-
2025-05-20
在本快速入门中,我们将指导你使用 Azure AI Foundry SDK 设置本地开发环境。 我们会编写提示,将其作为应用代码的一部分运行,跟踪进行的 LLM 调用,并在 LLM 的输出上运行基本评估。
小窍门
本文的其余部分介绍如何使用 基于中心的项目。 如果要改用 Foundry 项目 ,请选择本文顶部的 Foundry 项目。
先决条件
- 一个 Azure 订阅。 如果没有 Azure 订阅,请在开始操作前先创建一个免费帐户。
- 基于中心的项目。 如果你不熟悉 Azure AI Foundry 并且没有基于中心的项目,请选择本文顶部的 Foundry 项目 以改用 Foundry 项目。
设置开发环境
请确保安装以下包:
pip install azure-ai-projects azure-ai-inference azure-identity
部署模型
小窍门
由于可以在 Azure AI Foundry 门户中 自定义左窗格 ,因此你可能会看到与这些步骤中显示的项不同。 如果未看到要查找的内容,请选择 ... 左窗格底部的更多内容。
登录到 Azure AI Foundry。
选择基于中心的项目。 如果没有基于中心的项目,请选择本文顶部的“Foundry 项目”以改用 Foundry 项目。
从左窗格中选择 “模型目录 ”。
从模型列表中选择“gpt-4o-mini”模型。 可以使用搜索栏查找它。
在模型详细信息页上,选择“部署”。
保留默认部署名称。 选择“部署”。
部署模型后,选择“在操场中打开”以测试模型。
生成聊天应用
创建一个名为 chat.py 的文件。 将以下代码复制并粘贴到其中。
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
project_connection_string = "<your-connection-string-goes-here>"
project = AIProjectClient.from_connection_string(
conn_str=project_connection_string, credential=DefaultAzureCredential()
)
chat = project.inference.get_chat_completions_client()
response = chat.complete(
model="gpt-4o-mini",
messages=[
{
"role": "system",
"content": "You are an AI assistant that speaks like a techno punk rocker from 2350. Be cool but not too cool. Ya dig?",
},
{"role": "user", "content": "Hey, can you help me with my taxes? I'm a freelancer."},
],
)
print(response.choices[0].message.content)
插入连接字符串
项目连接字符串需要从代码调用 Azure AI Foundry 模型中的 Azure OpenAI。
在你曾在 Azure AI Foundry 操场快速入门中创建的 Azure AI Foundry 项目中查找连接字符串。 打开项目,然后在“概述”页上找到连接字符串。
复制连接字符串并替换 chat.py 文件中的 <your-connection-string-goes-here>
。
运行聊天脚本
运行脚本以查看来自模型的响应。
python chat.py
根据用户输入和提示模板生成提示
该脚本使用硬编码的输入和输出消息。 在实际应用中,你将从客户端应用程序获取输入,生成带有模型内部指令的系统消息,然后使用所有消息调用 LLM。
让我们更改脚本,从客户端应用程序获取输入,并使用提示模板生成系统消息。
删除输出响应的脚本的最后一行。
现在定义一个
get_chat_response
函数,该函数会接收消息和上下文、使用提示模板生成系统消息并调用模型。 将此代码添加到现有 chat.py 文件:from azure.ai.inference.prompts import PromptTemplate def get_chat_response(messages, context): # create a prompt template from an inline string (using mustache syntax) prompt_template = PromptTemplate.from_string( prompt_template=""" system: You are an AI assistant that speaks like a techno punk rocker from 2350. Be cool but not too cool. Ya dig? Refer to the user by their first name, try to work their last name into a pun. The user's first name is {{first_name}} and their last name is {{last_name}}. """ ) # generate system message from the template, passing in the context as variables system_message = prompt_template.create_messages(data=context) # add the prompt messages to the user messages return chat.complete( model="gpt-4o-mini", messages=system_message + messages, temperature=1, frequency_penalty=0.5, presence_penalty=0.5, )
注释
提示模板使用 mustache 格式。
get_chat_response 函数可以轻松添加为 FastAPI 或 Flask 应用的路由,以便从前端 Web 应用程序调用此函数。
现在模拟将信息从前端应用程序传递到此函数。 将以下代码添加到 chat.py 文件的末尾。 随意处理消息并添加自己的名字。
if __name__ == "__main__": response = get_chat_response( messages=[{"role": "user", "content": "what city has the best food in the world?"}], context={"first_name": "Jessie", "last_name": "Irwin"}, ) print(response.choices[0].message.content)
运行修订后的脚本以使用此新输入查看模型的响应。
python chat.py
后续步骤
在本快速入门中,你将使用 Azure AI Foundry :
- 创建项目
- 部署模型
- 运行聊天完成
- 创建并运行代理
- 将文件上传到代理
Azure AI Foundry SDK 以多种语言提供,包括 Python、Java、JavaScript 和 C# 。 本快速入门提供了每种语言的说明。
小窍门
本文的其余部分介绍如何使用 Foundry 项目。 如果要改用基于中心的项目,请选择本文顶部的“基于中心的项目”。
先决条件
- 一个 Azure 订阅。 如果没有 Azure 订阅,请在开始操作前先创建一个免费帐户。
- 必须是订阅的 所有者 才能接收使用项目所需的适当访问控制。
