.3f}秒, 结果数: {len(result_items)}")
+
+        return result_items
+
+    def delete_all_write_sessions(self) -> int:
+        """
+        删除所有 write 类型的会话
+        
+        Returns:
+            int: 删除的数量
+        """
+        keys = self.r.keys('session:write:*')
+        if keys:
+            return self.r.delete(*keys)
+        return 0
+
+
+class RedisCountStore:
+    """Redis Count 类型存储类，用于管理访问次数统计相关的数据"""
+    
+    def __init__(self, host='localhost', port=6379, db=0, password=None, session_id=''):
+        """
+        初始化 Redis 连接
+        
+        Args:
+            host: Redis 主机地址
+            port: Redis 端口
+            db: Redis 数据库编号
+            password: Redis 密码
+            session_id: 会话ID
+        """
+        self.r = redis.Redis(
+            host=host,
+            port=port,
+            db=db,
+            password=password,
+            decode_responses=True,
+            encoding='utf-8'
+        )
+        self.uudi = session_id
+
+    def save_sessions_count(self, end_user_id: str, count: int, messages: Any) -> str:
+        """
+        保存用户访问次数统计
+        
+        Args:
+            end_user_id: 终端用户ID
+            count: 访问次数
+            messages: 消息内容
+            
+        Returns:
+            str: 新生成的 session_id
+        """
+        session_id = str(uuid.uuid4())
+        key = generate_session_key(session_id, key_type="count")
+        index_key = f'session:count:index:{end_user_id}'  # 索引键
+        
+        pipe = self.r.pipeline()
+        pipe.hset(key, mapping={
+            "id": self.uudi,
+            "end_user_id": end_user_id,
+            "count": int(count),
+            "messages": serialize_messages(messages),
+            "starttime": get_current_timestamp()
+        })
+        pipe.expire(key, 30 * 24 * 60 * 60)  # 30天过期
+        
+        # 创建索引：end_user_id -> session_id 映射
+        pipe.set(index_key, session_id, ex=30 * 24 * 60 * 60)
+        
+        result = pipe.execute()
+        
+        print(f"[save_sessions_count] 保存结果: {result}, session_id: {session_id}")
+        return session_id
+
+    def get_sessions_count(self, end_user_id: str) -> Union[List[Any], bool]:
+        """
+        通过 end_user_id 查询访问次数统计
+        
+        Args:
+            end_user_id: 终端用户ID
+            
+        Returns:
+            list 或 False: 如果找到返回 [count, messages]，否则返回 False
+        """
+        try:
+            # 使用索引键快速查找
+            index_key = f'session:count:index:{end_user_id}'
+            
+            # 检查索引键类型，避免 WRONGTYPE 错误
+            try:
+                key_type = self.r.type(index_key)
+                if key_type != 'string' and key_type != 'none':
+                    self.r.delete(index_key)
+                    return False
+            except Exception as type_error:
+                print(f"[get_sessions_count] 检查键类型失败: {type_error}")
+            
+            session_id = self.r.get(index_key)
+            
+            if not session_id:
+                return False
+            
+            # 直接获取数据
+            key = generate_session_key(session_id, key_type="count")
+            data = self.r.hgetall(key)
+            
+            if not data:
+                # 索引存在但数据不存在，清理索引
+                self.r.delete(index_key)
+                return False
+            
+            count = data.get('count')
+            messages_str = data.get('messages')
+            
+            if count is not None:
+                messages = deserialize_messages(messages_str)
+                return [int(count), messages]
+            
+            return False
+        except Exception as e:
+            print(f"[get_sessions_count] 查询失败: {e}")
+            return False
+    def update_sessions_count(self, end_user_id: str, new_count: int, 
+                             messages: Any) -> bool:
+        """
+        通过 end_user_id 修改访问次数统计（优化版：使用索引）
+        
+        Args:
+            end_user_id: 终端用户ID
+            new_count: 新的 count 值
+            messages: 消息内容
+            
+        Returns:
+            bool: 更新成功返回 True，未找到记录返回 False
+        """
+        try:
+            # 使用索引键快速查找
+            index_key = f'session:count:index:{end_user_id}'
+            
+            # 检查索引键类型，避免 WRONGTYPE 错误
+            try:
+                key_type = self.r.type(index_key)
+                if key_type != 'string' and key_type != 'none':
+                    # 索引键类型错误，删除并返回 False
+                    print(f"[update_sessions_count] 索引键类型错误: {key_type}，删除索引")
+                    self.r.delete(index_key)
+                    print(f"[update_sessions_count] 未找到记录: end_user_id={end_user_id}")
+                    return False
+            except Exception as type_error:
+                print(f"[update_sessions_count] 检查键类型失败: {type_error}")
+            
+            session_id = self.r.get(index_key)
+            
+            if not session_id:
