fix(storage): 优化 SQL 存储兼容性

- 新增通用的 Content 清理逻辑，在 SQL Session 和 Memory
  存储读写时过滤空的 Content Part。
- 修复 MySQL 下 DynamicPickleType 写入 LONGBLOB
  时未序列化导致的类型错误。
- 支持通过 sessionmaker_kwargs 向 SqlStorage 传递 sessionmaker
  配置，例如 expire_on_commit。
- 将 SQL Session 中显式更新 update_time 的逻辑改为使用
  datetime.now()，避免 func.now() 带来的 ORM 状态问题。
- 统一通过当前事件循环创建 Session 和 Memory 服务的清理任务。
- 优化 SQL Session 示例输出，在助手回复前打印用户问题。
- 在 SQL Memory/Session 示例中关闭 thinking 输出，使演示结果更清晰。
- 增加 SQL Content 清理和 MySQL DynamicPickleType 序列化相关测试。

Author: raychen911

README.md

@@ -16,10 +16,10 @@ tRPC-Agent-Python provides an end-to-end foundation for agent building, orchestr
 
 - **Multi-paradigm agent orchestration**: Built-in orchestration supports `ChainAgent` / `ParallelAgent` / `CycleAgent` / `TransferAgent`, with `GraphAgent` for graph-based orchestration.
 - **Graph orchestration capability (`GraphAgent`)**: Use DSL to orchestrate `Agent` / `Tool` / `MCP` / `Knowledge` / `CodeExecutor` in one unified flow.
-- **Efficient integration with Python AI ecosystems**: Agent ecosystem extensions (`claude-agent-sdk` / `LangGraph`, etc.) / Tool ecosystem extensions (`mcp`, etc.) / Knowledge ecosystem extensions (`LangChain`, etc.) / Model ecosystem extensions (`LiteLLM`, etc.) / Memory ecosystem extensions (`Mem0`, etc.).
+- **Efficient integration with Python AI ecosystems**: Agent ecosystem extensions (`claude-agent-sdk` / `LangGraph`, etc.) / Tool ecosystem extensions (`mcp`, etc.) / Knowledge ecosystem extensions (`LangChain`, etc.) / Model ecosystem extensions (`LiteLLM`, etc.) / Memory ecosystem extensions (`Mem0`, `Mempalace`, etc.).
 - **Agent ecosystem extensions**: Supports `LangGraphAgent` / `ClaudeAgent` / `TeamAgent` (Agno-Like).
 - **Tool ecosystem extensions**: `FunctionTool` / File tools / `MCPToolset` / LangChain Tool / Agent-as-Tool.
-- **Complete memory capability (`Session` / `Memory`)**: `Session` manages messages and state within a single session, while `Memory` manages cross-session long-term memory and personalization. Persistence supports `InMemory` / `Redis` / `SQL`; `Memory` also supports `Mem0`.
+- **Complete memory capability (`Session` / `Memory`)**: `Session` manages messages and state within a single session, while `Memory` manages cross-session long-term memory and personalization. Persistence supports `InMemory` / `Redis` / `SQL`; `Memory` also supports `Mem0`、`Mempalace`.
 - **Production-grade knowledge capability**: Built on LangChain components with first-class RAG support.
 - **CodeExecutor extension capability**: Supports local / container executors for code execution and task grounding.
 - **Skills extension capability**: Supports `SKILL.md`-based skill systems for reusable capabilities and dynamic tooling.
@@ -84,7 +84,7 @@ pip install trpc-agent-py
 Install optional capabilities as needed:
 
 ```bash
-pip install trpc-agent-py[a2a,ag-ui,knowledge,agent-claude,mem0,langfuse]
+pip install trpc-agent-py[a2a,ag-ui,knowledge,agent-claude,mem0, Mempalace, langfuse]
 ```
 
 ### Develop Weather Agent
@@ -457,7 +457,7 @@ Related docs:
 This group helps you:
 
 - Session: manage per-session messages, summaries, and state
-- Memory: manage cross-session long-term memory (including Mem0)
+- Memory: manage cross-session long-term memory (including Mem0, Mempalace)
 - Knowledge: cover document loading, retrieval, RAG, and prompt templates
 