重要
本文中标记了“(预览版)”的项目目前为公共预览版。 此预览版未提供服务级别协议,不建议将其用于生产工作负载。 某些功能可能不受支持或者受限。 有关详细信息,请参阅 Microsoft Azure 预览版补充使用条款。
创建 Foundry 项目
在主页上,选择“ 创建代理”。 这会创建一个项目,并包括开始使用基本代理的步骤。
填写项目的名称,然后选择“ 创建”。
部署模型
小窍门
由于可以在 Azure AI Foundry 门户中 自定义左窗格 ,因此你可能会看到与这些步骤中显示的项不同。 如果未看到要查找的内容,请选择 ... 左窗格底部的更多内容。
- 如果刚刚使用 Azure AI Foundry 门户通过 “创建代理 ”链接创建项目,则已部署默认 gpt-4o 模型。
- 或者,登录到 Azure AI Foundry 门户,选择项目,然后选择 模型目录。
- 搜索要部署的模型。 对于本快速入门,请选择“gpt-4o”。
- 选择“确认”。
- 不要更改默认设置。 选择“部署”。
配置你的环境
无需安装 Azure AI Foundry 门户。
安装以下包:
pip install openai azure-ai-projects azure-identity
在运行 Python 脚本之前,请确保使用 CLI
az login
(或az login --use-device-code
) 命令登录以进行身份验证。
小窍门
本文中的所有代码都在 GitHub 快速入门中。
设置这些环境变量以在脚本中使用:
# Azure AI Foundry Endpoint AZURE_ENDPOINT=your_endpoint_here # Azure AI Foundry Model Deployment Name AZURE_DEPLOYMENT=your_deployment_name_here # Project ID (Optional - will be generated if not provided) PROJECT_ID=your_project_id_here MODEL_DEPLOYMENT_NAME=gpt-4o # Note: This sample uses DefaultAzureCredential for authentication # Please ensure you are logged in with Azure CLI using 'az login' # For more authentication options, see: https://learn.microsoft.com/en-us/java/api/overview/azure/identity-readme?view=azure-java-stable
在运行 Java 脚本之前,请确保使用 CLI
az login
(或az login --use-device-code
) 命令登录以进行身份验证。将POM.XML下载到 Java IDE。
小窍门
本文中的所有代码都在 GitHub 快速入门中。
在运行 JavaScript 脚本之前,请确保使用 CLI
az login
(或az login --use-device-code
) 命令登录以进行身份验证。下载 package.json。
使用
npm install
安装包设置这些环境变量以在脚本中使用:
MODEL_DEPLOYMENT_NAME=gpt-4o PROJECT_ENDPOINT=https://<your-foundry-resource-name>.services.ai.azure.com/api/projects/<your-foundry-project-name>
小窍门
本文中的所有代码都在 GitHub 快速入门中。
安装包:
若要在 .NET 项目中使用 Azure AI 服务,需要安装多个 NuGet 包。 在集成终端中使用 .NET CLI 添加 NuGet 包:
# Add Azure AI SDK packages dotnet add package Azure.Identity dotnet add package Azure.AI.Projects dotnet add package Azure.AI.Agents.Persistent dotnet add package Azure.AI.Inference
设置这些环境变量以在脚本中使用:
AZURE_AI_ENDPOINT=https://your.services.ai.azure.com/api/projects/project AZURE_AI_MODEL=your_model_name
在运行 C# 脚本之前,请确保使用 CLI
az login
(或az login --use-device-code
) 命令登录以进行身份验证。
小窍门
本文中的所有代码都在 GitHub 快速入门中。
在运行下一个命令之前,请确保使用 CLI
az login
(或az login --use-device-code
) 命令登录以进行身份验证。获取临时访问令牌。 它将在 60-90 分钟内过期,之后需要刷新。
az account get-access-token --scope https://ai.azure.com/.default
将结果保存为环境变量
AZURE_AI_AUTH_TOKEN
。
小窍门
本文中的所有代码都在 GitHub 快速入门中。
运行聊天完成
聊天完成是 AI 应用程序的基本构建基块。 使用聊天完成,可以发送消息列表并获取响应。
- 如果使用了“创建代理”来创建项目,则你现在将位于代理操场并已准备好试用它。稍后你会回到这里,但首先让我们来试用一下此模型。
- 在左窗格中,选择 游乐场。
- 选择“试用聊天操场”。
- 填写提示并选择 “发送 ”按钮。
- 模型在 “响应 ”窗格中返回响应。
将此代码中的端点替换为你的值:
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
project = AIProjectClient(
endpoint="https://your-foundry-resource-name.ai.azure.com/api/projects",
credential=DefaultAzureCredential(),
)
models = project.inference.get_azure_openai_client(api_version="2025-04-01-preview")
response = models.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system", "content": "You are a helpful writing assistant"},
{"role": "user", "content": "Write me a poem about flowers"},
],
)
print(response.choices[0].message.content)
package com.azure.ai.foundry.samples;
import com.azure.ai.foundry.samples.utils.ConfigLoader;
import com.azure.ai.projects.ProjectsClient;
import com.azure.ai.projects.ProjectsClientBuilder;
import com.azure.ai.projects.models.chat.ChatClient;
import com.azure.ai.projects.models.chat.ChatCompletion;