+                print(f"[update_sessions_count] 未找到记录: end_user_id={end_user_id}")
+                return False
+            
+            # 直接更新数据
+            key = generate_session_key(session_id, key_type="count")
+            messages_str = serialize_messages(messages)
+            
+            pipe = self.r.pipeline()
+            pipe.hset(key, 'count', int(new_count))
+            pipe.hset(key, 'messages', messages_str)
+            result = pipe.execute()
+            
+            print(f"[update_sessions_count] 更新成功: end_user_id={end_user_id}, new_count={new_count}, key={key}")
+            return True
+            
+        except Exception as e:
+            print(f"[update_sessions_count] 更新失败: {e}")
+            return False
+
+    def delete_all_count_sessions(self) -> int:
+        """
+        删除所有 count 类型的会话
+        
+        Returns:
+            int: 删除的数量
+        """
+        keys = self.r.keys('session:count:*')
+        if keys:
+            return self.r.delete(*keys)
+        return 0
+
+
+class RedisSessionStore:
+    """Redis 会话存储类，用于管理会话数据"""
+    
+    def __init__(self, host='localhost', port=6379, db=0, password=None, session_id=''):
+        """
+        初始化 Redis 连接
+        
+        Args:
+            host: Redis 主机地址
+            port: Redis 端口
+            db: Redis 数据库编号
+            password: Redis 密码
+            session_id: 会话ID
+        """
+        self.r = redis.Redis(
+            host=host,
+            port=port,
+            db=db,
+            password=password,
+            decode_responses=True,
+            encoding='utf-8'
+        )
+        self.uudi = session_id
+
+    # ==================== 写入操作 ====================
+    
+    def save_session(self, userid: str, messages: str, aimessages: str, 
+                    apply_id: str, end_user_id: str) -> str:
+        """
+        写入一条会话数据，返回 session_id
+        
+        Args:
+            userid: 用户ID
+            messages: 用户消息
+            aimessages: AI回复消息
+            apply_id: 应用ID
+            end_user_id: 终端用户ID
+            
+        Returns:
+            str: 新生成的 session_id
+        """
+        try:
+            session_id = str(uuid.uuid4())
+            key = generate_session_key(session_id, key_type="read")
+
+            pipe = self.r.pipeline()
+            pipe.hset(key, mapping={
+                "id": self.uudi,
+                "sessionid": userid,
+                "apply_id": apply_id,
+                "end_user_id": end_user_id,
+                "messages": messages,
+                "aimessages": aimessages,
+                "starttime": get_current_timestamp()
+            })
+            result = pipe.execute()
+
+            print(f"[save_session] 保存结果: {result[0]}, session_id: {session_id}")
+            return session_id
+        except Exception as e:
+            print(f"[save_session] 保存会话失败: {e}")
+            raise e
+
+    # ==================== 读取操作 ====================
+    
+    def get_session(self, session_id: str) -> Optional[Dict[str, Any]]:
+        """
+        读取一条会话数据
+        
+        Args:
+            session_id: 会话ID
+            
+        Returns:
+            Dict 或 None: 会话数据
+        """
+        key = generate_session_key(session_id)
+        data = self.r.hgetall(key)
+        return data if data else None
+
+    def get_all_sessions(self) -> Dict[str, Dict[str, Any]]:
+        """
+        获取所有会话数据（不包括 count 和 write 类型）
+        
+        Returns:
+            Dict: 所有会话数据，key 为 session_id
+        """
+        sessions = {}
+        for key in self.r.keys('session:*'):
+            # 排除 count 和 write 类型的 key
+            if ':count:' not in key and ':write:' not in key:
+                sid = key.split(':')[1]
+                sessions[sid] = self.get_session(sid)
+        return sessions
+
+    def find_user_apply_group(self, sessionid: str, apply_id: str, 
+                             end_user_id: str) -> List[Dict[str, str]]:
+        """
+        根据 sessionid、apply_id 和 end_user_id 查询会话数据，返回最新的6条
+        
+        Args:
+            sessionid: 会话ID（支持模糊匹配）
+            apply_id: 应用ID
+            end_user_id: 终端用户ID
+            
+        Returns:
+            List[Dict]: 会话列表 [{"Query": "...", "Answer": "..."}, ...]
+        """
+        import time
+        start_time = time.time()
+        
+        keys = self.r.keys('session:*')
+        if not keys:
+            print(f"[find_user_apply_group] 查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
+            return []
+
+        # 批量获取数据
+        pipe = self.r.pipeline()
+        for key in keys:
+            # 排除 count 和 write 类型
+            if ':count:' not in key and ':write:' not in key:
+                pipe.hgetall(key)
+        all_data = pipe.execute()
+
+        # 筛选符合条件的数据
+        matched_items = []
+        for data in all_data:
             if not data:
                 continue
 