 ### 10. Serving and Protocols
@@ -545,12 +545,12 @@ The framework is organized in an event-driven architecture where each layer can
 - **Runner layer**: Unified execution entry, coordinating Session / Memory / Artifact services
 - **Tool layer**: FunctionTool / file tools / MCPToolset / Skill tools
 - **Model layer**: OpenAIModel / AnthropicModel / LiteLLMModel
-- **Memory layer**: SessionService / MemoryService / SessionSummarizer / Mem0MemoryService
+- **Memory layer**: SessionService / MemoryService / SessionSummarizer / Mem0MemoryService / MempalaceMemoryService
 - **Knowledge layer**: Production-grade LangChain-based knowledge and RAG capability
 - **Execution and skill layer**: CodeExecutor (local / container) / Skills
 - **Service layer**: FastAPI / A2A / AG-UI
 - **Observability layer**: OpenTelemetry tracing/metrics, integrable with platforms like Langfuse
-- **Ecosystem adapter layer**: claude-agent-sdk / mcp / LangChain / LiteLLM / Mem0 plugged into the main chain through model/tool/memory adapters
+- **Ecosystem adapter layer**: claude-agent-sdk / mcp / LangChain / LiteLLM / Mem0 / Mempalace plugged into the main chain through model/tool/memory adapters
 
 Key packages:
 

README.zh_CN.md

@@ -16,10 +16,10 @@ tRPC-Agent-Python 提供从 Agent 构建、编排、工具接入、会话记忆
 
 - **多范式 Agent 编排**：预设编排支持 ChainAgent / ParallelAgent / CycleAgent / TransferAgent，同时支持 GraphAgent 图编排
 - **图编排能力（GraphAgent）**：通过 DSL 统一编排 Agent / Tool / MCP / Knowledge / CodeExecutor
-- **高效接入 Python AI 生态扩展**：Agent 生态扩展（claude-agent-sdk / LangGraph 等）/ 工具生态扩展（mcp 等）/ 知识库生态扩展（LangChain 等）/ 模型生态扩展（LiteLLM 等）/ 记忆生态扩展（Mem0 等）
+- **高效接入 Python AI 生态扩展**：Agent 生态扩展（claude-agent-sdk / LangGraph 等）/ 工具生态扩展（mcp 等）/ 知识库生态扩展（LangChain 等）/ 模型生态扩展（LiteLLM 等）/ 记忆生态扩展（Mem0、Mempalace等）
 - **Agent 生态扩展**：支持 LangGraphAgent / ClaudeAgent / TeamAgent（Agno-Like）
 - **Tool 生态扩展**：FunctionTool / 文件工具 / MCPToolset / LangChain Tool / Agent-as-Tool
-- **完善的记忆能力（Session / Memory）**：Session 负责单会话内的消息与状态管理，Memory 负责跨会话长期记忆与个性化信息沉淀。持久化支持 InMemory / Redis / SQL，Memory 还支持 Mem0
+- **完善的记忆能力（Session / Memory）**：Session 负责单会话内的消息与状态管理，Memory 负责跨会话长期记忆与个性化信息沉淀。持久化支持 InMemory / Redis / SQL，Memory 还支持 Mem0、Mempalace
 - **生产级知识库能力**：知识库能力基于 LangChain 组件构建，支持 RAG 场景
 - **CodeExecutor 扩展能力**：支持本地 / 容器执行器，用于支持 Agent 的代码执行与任务落地能力
 - **Skills 扩展能力**：支持 SKILL.md 技能体系，用于支持 Agent 的技能复用与动态工具化能力
@@ -84,7 +84,7 @@ pip install trpc-agent-py
 按需安装扩展能力：
 
 ```bash
-pip install trpc-agent-py[a2a,ag-ui,knowledge,agent-claude,mem0,langfuse]
+pip install trpc-agent-py[a2a,ag-ui,knowledge,agent-claude,mem0, Mempalace, langfuse]
 ```
 
