快速开始

环境准备

• 安装Python：请确保Python 3.10及以上版本已安装。
• 操作系统: Linux，macOS，并安装docker。
• 执行以下命令安装SDK。（强烈建议在Python虚拟环境中安装，以避免与系统包产生冲突）

  # 创建并激活虚拟环境 (linux) 
  python -m venv venv 
  source venv/bin/activate
  
  # 安装sdk
  pip install hw-agentrun 

• 执行以下命令配置华为云凭证，获取华为云凭证请参考认证鉴权。

  export HUAWEICLOUD_SDK_AK="your-access-key"
  export HUAWEICLOUD_SDK_SK="your-secret-key"

本地开发

1. 执行以下命令初始化一个langgraph项目，支持基本的问答能力。

   agentarts init -n my_agent -t langgraph

   执行完成后会在目录下创建如下目录和文件：

   my_agent/
   ├── agent.py              # Agent implementation
   ├── requirements.txt      # Python dependencies
   ├── Dockerfile            
   └── .agentarts_config.yaml # Project configuration

   其中，agent.py代码如下：

   # agent.py
   """
   my_agent - LangGraph Agent Implementation
   
   An agent built using LangGraph for stateful workflows, wrapped with AgentArts SDK runtime.
   
   Environment Variables:
       OPENAI_API_KEY: Your OpenAI API key (required)
           - Get it from: https://platform.openai.com/api-keys
           - Set in .agentarts_config.yaml under runtime.environment_variables
       OPENAI_BASE_URL: Custom API endpoint URL (optional)
           - Use this to connect to OpenAI-compatible APIs (e.g., Azure OpenAI, local LLMs)
           - Example: https://your-api-endpoint.com/v1
           - Leave empty to use default OpenAI API
       OPENAI_MODEL_NAME: Model name to use (optional, default: gpt-4o-mini)
           - Examples: gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-3.5-turbo
           - Or custom model names for OpenAI-compatible APIs
   
   Configuration:
       Edit .agentarts_config.yaml to customize:
       - runtime.environment_variables: Set OPENAI_API_KEY, OPENAI_MODEL_NAME and optionally OPENAI_BASE_URL
       - runtime.invoke_config: Change protocol or port
       - runtime.network_config: Configure VPC if needed
   
   Usage:
       1. Set your OPENAI_API_KEY in .agentarts_config.yaml
       2. (Optional) Set OPENAI_MODEL_NAME to use a different model
       3. (Optional) Set OPENAI_BASE_URL for custom endpoints
       4. Run: agentarts deploy
       5. Invoke: agentarts invoke '{"message": "Hello"}'
   """
   import os
   from typing import Dict, Any, TypedDict, Annotated
   from operator import add
   # 步骤1 导入sdk相关包
   from agentarts.sdk import AgentArtsRuntimeApp, RequestContext    
   # AgentArtsSDK runtime执行入口app
   app = AgentArtsRuntimeApp()
   
   # 步骤2 定义主体Agent业务逻辑
   class State(TypedDict):
       messages: Annotated[list, add]
       query: str
       response: str
   
   class LangGraphAgent:
       def __init__(self):
           self.model_name = os.environ.get("OPENAI_MODEL_NAME", "gpt-4o-mini")
           self._graph = None
   
       def _build_graph(self):
           from langgraph.graph import StateGraph, END
           from langchain_openai import ChatOpenAI
           from langchain_core.messages import HumanMessage, AIMessage
           llm = ChatOpenAI(
               model=self.model_name,
               api_key=os.environ.get("OPENAI_API_KEY"),
               base_url=os.environ.get("OPENAI_BASE_URL")
           )
           async def process_node(state: State) -> Dict[str, Any]:
               query = state.get("query", "")
               messages = state.get("messages", []) or [HumanMessage(content=query)]
               response = await llm.ainvoke(messages)
               return {
                   "messages": [AIMessage(content=response.content)],
                   "response": response.content,
               }
           workflow = StateGraph(State)
           workflow.add_node("process", process_node)
           workflow.set_entry_point("process")
           workflow.add_edge("process", END)
           return workflow.compile()
   
       async def run(self, query: str) -> Dict[str, Any]:
           graph = self._graph or self._build_graph()
           self._graph = graph
           result = await graph.ainvoke({"messages": [], "query": query, "response": ""})
           return {"response": result.get("response", "")}
   
   _agent = LangGraphAgent()
   
   # 步骤3 定义http执行入口
   @app.entrypoint  # 装饰器/invocations端点
   async def handler(payload: Dict[str, Any], context: RequestContext = None) -> Dict[str, Any]:
       query = payload.get("message", "")
       return await _agent.run(query)
   
   if __name__ == "__main__":
       app.run()

2. 配置环境变量等信息，跳转到my_agent目录下，编辑.agentarts_config.yaml如下：

   runtime:
     environment_variables:
       - key: OPENAI_API_KEY
         value: "your-openai-api-key"
       - key: OPENAI_MODEL_NAME
         value: "gpt-4o-mini"  # Optional: gpt-4o, gpt-4-turbo, etc.
       - key: OPENAI_BASE_URL
         value: ""  # Optional: custom API endpoint

3. 执行以下命令进行本地调试。

   pip install -r requirements.txt
   
   agentarts dev

   在服务启动后，可以访问如下地址：

   • 健康检查

     Curl调用示例

     curl http://localhost:8080/ping

     检查结果示例

     {"status":"Healthy","time_of_last_update":1775718928}

   • Agent调用

     Curl请求示例

     curl --location --request POST 'http://localhost:8080/invocations' --data-raw '{"message": "你好，你是谁"}'

     调用成功结果示例

     {"response": "你好！我是 **xxx**"}

云端部署

1. 配置Agent部署region

   agentarts config set region cn-southwest-2

   也可执行agentarts config命令进行交互式配置，配置向导会引导您完成以下设置：（按“回车”使用默认值）

   • 部署region: 默认西南贵阳一cn-southwest-2
   • SWR组织：默认自动创建
   • SWR仓库：默认自动创建
   • 依赖文件：默认requirements.txt

   如果需要自定义配置（例如，指定部署区域、镜像仓库或者agent入站认证、环境变量等），可以手动编辑 .agentarts_config.yaml文件中的配置。

2. 配置完成后，通过命令一键部署到AgentArts平台的云端(需要本地安装docker)

   agentarts launch 

   该命令会自动完成以下步骤：

   1. 本地构建Docker镜像。
   2. 将Docker镜像推送到SWR仓库。
   3. 部署到AgentArts运行时。

3. 调用云端Agent

   agentarts invoke '{"message": "Hello World"}'