-            # 获取五个字段的值
-            sessionid = data.get('sessionid', '')
-            user_id = data.get('id', '')
-            group_id = data.get('group_id', '')
-            messages = data.get('messages', '')
-            aimessages = data.get('aimessages', '')
+            if (data.get('apply_id') == apply_id and
+                data.get('end_user_id') == end_user_id):
+                # 支持模糊匹配或完全匹配 sessionid
+                if sessionid in data.get('sessionid', '') or data.get('sessionid') == sessionid:
+                    matched_items.append(format_session_data(data, include_time=True))
+        
+        # 排序、限制数量并移除时间字段
+        result_items = sort_and_limit_results(matched_items, limit=6)
+
+        elapsed_time = time.time() - start_time
+        print(f"[find_user_apply_group] 查询耗时: {elapsed_time:.3f}秒, 结果数: {len(result_items)}")
+
+        return result_items
+
+    # ==================== 更新操作 ====================
+    
+    def update_session(self, session_id: str, field: str, value: Any) -> bool:
+        """
+        更新单个字段
+        
+        Args:
+            session_id: 会话ID
+            field: 字段名
+            value: 字段值
+            
+        Returns:
+            bool: 是否更新成功
+        """
+        key = generate_session_key(session_id)
+        pipe = self.r.pipeline()
+        pipe.exists(key)
+        pipe.hset(key, field, value)
+        results = pipe.execute()
+        return bool(results[0])
+
+    # ==================== 删除操作 ====================
+    
+    def delete_session(self, session_id: str) -> int:
+        """
+        删除单条会话
+        
+        Args:
+            session_id: 会话ID
+            
+        Returns:
+            int: 删除的数量
+        """
+        key = generate_session_key(session_id)
+        return self.r.delete(key)
+
+    def delete_all_sessions(self) -> int:
+        """
+        删除所有会话（不包括 count 和 write 类型）
+        
+        Returns:
+            int: 删除的数量
+        """
+        keys = self.r.keys('session:*')
+        # 过滤掉 count 和 write 类型
+        keys_to_delete = [k for k in keys if ':count:' not in k and ':write:' not in k]
+        if keys_to_delete:
+            return self.r.delete(*keys_to_delete)
+        return 0
+
+    def delete_duplicate_sessions(self) -> int:
+        """
+        删除重复会话数据（不包括 count 和 write 类型）
+        条件：sessionid、user_id、end_user_id、messages、aimessages 五个字段都相同的只保留一个
+        
+        Returns:
+            int: 删除的数量
+        """
+        import time
+        start_time = time.time()
+
+        keys = self.r.keys('session:*')
+        if not keys:
+            print("[delete_duplicate_sessions] 没有会话数据")
+            return 0
+
+        # 批量获取所有数据
+        pipe = self.r.pipeline()
+        for key in keys:
+            # 排除 count 和 write 类型
+            if ':count:' not in key and ':write:' not in key:
+                pipe.hgetall(key)
+        all_data = pipe.execute()
+
+        # 识别重复数据
+        seen = {}
+        keys_to_delete = []
+
+        for key, data in zip([k for k in keys if ':count:' not in k and ':write:' not in k], all_data, strict=False):
+            if not data:
+                continue
 