 
@@ -458,7 +458,7 @@ skill_tool_set = SkillToolSet(repository=repository, run_tool_kwargs=tool_kwargs
 这组示例可以帮你：
 
 - Session：管理单会话的消息、摘要与状态
-- Memory：管理跨会话长期记忆（含 Mem0）
+- Memory：管理跨会话长期记忆（含 Mem0, Mempalace）
 - Knowledge：覆盖文档加载、检索、RAG、提示模板等链路
 
 ### 10. 服务化与协议
@@ -546,12 +546,12 @@ skill_tool_set = SkillToolSet(repository=repository, run_tool_kwargs=tool_kwargs
 - **Runner 层**：统一执行入口，负责 Session/Memory/Artifact 等服务协同
 - **Tool 层**：FunctionTool / 文件工具 / MCPToolset / Skill 工具
 - **Model 层**：OpenAIModel / AnthropicModel / LiteLLMModel
-- **Memory 层**：SessionService / MemoryService / SessionSummarizer / Mem0MemoryService
+- **Memory 层**：SessionService / MemoryService / SessionSummarizer / Mem0MemoryService / MempalaceMemoryService
 - **Knowledge 层**：基于 LangChain 的生产级知识库能力（RAG）
 - **执行与技能层**：CodeExecutor（本地/容器）/ Skills
 - **服务层**：FastAPI / A2A / AG-UI
 - **观测层**：OpenTelemetry tracing/metrics，可对接 Langfuse 等平台
-- **生态适配层**：claude-agent-sdk / mcp / LangChain / LiteLLM / Mem0，通过模型/工具/记忆适配器接入主链路
+- **生态适配层**：claude-agent-sdk / mcp / LangChain / LiteLLM / Mem0 / MemoryService，通过模型/工具/记忆适配器接入主链路
 
 关键包一览：
 

examples/memory_service_with_sql/agent/agent.py

@@ -10,6 +10,8 @@
 from trpc_agent_sdk.models import OpenAIModel
 from trpc_agent_sdk.tools import FunctionTool
 from trpc_agent_sdk.tools import load_memory_tool
+from trpc_agent_sdk.types import GenerateContentConfig
+from trpc_agent_sdk.types import HttpOptions
 
 from .config import get_model_config
 from .prompts import INSTRUCTION
@@ -25,12 +27,18 @@ def _create_model() -> LLMModel:
 
 def create_agent() -> LlmAgent:
     """ Create an agent"""
+    generate_content_config = GenerateContentConfig(
+        http_options=HttpOptions(extra_body={"chat_template_kwargs": {
+                                     "enable_thinking": False
+                                 }}),
+    )
     agent = LlmAgent(
         name="assistant",
         description="A helpful assistant for conversation",
         model=_create_model(),  # You can change this to your preferred model
         instruction=INSTRUCTION,
         tools=[FunctionTool(get_weather_report), load_memory_tool],
+        generate_content_config=generate_content_config,
     )
     return agent
 

examples/session_service_with_sql/agent/agent.py

@@ -9,6 +9,8 @@
 from trpc_agent_sdk.models import LLMModel
 from trpc_agent_sdk.models import OpenAIModel
 from trpc_agent_sdk.tools import FunctionTool
+from trpc_agent_sdk.types import GenerateContentConfig
+from trpc_agent_sdk.types import HttpOptions
 
 from .config import get_model_config
 from .prompts import INSTRUCTION
@@ -24,12 +26,18 @@ def _create_model() -> LLMModel:
 
 def create_agent() -> LlmAgent:
     """ Create an agent"""
+    generate_content_config = GenerateContentConfig(
+        http_options=HttpOptions(extra_body={"chat_template_kwargs": {
+                                     "enable_thinking": False
+                                 }}),
+    )
     agent = LlmAgent(
         name="assistant",
         description="A helpful assistant for conversation",
         model=_create_model(),  # You can change this to your preferred model
         instruction=INSTRUCTION,
         tools=[FunctionTool(get_weather_report)],
+        generate_content_config=generate_content_config,
     )
     return agent
 

examples/session_service_with_sql/run_agent.py

@@ -76,10 +76,9 @@ async def run_weather_agent():
     ]
 
     for query in demo_queries:
-        # Use a new session for each query
-
         user_content = Content(parts=[Part.from_text(text=query)])
 
