docs: 更新文档

Author: raylchen

docs/mkdocs/en/knowledge_vectorstore.md

@@ -614,7 +614,7 @@ The complete example is available at [knowledge_with_vectorstore](../../../examp
 
 Steps to run the example:
 
-1. Configure environment variables in `examples/knowledge_with_vectorstore/.env`, set `VECTORSTORE_TYPE=tencentvdb` and fill in the Tencent Cloud VectorDB connection parameters:
+1. Configure environment variables in [examples/knowledge_with_vectorstore/.env](../../../examples/knowledge_with_vectorstore/.env), set `VECTORSTORE_TYPE=tencentvdb` and fill in the Tencent Cloud VectorDB connection parameters:
 
 ```bash
 VECTORSTORE_TYPE=tencentvdb

docs/mkdocs/en/langgraph_agent.md

@@ -65,6 +65,24 @@ A graph is the core structure of a workflow, composed of nodes and edges, used t
 
 ```python
 from langgraph.graph import StateGraph, START, END
+from langgraph.prebuilt import ToolNode
+
+from trpc_agent_sdk.agents import langgraph_llm_node
+from trpc_agent_sdk.dsl.graph import State
+
+class MyState(State):
+    result: str
+
+@langgraph_llm_node
+def chatbot_node(state: State):
+    """Chatbot node that can use tools."""
+    return {"messages": [llm_with_tools.invoke(state["messages"])]}
+
+graph_builder = StateGraph(State)
+
+# Add tool node
+tool_node = ToolNode(tools=[])
+graph_builder.add_node("tools", tool_node)
 
 graph_builder = StateGraph(MyState)
 graph_builder.add_node("chatbot", chatbot_node)
@@ -99,7 +117,7 @@ For tool nodes, they are typically used with the `@tool` and `@tool_node` decora
 
 ```python
 from langchain_core.tools import tool
-from trpc_agent_sdk.agents.langgraph_agent import tool_node
+from trpc_agent_sdk.agents import tool_node
 
 @tool
 @tool_node
@@ -154,7 +172,7 @@ Its purpose is to:
 What `LangGraphAgent` actually accepts is not the `StateGraph` builder itself, but the graph object after `compile()`. The framework's `LangGraphAgent` explicitly requires a compiled graph as input.
 
 ```python
-from trpc_agent_sdk.agents.langgraph_agent import LangGraphAgent
+from trpc_agent_sdk.agents import LangGraphAgent
 
 graph = graph_builder.compile()
 
@@ -180,7 +198,7 @@ A LangGraphAgent can be created by providing a compiled LangGraph graph, as show
 The other parameters follow the same definitions as in `LlmAgent`.
 
 ```python
-from trpc_agent_sdk.agents.langgraph_agent import LangGraphAgent
+from trpc_agent_sdk.agents import LangGraphAgent
 
 # Assume the LangGraph has already been built
 graph = build_your_langgraph()
@@ -198,13 +216,13 @@ agent = LangGraphAgent(
 ### Basic Graph Structure
 
 ```python
+from typing import Annotated
+from typing_extensions import TypedDict
 from langchain.chat_models import init_chat_model
 from langchain_core.tools import tool
 from langgraph.graph import StateGraph, START
 from langgraph.graph.message import add_messages
 from langgraph.prebuilt import tools_condition, ToolNode
-from typing_extensions import TypedDict
-from typing import Annotated
 
 # Define the state structure
 class State(TypedDict):
@@ -256,7 +274,7 @@ def custom_chatbot(state: CustomState):
 Used to decorate tool execution nodes, automatically recording tool invocation information. Note that @tool_node must be placed after LangGraph's @tool decorator:
 
 ```python
-from trpc_agent_sdk.agents.langgraph_agent import tool_node
+from trpc_agent_sdk.agents import tool_node
 from langchain_core.tools import tool
 
 @tool
@@ -303,7 +321,7 @@ runner.run_async(..., run_config=run_config)
 After receiving an `Event`, you can access the raw LangGraph payload as shown below:
 