             # 用五元组作为唯一标识
-            identifier = (sessionid, user_id, group_id, messages, aimessages)
+            identifier = (
+                data.get('sessionid', ''),
+                data.get('id', ''),
+                data.get('end_user_id', ''),
+                data.get('messages', ''),
+                data.get('aimessages', '')
+            )
 
             if identifier in seen:
-                # 重复，标记为待删除
                 keys_to_delete.append(key)
             else:
-                # 第一次出现，记录
                 seen[identifier] = key
 
-        # 第四步：使用 pipeline 批量删除重复的 key
+        # 批量删除重复的 key
         deleted_count = 0
         if keys_to_delete:
-            # 分批删除，避免单次操作过大
             batch_size = 1000
             for i in range(0, len(keys_to_delete), batch_size):
                 batch = keys_to_delete[i:i + batch_size]
@@ -233,79 +681,28 @@ class RedisSessionStore:
         print(f"[delete_duplicate_sessions] 删除重复会话数量: {deleted_count}, 耗时: {elapsed_time:.3f}秒")
         return deleted_count
 
-    def find_user_session(self, sessionid):
-        user_id = sessionid
-
-        result_items = []
-        for key, values in store.get_all_sessions().items():
-            history = {}
-            if user_id == str(values['sessionid']):
-                history["Query"] = values['messages']
-                history["Answer"] = values['aimessages']
-                result_items.append(history)
-
-        if len(result_items) <= 1:
-            result_items = []
-        return (result_items)
-
-    def find_user_apply_group(self, sessionid, apply_id, group_id):
-        """
-        根据 sessionid、apply_id 和 group_id 三个条件查询会话数据，返回最新的6条
-        """
-        import time
-        start_time = time.time()
-        # 使用 pipeline 批量获取数据，提高性能
-        keys = self.r.keys('session:*')
-
-        if not keys:
-            print(f"查询耗时: {time.time() - start_time:.3f}秒, 结果数: 0")
-            return []
-
-        # 使用 pipeline 批量获取所有 hash 数据
-        pipe = self.r.pipeline()
-        for key in keys:
-            pipe.hgetall(key)
-        all_data = pipe.execute()
-
-        # 解析并筛选符合条件的数据
-        matched_items = []
-        for data in all_data:
-            if not data:
-                continue
-
-            # 检查是否符合三个条件
-
-            if (data.get('apply_id') == apply_id and
-                    data.get('group_id') == group_id):
-                # 支持模糊匹配 sessionid 或者完全匹配
-                if sessionid in data.get('sessionid', '') or data.get('sessionid') == sessionid:
-                    matched_items.append({
-                        "Query": self._fix_encoding(data.get('messages')),
-                        "Answer": self._fix_encoding(data.get('aimessages')),
-                        "starttime": data.get('starttime', '')
-                    })
-        # 按时间降序排序（最新的在前）
-        matched_items.sort(key=lambda x: x.get('starttime', ''), reverse=True)
-        # 只保留最新的6条
-        result_items = matched_items[:6]
-        # # 移除 starttime 字段
-        for item in result_items:
-            item.pop('starttime', None)
-
-        # 如果结果少于等于1条，返回空列表
-        if len(result_items) <= 1:
-            result_items = []
-
-        elapsed_time = time.time() - start_time
-        print(f"查询耗时: {elapsed_time:.3f}秒, 结果数: {len(result_items)}")
-
-        return result_items
-
 