+        print(f"👤 User: {query}")
         print("🤖 Assistant: ", end="", flush=True)
         async for event in runner.run_async(user_id=user_id, session_id=current_session_id, new_message=user_content):
             # Check if event.content exists

tests/memory/test_sql_memory_service.py

@@ -278,17 +278,6 @@ async def test_store_skips_events_without_content(self):
         svc._sql_storage.add.assert_not_called()
         svc._sql_storage.commit.assert_not_called()
 
-    async def test_store_raises_on_non_session(self):
-        svc = SqlMemoryService.__new__(SqlMemoryService)
-        svc._memory_service_config = _make_config_no_ttl()
-        svc._sql_storage = _patch_sql_storage()
-        svc._SqlMemoryService__cleanup_task = None
-        svc._SqlMemoryService__cleanup_stop_event = None
-
-        with pytest.raises(TypeError, match="Content must be a Session"):
-            await svc.store_session("not a session")
-
-
 # ---------------------------------------------------------------------------
 # SqlMemoryService — search_memory
 # ---------------------------------------------------------------------------

tests/sessions/test_sql_session_service.py

@@ -90,12 +90,39 @@ def test_from_event_with_function_call(self):
         storage_event = SessionStorageEvent.from_event(session, event)
         assert storage_event.content is not None
 
+    def test_from_event_drops_empty_parts(self):
+        session = Session(id="s1", app_name="app", user_id="user", save_key="k")
+        event = Event(
+            invocation_id="inv-1",
+            author="agent",
+            content=Content(parts=[Part()]),
+        )
+        storage_event = SessionStorageEvent.from_event(session, event)
+        assert storage_event.content is None
+
     def test_from_event_no_content(self):
         session = Session(id="s1", app_name="app", user_id="user", save_key="k")
         event = Event(invocation_id="inv-1", author="agent", actions=EventActions())
         storage_event = SessionStorageEvent.from_event(session, event)
         assert storage_event.content is None
 
+    def test_to_event_drops_legacy_empty_parts(self):
+        storage_event = SessionStorageEvent(
+            id="e1",
+            app_name="app",
+            user_id="user",
+            session_id="s1",
+            invocation_id="inv-1",
+            author="agent",
+            actions=EventActions(),
+            long_running_tool_ids=set(),
+            timestamp=datetime.now(),
+            model_flags=1,
+            content={"parts": [{}], "role": "model"},
+        )
+        event = storage_event.to_event()
+        assert event.content is None
+
     def test_long_running_tool_ids_property(self):
         session = Session(id="s1", app_name="app", user_id="user", save_key="k")
         event = _make_event()

tests/storage/test_sql_common.py

@@ -332,6 +332,14 @@ def test_process_bind_param_spanner(self):
         result = dpt.process_bind_param(value, dialect)
         assert pickle.loads(result) == value
 
+    def test_process_bind_param_mysql(self):
+        dpt = DynamicPickleType()
+        dialect = _make_dialect("mysql")
+        value = {"key": "value", "nums": [1, 2, 3]}
+        result = dpt.process_bind_param(value, dialect)
+        assert isinstance(result, bytes)
+        assert pickle.loads(result) == value
+
     def test_process_bind_param_non_spanner(self):
         dpt = DynamicPickleType()
         dialect = _make_dialect("sqlite")
@@ -352,6 +360,14 @@ def test_process_result_value_spanner(self):
         result = dpt.process_result_value(pickled, dialect)
         assert result == original
 