 ```python
-from trpc_agent_sdk.agents.langgraph_agent import get_langgraph_payload
+from trpc_agent_sdk.agents import get_langgraph_payload
 
 async for event in runner.run_async(...):
     langgraph_payload = get_langgraph_payload(event)
@@ -318,7 +336,7 @@ async for event in runner.run_async(...):
 Within LangGraph nodes, you can access `trpc_agent` context information:
 
 ```python
-from trpc_agent_sdk.agents.langgraph_agent import get_langgraph_agent_context
+from trpc_agent_sdk.agents import get_langgraph_agent_context
 
 @langgraph_llm_node
 def context_aware_node(state: State, config: RunnableConfig):
@@ -448,7 +466,7 @@ from typing import AsyncGenerator
 from ag_ui.core import BaseEvent, EventType, CustomEvent
 from trpc_agent_sdk.events import Event as TrpcEvent
 from trpc_agent_sdk.agents.utils import LangGraphEventType, get_event_type
-from trpc_agent_sdk.server.ag_ui._plugin._langgraph_event_translator import (
+from trpc_agent_sdk.server.ag_ui import (
     AgUiLangGraphEventTranslator,
     AgUiTranslationContext,
 )

docs/mkdocs/en/llm_agent.md

@@ -505,7 +505,7 @@ Using thinking mode requires configuring the following components:
   - `thinking_budget`: Token budget for the thinking process, must be less than `max_output_tokens`
 
 ```python
-from trpc_agent_sdk.agents.llm_agent import LlmAgent
+from trpc_agent_sdk.agents import LlmAgent
 from trpc_agent_sdk.models import LLMModel, OpenAIModel
 from trpc_agent_sdk.tools import FunctionTool
 from trpc_agent_sdk.types import GenerateContentConfig
@@ -577,6 +577,7 @@ The framework supports passing an async function as a model creation callback to
 from trpc_agent_sdk.agents import LlmAgent
 from trpc_agent_sdk.models import OpenAIModel, LLMModel
 from trpc_agent_sdk.configs import RunConfig
+from trpc_agent_sdk.runners import Runner
 
 async def create_model(custom_data: dict) -> LLMModel:
     """Model creation callback function
@@ -637,6 +638,8 @@ The implementation mechanism of output_schema has two different methods dependin
 from pydantic import BaseModel
 from typing import List
 
+from trpc_agent_sdk.agents import LlmAgent
+
 class UserProfileOutput(BaseModel):
     """Output schema for user profile analysis."""
     user_name: str
@@ -671,6 +674,8 @@ LlmAgent also supports structured input (input_schema), which is generally used
 ```python
 from pydantic import BaseModel
 from typing import List, Optional
+
+from trpc_agent_sdk.agents import LlmAgent
 from trpc_agent_sdk.tools import AgentTool
 
 class UserProfileInput(BaseModel):

docs/mkdocs/en/memory.md

@@ -18,7 +18,7 @@
 
 ## Core Capabilities of MemoryService
 
-Based on the implementation in `trpc_agent_sdk/memory/`, MemoryService provides the following core capabilities:
+Based on the implementation in [trpc_agent_sdk/memory/](../../../trpc_agent_sdk/memory/), MemoryService provides the following core capabilities:
 
 ### 1. Storing Session Memory
 
@@ -76,7 +76,7 @@ async def search_memory(self, key: str, query: str, limit: int = 10, ...) -> Sea
                 event_ttl.update_expired_at()
 ```
 
-**Keyword Extraction** (`_utils.py`):
+**Keyword Extraction** ([_utils.py](../../../trpc_agent_sdk/memory/_utils.py)):
 ```python
 def extract_words_lower(text: str) -> set[str]:
     """Extract English words and Chinese characters"""
@@ -234,8 +234,8 @@ memory_service = InMemoryMemoryService(memory_service_config=memory_service_conf
 
 **Configuration Example**:
 ```python
-from trpc_agent_sdk.memory import RedisMemoryService, MemoryServiceConfig
 import os
+from trpc_agent_sdk.memory import RedisMemoryService, MemoryServiceConfig
 
 # Read Redis configuration from environment variables
 db_host = os.environ.get("REDIS_HOST", "127.0.0.1")
@@ -308,8 +308,8 @@ EXPIRE memory:weather_app/user_001:session_1 86400  # Expires after 24 hours
 
 **Configuration Example**:
 ```python
-from trpc_agent_sdk.memory import SqlMemoryService, MemoryServiceConfig
 import os
+from trpc_agent_sdk.memory import SqlMemoryService, MemoryServiceConfig
 
 # Read MySQL configuration from environment variables
 db_user = os.environ.get("MYSQL_USER", "root")
@@ -413,7 +413,8 @@ async def _cleanup_expired_async(self) -> None:
 
 ```python
 from trpc_agent_sdk.sessions import InMemorySessionService
-from trpc_agent_sdk.memory import InMemoryMemoryService, MemoryServiceConfig
+from trpc_agent_sdk.memory import MemoryServiceConfig
+from trpc_agent_sdk.memory import InMemoryMemoryService
 from trpc_agent_sdk.runners import Runner
 from trpc_agent_sdk.types import Content, Part
 
@@ -1234,7 +1235,7 @@ Through proper use of MemoryService, you can achieve:
 - Cross-session knowledge sharing
 - Intelligent conversation context
 
-For more detailed usage examples, please refer to the related examples in the `examples/` directory.
+For more detailed usage examples, please refer to the related examples in the [examples/](../../../examples/) directory.
 
 - [examples/memory_service_with_in_memory/run_agent.py](../../../examples/memory_service_with_in_memory/run_agent.py)
 - [examples/memory_service_with_redis/run_agent.py](../../../examples/memory_service_with_redis/run_agent.py)

docs/mkdocs/en/model.md

@@ -221,7 +221,7 @@ export TRPC_AGENT_MODEL_NAME="your-model-name"
 
 ```python
 from trpc_agent_sdk.models import LiteLLMModel
-from trpc_agent_sdk.agents.llm_agent import LlmAgent
+from trpc_agent_sdk.agents import LlmAgent
 
 model = LiteLLMModel(
     model_name="openai/gpt-4o",   # Required: provider/model
@@ -236,7 +236,7 @@ LlmAgent(..., model=model, instruction="...")
 Without explicitly instantiating `LiteLLMModel`, you can pass only the model name string; the framework uses `ModelRegistry`'s `supported_models` regex patterns to match and create a LiteLLMModel instance. In this case, the API Key, base_url, and other settings rely on environment variables (e.g., `OPENAI_API_KEY`, `OPENAI_API_BASE`).
 
 ```python
-from trpc_agent_sdk.agents.llm_agent import LlmAgent
+from trpc_agent_sdk.agents import LlmAgent
 
 LlmAgent(..., model="openai/gpt-4o", instruction="...")
 ```

docs/mkdocs/en/multi_agents.md

@@ -442,7 +442,7 @@ Content Processing Assistant (Main Agent)
 
 TransferAgent is a transfer proxy Agent designed to enable custom Agents without transfer capability (such as `TrpcRemoteA2aAgent`) to gain transfer capability, thereby integrating them into the tRPC-Agent framework's multi-Agent system.
 
-`TrpcRemoteA2aAgent` can be used to simulate a remote Agent scenario (first start an A2A Agent service with `examples/a2a/run_server.py` to mimic a remote Agent).
+`TrpcRemoteA2aAgent` can be used to simulate a remote Agent scenario (first start an A2A Agent service with [examples/a2a/run_server.py](../../../examples/a2a/run_server.py) to mimic a remote Agent).
 
 Through TransferAgent, custom Agents can:
 - **Act as sub_agent**: The parent Agent can transfer control to this Agent, and this Agent can transfer control to parent/sibling Agents

docs/mkdocs/en/session.md

@@ -22,7 +22,7 @@ In trpc-agent, `SessionService` is used to manage `Session` (sessions). A `Sessi
 
 ### Core Structure of Session
 
-Based on the implementation in `trpc_agent/sessions/_session.py`, a `Session` contains the following key fields:
+Based on the implementation in [trpc_agent_sdk/sessions/_session.py](../../../trpc_agent_sdk/sessions/_session.py), a `Session` contains the following key fields:
 
 #### 1. Identity
 
@@ -84,7 +84,7 @@ def extract_state_delta(state_delta: Optional[dict[str, Any]]) -> StateStorageEn
 
 ### Core Features of SessionService
 
-Based on the implementation in `trpc_agent/sessions/`, `SessionService` provides the following core features:
+Based on the implementation in [trpc_agent_sdk/sessions/](../../../trpc_agent_sdk/sessions/), `SessionService` provides the following core features:
 
 #### 1. Session Management (CRUD)
 
@@ -700,7 +700,7 @@ The following examples demonstrate the usage of different SessionService impleme
 
 #### InMemorySessionService
 
-📁 **Example Path**: `examples/session_service_with_in_memory/`
+📁 **Example Path**: [examples/session_service_with_in_memory/](../../../examples/session_service_with_in_memory/)
 
 **Description**:
 - Demonstrates basic usage of In-Memory Session Service
@@ -718,7 +718,7 @@ python3 run_agent.py
 
 #### RedisSessionService
 
-📁 **Example Path**: `examples/session_service_with_redis/`
+📁 **Example Path**: [examples/session_service_with_redis/](../../../examples/session_service_with_redis/)
 
 **Description**:
 - Demonstrates Redis Session Service usage
@@ -736,7 +736,7 @@ python3 run_agent.py
 
 #### SqlSessionService
 
-📁 **Example Path**: `examples/session_service_with_sql/`
+📁 **Example Path**: [examples/session_service_with_sql/](../../../examples/session_service_with_sql/)
 
 **Description**:
 - Demonstrates SQL Session Service usage
@@ -808,7 +808,7 @@ By properly using SessionService, you can achieve:
 - Application-level state sharing
 - Session lifecycle management
 
-For more detailed usage examples, refer to the related examples in the `examples/` directory.
+For more detailed usage examples, refer to the related examples in the [examples/](../../../examples/) directory.
 
 - [examples/session_service_with_in_memory/run_agent.py](../../../examples/session_service_with_in_memory/run_agent.py)
 - [examples/session_service_with_redis/run_agent.py](../../../examples/session_service_with_redis/run_agent.py)

docs/mkdocs/en/skill.md

@@ -96,7 +96,7 @@ Create a skill repository and a workspace executor. If no executor is specified,
 