+# 全局实例
 store = RedisSessionStore(
     host=settings.REDIS_HOST,
     port=settings.REDIS_PORT,
     db=settings.REDIS_DB,
     password=settings.REDIS_PASSWORD if settings.REDIS_PASSWORD else None,
     session_id=str(uuid.uuid4())
-)
+)
+
+write_store = RedisWriteStore(
+    host=settings.REDIS_HOST,
+    port=settings.REDIS_PORT,
+    db=settings.REDIS_DB,
+    password=settings.REDIS_PASSWORD if settings.REDIS_PASSWORD else None,
+    session_id=str(uuid.uuid4())
+)
+
+count_store = RedisCountStore(
+    host=settings.REDIS_HOST,
+    port=settings.REDIS_PORT,
+    db=settings.REDIS_DB,
+    password=settings.REDIS_PASSWORD if settings.REDIS_PASSWORD else None,
+    session_id=str(uuid.uuid4())
+)

api/app/core/memory/agent/utils/session_tools.py

@@ -0,0 +1,169 @@
+"""
+Session Service for managing user sessions and conversation history.
+
+This service provides clean Redis interactions with error handling and
+session management utilities.
+"""
+from typing import List, Optional
+
+from app.core.logging_config import get_agent_logger
+from app.core.memory.agent.utils.redis_tool import RedisSessionStore
+
+
+logger = get_agent_logger(__name__)
+
+
+class SessionService:
+    """Service for managing user sessions and conversation history."""
+    
+    def __init__(self, store: RedisSessionStore):
+        """
+        Initialize the session service.
+        
+        Args:
+            store: Redis session store instance
+        """
+        self.store = store
+        logger.info("SessionService initialized")
+    
+    def resolve_user_id(self, session_string: str) -> str:
+        """
+        Extract user ID from session string.
+        
+        Handles formats like:
+        - 'call_id_user123' -> 'user123'
+        - 'prefix_id_user456_suffix' -> 'user456_suffix'
+        
+        Args:
+            session_string: Session identifier string
+            
+        Returns:
+            Extracted user ID
+        """
+        try:
+            # Split by '_id_' and take everything after it
+            parts = session_string.split('_id_')
+            if len(parts) > 1:
+                return parts[1]
+            
+            # Fallback: return original string
+            return session_string
+            
+        except Exception as e:
+            logger.warning(
+                f"Failed to parse user ID from session string '{session_string}': {e}"
+            )
+            return session_string
+    
+    async def get_history(
+        self,
+        user_id: str,
+        apply_id: str,
+        end_user_id: str
+    ) -> List[dict]:
+        """
+        Retrieve conversation history from Redis.
+        
+        Args:
+            user_id: User identifier
+            apply_id: Application identifier
+            end_user_id: Group identifier
+            
+        Returns:
+            List of conversation history items with Query and Answer keys
+            Returns empty list if no history found or on error
+        """
+        try:
+            history = self.store.find_user_apply_group(user_id, apply_id, end_user_id)
+            
+            # Validate history structure
+            if not isinstance(history, list):
+                logger.warning(
+                    f"Invalid history format for user {user_id}, "
+                    f"apply {apply_id}, group {end_user_id}: expected list, got {type(history)}"
+                )
+                return []
+            
+            return history
+            
+        except Exception as e:
+            logger.error(
+                f"Failed to retrieve history for user {user_id}, "
+                f"apply {apply_id}, group {end_user_id}: {e}",
+                exc_info=True
+            )
+            # Return empty list on error to allow execution to continue
+            return []
+    
+    async def save_session(
+        self,
+        user_id: str,
+        query: str,
+        apply_id: str,
+        end_user_id: str,
+        ai_response: str
+    ) -> Optional[str]:
+        """
+        Save conversation turn to Redis.
+        
+        Args:
+            user_id: User identifier
+            query: User query/message
+            apply_id: Application identifier
+            end_user_id: Group identifier
+            ai_response: AI response/answer
+            
+        Returns:
+            Session ID if successful, None on error
+        """
+        try:
+            # Validate required fields
+            if not user_id:
+                logger.warning("Cannot save session: user_id is empty")
+                return None
+            
+            if not query:
+                logger.warning("Cannot save session: query is empty")
+                return None
+            
+            # Save session
+            session_id = self.store.save_session(
+                userid=user_id,
+                messages=query,
+                apply_id=apply_id,
+                end_user_id=end_user_id,
+                aimessages=ai_response
+            )
+            
+            logger.info(f"Session saved successfully: {session_id}")
+            return session_id
+            
+        except Exception as e:
+            logger.error(
+                f"Failed to save session for user {user_id}: {e}",
+                exc_info=True
+            )
+            return None
+    
+    async def cleanup_duplicates(self) -> int:
+        """
+        Remove duplicate session entries.
+        
+        Duplicates are identified by matching:
+        - sessionid
+        - user_id (id field)
+        - end_user_id
+        - messages
+        - aimessages
+        
+        Returns:
+            Number of duplicate sessions deleted
+        """
+        try:
+            deleted_count = self.store.delete_duplicate_sessions()
+            logger.info(f"Cleaned up {deleted_count} duplicate sessions")
+            return deleted_count
+            
+        except Exception as e:
+            logger.error(f"Failed to cleanup duplicate sessions: {e}", exc_info=True)
+            return 0

api/app/core/memory/agent/utils/template_tools.py

@@ -0,0 +1,117 @@
+"""
+Template Service for loading and rendering Jinja2 templates.
+
+This service provides centralized template management with caching and error handling.
+"""
+# 标准库
+import os
+from functools import lru_cache
+
+from jinja2 import Environment, FileSystemLoader, Template, TemplateNotFound
+
+from app.core.logging_config import get_agent_logger, log_prompt_rendering
+
+
+logger = get_agent_logger(__name__)
+
+
+class TemplateRenderError(Exception):
+    """Exception raised when template rendering fails."""
+    
+    def __init__(self, template_name: str, error: Exception, variables: dict):
+        self.template_name = template_name
+        self.error = error
+        self.variables = variables
+        super().__init__(
+            f"Failed to render template '{template_name}': {str(error)}"
+        )
+
+
+class TemplateService:
+    """Service for loading and rendering Jinja2 templates with caching."""
+    
+    def __init__(self, template_root: str):
+        """
+        Initialize the template service.
+        
+        Args:
+            template_root: Root directory containing template files
+        """
+        self.template_root = template_root
+        self.env = Environment(
+            loader=FileSystemLoader(template_root),
+            autoescape=False  # Disable autoescape for prompt templates
+        )
+        logger.info(f"TemplateService initialized with root: {template_root}")
+    
+    @lru_cache(maxsize=128)
+    def _load_template(self, template_name: str) -> Template:
+        """
+        Load a template from disk with caching.
+        
+        Args:
+            template_name: Relative path to template file
+            
+        Returns:
+            Loaded Jinja2 Template object
+            
+        Raises:
+            TemplateNotFound: If template file doesn't exist
+        """
+        try:
+            return self.env.get_template(template_name)
+        except TemplateNotFound as e:
+            expected_path = os.path.join(self.template_root, template_name)
+            logger.error(
+                f"Template not found: {template_name}. "
+                f"Expected path: {expected_path}"
+            )
+            raise
+    
+    async def render_template(
+        self,
+        template_name: str,
+        operation_name: str,
+        **variables
+    ) -> str:
+        """
+        Load and render a Jinja2 template.
+        
+        Args:
+            template_name: Relative path to template file
+            operation_name: Name for logging (e.g., "split_the_problem")
+            **variables: Template variables to render
+            
+        Returns:
+            Rendered template string
+            
+        Raises:
+            TemplateRenderError: If template loading or rendering fails
+        """
+        try:
+            # Load template (cached)
+            template = self._load_template(template_name)
+            
+            # Render template
+            rendered = template.render(**variables)
+            
+            # Log rendered prompt
+            log_prompt_rendering(operation_name, rendered)
+            
+            return rendered
+            
+        except TemplateNotFound as e:
+            logger.error(
+                f"Template rendering failed for {operation_name} "
+                f"({template_name}): Template not found",
+                exc_info=True
+            )
+            raise TemplateRenderError(template_name, e, variables)
+            
+        except Exception as e:
+            logger.error(
+                f"Template rendering failed for {operation_name} "
+                f"({template_name}): {e}",
+                exc_info=True
+            )
+            raise TemplateRenderError(template_name, e, variables)