+    def test_process_result_value_mysql(self):
+        dpt = DynamicPickleType()
+        dialect = _make_dialect("mysql")
+        original = {"key": "value", "nums": [1, 2, 3]}
+        pickled = pickle.dumps(original)
+        result = dpt.process_result_value(pickled, dialect)
+        assert result == original
+
     def test_process_result_value_non_spanner(self):
         dpt = DynamicPickleType()
         dialect = _make_dialect("sqlite")

trpc_agent_sdk/memory/_in_memory_memory_service.py

@@ -200,7 +200,7 @@ def _start_cleanup_task(self) -> None:
             return
 
         self.__cleanup_stop_event = asyncio.Event()
-        self.__cleanup_task = asyncio.create_task(self._cleanup_loop())
+        self.__cleanup_task = asyncio.get_event_loop().create_task(self._cleanup_loop())
         logger.debug("Cleanup task created")
 
     def _stop_cleanup_task(self) -> None:

trpc_agent_sdk/memory/_sql_memory_service.py

@@ -43,6 +43,7 @@
 from trpc_agent_sdk.storage import SqlStorage
 from trpc_agent_sdk.storage import decode_content
 from trpc_agent_sdk.storage import decode_grounding_metadata
+from trpc_agent_sdk.storage import sanitize_content_json
 
 from ._utils import extract_words_lower
 from ._utils import format_timestamp
@@ -110,7 +111,7 @@ def update_event(self, session: Session, event: Event):
         self.error_message = event.error_message
         self.interrupted = event.interrupted
         if event.content:
-            self.content = event.content.model_dump(exclude_none=True, mode="json")
+            self.content = sanitize_content_json(event.content.model_dump(exclude_none=True, mode="json"))
         if event.grounding_metadata:
             self.grounding_metadata = event.grounding_metadata.model_dump(exclude_none=True, mode="json")
         if event.custom_metadata:
@@ -135,7 +136,7 @@ def from_event(cls, session: Session, event: Event) -> MemStorageEvent:
             interrupted=event.interrupted,
         )
         if event.content:
-            storage_event.content = event.content.model_dump(exclude_none=True, mode="json")
+            storage_event.content = sanitize_content_json(event.content.model_dump(exclude_none=True, mode="json"))
         if event.grounding_metadata:
             storage_event.grounding_metadata = event.grounding_metadata.model_dump(exclude_none=True, mode="json")
         if event.custom_metadata:
@@ -150,7 +151,7 @@ def to_event(self) -> Event:
             branch=self.branch,
             actions=self.actions,  # type: ignore
             timestamp=self.timestamp.timestamp(),
-            content=decode_content(self.content),
+            content=decode_content(sanitize_content_json(self.content)),
             long_running_tool_ids=self.long_running_tool_ids,
             partial=self.partial,
             turn_complete=self.turn_complete,
@@ -194,15 +195,15 @@ async def store_session(self, session: Session, agent_context: Optional[AgentCon
 
         Only stores events that are not expired based on event_ttl_seconds.
         """
-        if not isinstance(session, Session):
-            raise TypeError(f"Content must be a Session, got {type(session)}")
-
         async with self._sql_storage.create_db_session() as sql_session:
             is_exist = False
             for event in session.events:
                 if not event.is_model_visible():
                     continue
-                if event.content and event.content.parts:
+                if not event.content or not event.content.parts:
+                    continue
+                content = sanitize_content_json(event.content.model_dump(exclude_none=True, mode="json"))
+                if content:
                     is_exist = True
                     # Check if the event already exists
                     event_key = SqlKey(key=(event.id, session.save_key, session.id), storage_cls=MemStorageEvent)
@@ -324,7 +325,7 @@ def _start_cleanup_task(self) -> None:
             return
 
         self.__cleanup_stop_event = asyncio.Event()
-        self.__cleanup_task = asyncio.create_task(self._cleanup_loop())
+        self.__cleanup_task = asyncio.get_event_loop().create_task(self._cleanup_loop())
         logger.debug("Memory cleanup task created")
 
     def _stop_cleanup_task(self) -> None: