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>
This commit is contained in:
Ke Sun
@@ -7,29 +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
|
||||
|
||||
from app.utils.config_utils import resolve_config_id
|
||||
|
||||
logger = get_business_logger()
|
||||
|
||||
|
||||
@@ -106,7 +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)
|
||||
}
|
||||
)
|
||||
|
||||
@@ -145,106 +137,8 @@ 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
|
||||
# TODO 乐力齐 - 累积多组对话批量写入功能已禁用
|
||||
# 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,
|
||||
# end_user_id=end_user_end,
|
||||
# aimessages=aimessages
|
||||
# )
|
||||
# store.delete_duplicate_sessions()
|
||||
# # logger.info(f'Redis_Agent:{end_user_end};{session_id}')
|
||||
# return session_id
|
||||
|
||||
# TODO 乐力齐 - 累积多组对话批量写入功能已禁用
|
||||
# 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, user_message, ai_message, user_rag_memory_id, actual_end_user_id, actual_config_id):
|
||||
"""
|
||||
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)
|
||||
|
||||
if storage_type == "rag":
|
||||
# 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}')
|
||||
else:
|
||||
# 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 user_message:
|
||||
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 ai_message:
|
||||
structured_messages.append({"role": "assistant", "content": ai_message})
|
||||
|
||||
# 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
|
||||
|
||||
# 调用 Celery 任务，传递结构化消息列表
|
||||
# 数据流：
|
||||
# 1. structured_messages 传递给 write_message_task
|
||||
# 2. write_message_task 调用 memory_agent_service.write_memory
|
||||
# 3. write_memory 调用 write_tools.write，传递 messages 参数
|
||||
# 4. write_tools.write 调用 get_chunked_dialogs，传递 messages 参数
|
||||
# 5. get_chunked_dialogs 为每条消息创建独立的 Chunk，设置 speaker 字段
|
||||
# 6. 每个 Chunk 保存到 Neo4j，包含 speaker 字段
|
||||
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: 结构化消息列表 [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]
|
||||
actual_config_id, # config_id: 配置ID
|
||||
storage_type, # storage_type: "neo4j"
|
||||
user_rag_memory_id # 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 chat(
|
||||
self,
|
||||
message: str,
|
||||
@@ -288,30 +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)}')
|
||||
# # TODO 乐力齐，在长短期记忆存储的时候再使用此代码
|
||||
# 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()
|
||||
|
||||
# # 长期记忆写入（
|
||||
# await self.write(storage_type, actual_end_user_id, history_term_memory, "", user_rag_memory_id, actual_end_user_id, 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)
|
||||
@@ -332,17 +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:
|
||||
# AI 回复写入（用户消息和 AI 回复配对，一次性写入完整对话）
|
||||
await self.write(storage_type, actual_end_user_id, message_chat, content, user_rag_memory_id, actual_end_user_id, actual_config_id)
|
||||
# TODO 乐力齐 - 累积多组对话批量写入功能已禁用
|
||||
# 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,
|
||||
@@ -350,7 +220,7 @@ class LangChainAgent:
|
||||
"usage": {
|
||||
"prompt_tokens": 0,
|
||||
"completion_tokens": 0,
|
||||
"total_tokens": 0
|
||||
"total_tokens": total_tokens
|
||||
}
|
||||
}
|
||||
|
||||
@@ -410,25 +280,7 @@ class LangChainAgent:
|
||||
db.close()
|
||||
except Exception as e:
|
||||
logger.warning(f"Failed to get db session: {e}")
|
||||
# # TODO 乐力齐
|
||||
# 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)}')
|
||||
# # 长期记忆写入
|
||||
# await self.write(storage_type, end_user_id, history_term_memory, "", user_rag_memory_id, end_user_id, actual_config_id)
|
||||
# except Exception as e:
|
||||
# logger.error(f"Failed to write to long term memory: {e}")
|
||||
# finally:
|
||||
# db_for_memory.close()
|
||||
|
||||
|
||||
# 注意：不在这里写入用户消息，等 AI 回复后一起写入
|
||||
try:
|
||||
@@ -444,7 +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},
|
||||
@@ -481,12 +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:
|
||||
# AI 回复写入（用户消息和 AI 回复配对，一次性写入完整对话）
|
||||
await self.write(storage_type, end_user_id, message_chat, full_content, user_rag_memory_id, end_user_id, actual_config_id)
|
||||
# TODO 乐力齐 - 累积多组对话批量写入功能已禁用
|
||||
# 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
|
||||
|
||||
Reference in New Issue
Block a user