api/app/core/memory/agent/utils/type_classifier.py

@@ -1,10 +1,9 @@
 """
 Type classification utility for distinguishing read/write operations.
 """
-from app.core.config import settings
 from app.core.logging_config import get_agent_logger, log_prompt_rendering
 from app.core.memory.agent.utils.llm_tools import PROJECT_ROOT_
-from app.core.memory.agent.utils.messages_tool import read_template_file
+from app.core.memory.agent.utils.messages_tools import read_template_file
 from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
 from app.db import get_db_context
 from jinja2 import Template

api/app/core/memory/agent/utils/write_to_database.py

@@ -1,49 +0,0 @@
-import os
-import uuid
-from datetime import datetime
-from typing import Any
-from sqlalchemy.orm import Session
-import logging
-import json
-
-from app.db import get_db
-from app.models.retrieval_info import RetrievalInfo
-
-logger = logging.getLogger(__name__)
-
-async def write_to_database(host_id: uuid.UUID, data: Any) -> str:
-    """
-    将数据写入数据库
-    :param host_id: 宿主 ID
-    :param data: 要写入的数据
-    :return: 写入数据库的结果
-    """
-    # 从数据库会话中获取会话
-    db: Session = next(get_db())
-    try:
-        if isinstance(data, (dict, list)):
-            serialized = json.dumps(data, ensure_ascii=False)
-        elif isinstance(data, str):
-            serialized = data
-        else:
-            serialized = str(data)
-
-        new_retrieval_info = RetrievalInfo(
-            # host_id=host_id,
-            host_id=uuid.UUID("2f6ff1eb-50c7-4765-8e89-e4566be19122"),
-            retrieve_info=serialized,
-            created_at=datetime.now()
-        )
-        db.add(new_retrieval_info)
-        db.commit()
-        logger.info(f"success to write data to database, host_id: {host_id}, retrieve_info: {serialized}")
-        return "success to write data to database"
-    except Exception as e:
-        db.rollback()
-        logger.error(f"failed to write data to database, host_id: {host_id}, retrieve_info: {data}, error: {e}")
-        raise e
-    finally:
-        try:
-            db.close()
-        except Exception:
-            pass

api/app/core/memory/agent/utils/write_tools.py

@@ -4,17 +4,16 @@ Write Tools for Memory Knowledge Extraction Pipeline
 This module provides the main write function for executing the knowledge extraction
 pipeline. Only MemoryConfig is needed - clients are constructed internally.
 """
+import asyncio
 import time
 from datetime import datetime
 
+from dotenv import load_dotenv
+
 from app.core.logging_config import get_agent_logger
 from app.core.memory.agent.utils.get_dialogs import get_chunked_dialogs
-from app.core.memory.storage_services.extraction_engine.extraction_orchestrator import (
-    ExtractionOrchestrator,
-)
-from app.core.memory.storage_services.extraction_engine.knowledge_extraction.memory_summary import (
-    memory_summary_generation,
-)
+from app.core.memory.storage_services.extraction_engine.extraction_orchestrator import ExtractionOrchestrator
+from app.core.memory.storage_services.extraction_engine.knowledge_extraction.memory_summary import memory_summary_generation
 from app.core.memory.utils.llm.llm_utils import MemoryClientFactory
 from app.core.memory.utils.log.logging_utils import log_time
 from app.db import get_db_context
@@ -23,7 +22,7 @@ from app.repositories.neo4j.add_nodes import add_memory_summary_nodes
 from app.repositories.neo4j.graph_saver import save_dialog_and_statements_to_neo4j
 from app.repositories.neo4j.neo4j_connector import Neo4jConnector
 from app.schemas.memory_config_schema import MemoryConfig
-from dotenv import load_dotenv
+
 
 load_dotenv()
 
@@ -31,39 +30,34 @@ logger = get_agent_logger(__name__)
 
 
 async def write(
-    content: str,
-    user_id: str,
-    apply_id: str,
-    group_id: str,
+    end_user_id: str,
     memory_config: MemoryConfig,
+    messages: list,
     ref_id: str = "wyl20251027",
 ) -> None:
     """
     Execute the complete knowledge extraction pipeline.
-    
-    Only MemoryConfig is needed - LLM and embedding clients are constructed
-    internally from the config.
 