 ```python
 import os
-from trpc_agent_sdk.agents.llm_agent import LlmAgent
+from trpc_agent_sdk.agents import LlmAgent
 from trpc_agent_sdk.models import OpenAIModel
 from trpc_agent_sdk.skills import SkillToolSet
 from trpc_agent_sdk.skills import create_default_skill_repository
@@ -1428,7 +1428,7 @@ def create_skill_dynamic_tool_set(skill_repository: BaseSkillRepository, only_ac
     )
 
 # agent/agent.py
-from trpc_agent_sdk.agents.llm_agent import LlmAgent
+from trpc_agent_sdk.agents import LlmAgent
 from .tools import create_skill_tool_set, create_skill_dynamic_tool_set
 
 def create_agent():
@@ -1692,7 +1692,7 @@ Here are the results for your requests:
 
 #### Scenario: Real-World Project Example
 
-Based on the actual implementation in `examples/skills_with_dynamic_tools/`:
+Based on the actual implementation in [examples/skills_with_dynamic_tools/](../../../examples/skills_with_dynamic_tools/):
 
 ```python
 # 1. Define all tools (agent/tools/_tools.py)
@@ -2001,7 +2001,7 @@ print(f"Tools state: {ctx.session_state.get(key)}")
 
 #### ✅ Run Results Match Expectations Exactly
 
-Based on actual run results from `examples/skills_with_dynamic_tools/run_agent.py`:
+Based on actual run results from [examples/skills_with_dynamic_tools/run_agent.py](../../../examples/skills_with_dynamic_tools/run_agent.py):
 
 ##### 1. Tool Discovery Mechanism Works Correctly
 ```

docs/mkdocs/en/tool.md

@@ -971,7 +971,7 @@ If `cwd` is not specified, the tools will use the current working directory.
 If you need to use a specific tool individually, you can import and add them one by one:
 
 ```python
-from trpc_agent_sdk.agents.llm_agent import LlmAgent
+from trpc_agent_sdk.agents import LlmAgent
 from trpc_agent_sdk.tools.file_tools import ReadTool, WriteTool, EditTool, GrepTool, BashTool, GlobTool
 
 # Create the working directory

docs/mkdocs/zh/knowledge_vectorstore.md

@@ -614,7 +614,7 @@ print(docs[0].page_content)
 
 示例运行步骤：
 
-1. 在 `examples/knowledge_with_vectorstore/.env` 中配置环境变量，设置 `VECTORSTORE_TYPE=tencentvdb` 并填写腾讯云向量数据库连接参数：
+1. 在 [examples/knowledge_with_vectorstore/.env](../../../examples/knowledge_with_vectorstore/.env) 中配置环境变量，设置 `VECTORSTORE_TYPE=tencentvdb` 并填写腾讯云向量数据库连接参数：
 
 ```bash
 VECTORSTORE_TYPE=tencentvdb