     Args:
-        content: Dialogue content to process
         user_id: User identifier
         apply_id: Application identifier
-        group_id: Group identifier
+        end_user_id: Group identifier
         memory_config: MemoryConfig object containing all configuration
+        messages: Structured message list [{"role": "user", "content": "..."}, ...]
         ref_id: Reference ID, defaults to "wyl20251027"
     """
     # Extract config values
     embedding_model_id = str(memory_config.embedding_model_id)
     chunker_strategy = memory_config.chunker_strategy
     config_id = str(memory_config.config_id)
-    
+
     logger.info("=== MemSci Knowledge Extraction Pipeline ===")
     logger.info(f"Config: {memory_config.config_name} (ID: {config_id})")
     logger.info(f"Workspace: {memory_config.workspace_name}")
     logger.info(f"LLM model: {memory_config.llm_model_name}")
     logger.info(f"Embedding model: {memory_config.embedding_model_name}")
     logger.info(f"Chunker strategy: {chunker_strategy}")
-    logger.info(f"Group ID: {group_id}")
+    logger.info(f"end_user_id ID: {end_user_id}")
 
     # Construct clients from memory_config using factory pattern with db session
     with get_db_context() as db:
@@ -88,10 +82,8 @@ async def write(
     step_start = time.time()
     chunked_dialogs = await get_chunked_dialogs(
         chunker_strategy=chunker_strategy,
-        group_id=group_id,
-        user_id=user_id,
-        apply_id=apply_id,
-        content=content,
+        end_user_id=end_user_id,
+        messages=messages,
         ref_id=ref_id,
         config_id=config_id,
     )
@@ -132,23 +124,48 @@ async def write(
     except Exception as e:
         logger.error(f"Error creating indexes: {e}", exc_info=True)
 
+    # 添加死锁重试机制
+    max_retries = 3
+    retry_delay = 1  # 秒
+
+    for attempt in range(max_retries):
+        try:
+            success = await save_dialog_and_statements_to_neo4j(
+                dialogue_nodes=all_dialogue_nodes,
+                chunk_nodes=all_chunk_nodes,
+                statement_nodes=all_statement_nodes,
+                entity_nodes=all_entity_nodes,
+                statement_chunk_edges=all_statement_chunk_edges,
+                statement_entity_edges=all_statement_entity_edges,
+                entity_edges=all_entity_entity_edges,
+                connector=neo4j_connector
+            )
+            if success:
+                logger.info("Successfully saved all data to Neo4j")
+                break
+            else:
+                logger.warning("Failed to save some data to Neo4j")
+                if attempt < max_retries - 1:
+                    logger.info(f"Retrying... (attempt {attempt + 2}/{max_retries})")
+                    await asyncio.sleep(retry_delay * (attempt + 1))  # 指数退避
+        except Exception as e:
+            error_msg = str(e)
+            # 检查是否是死锁错误
+            if "DeadlockDetected" in error_msg or "deadlock" in error_msg.lower():
+                if attempt < max_retries - 1:
+                    logger.warning(f"Deadlock detected, retrying... (attempt {attempt + 2}/{max_retries})")
+                    await asyncio.sleep(retry_delay * (attempt + 1))  # 指数退避
+                else:
+                    logger.error(f"Failed after {max_retries} attempts due to deadlock: {e}")
+                    raise
+            else:
+                # 非死锁错误，直接抛出
+                raise
+
     try:
-        success = await save_dialog_and_statements_to_neo4j(
-            dialogue_nodes=all_dialogue_nodes,
-            chunk_nodes=all_chunk_nodes,
-            statement_nodes=all_statement_nodes,
-            entity_nodes=all_entity_nodes,
-            statement_chunk_edges=all_statement_chunk_edges,
-            statement_entity_edges=all_statement_entity_edges,
-            entity_edges=all_entity_entity_edges,
-            connector=neo4j_connector
-        )
-        if success:
-            logger.info("Successfully saved all data to Neo4j")
-        else:
-            logger.warning("Failed to save some data to Neo4j")
-    finally:
         await neo4j_connector.close()
+    except Exception as e:
+        logger.error(f"Error closing Neo4j connector: {e}")
 
     log_time("Neo4j Database Save", time.time() - step_start, log_file)