import com.azure.ai.projects.models.chat.ChatCompletionOptions;
import com.azure.ai.projects.models.chat.ChatMessage;
import com.azure.ai.projects.models.chat.ChatRole;
import com.azure.identity.DefaultAzureCredential;
import com.azure.identity.DefaultAzureCredentialBuilder;
import com.azure.ai.projects.models.chat.ChatCompletionResponse;
import java.util.Arrays;
import java.util.List;
/**
* This sample demonstrates how to use the chat completion API with the Azure AI Foundry SDK.
*
* Chat completions allow you to have interactive, conversational interactions with AI models
* by providing a list of messages and receiving AI-generated responses that maintain context
* across the conversation.
*
* This sample shows:
* 1. How to authenticate with Azure AI Foundry using DefaultAzureCredential
* 2. How to create a chat client for a specific model deployment
* 3. How to structure a conversation with system and user messages
* 4. How to configure and send a chat completion request
* 5. How to process and display the AI-generated response
*
* Prerequisites:
* - An Azure account with access to Azure AI Foundry
* - Azure CLI installed and logged in ('az login')
* - Environment variables set in .env file (AZURE_ENDPOINT, AZURE_DEPLOYMENT)
*/
public class ChatCompletionSample {
public static void main(String[] args) {
// Load configuration values from the .env file
// These include the service endpoint and the deployment name of the model to use
String endpoint = ConfigLoader.getAzureEndpoint();
String deploymentName = ConfigLoader.getAzureDeployment();
// Get DefaultAzureCredential for authentication
// This uses the most appropriate authentication method based on the environment
// For local development, it will use your Azure CLI login credentials
DefaultAzureCredential credential = ConfigLoader.getDefaultCredential();
// Create a projects client to interact with Azure AI Foundry services
// The client requires an authentication credential and an endpoint
ProjectsClient client = new ProjectsClientBuilder()
.credential(credential)
.endpoint(endpoint)
.buildClient();
// Get a chat client for the specified model deployment
// This client provides access to chat completion functionality
ChatClient chatClient = client.getChatClient(deploymentName);
// Create a list of chat messages to form the conversation
// This includes a system message to set the assistant's behavior
// and a user message containing the user's question or prompt
List<ChatMessage> messages = Arrays.asList(
new ChatMessage(ChatRole.SYSTEM, "You are a helpful assistant."),
new ChatMessage(ChatRole.USER, "Tell me about Azure AI Foundry.")
);
// Configure chat completion options including the messages, temperature, and token limit
// - Temperature controls randomness: lower values (like 0.2) give more focused responses,
// higher values (like 0.8) give more creative responses
// - MaxTokens limits the length of the response
ChatCompletionOptions options = new ChatCompletionOptions(messages)
.setTemperature(0.7) // Balanced between deterministic and creative
.setMaxTokens(800); // Limit response length
System.out.println("Sending chat completion request...");
// Send the request and get the AI-generated completion
ChatCompletion completion = chatClient.getChatCompletion(options);
// Display the response
System.out.println("\nResponse from assistant:");
System.out.println(completion.getChoices().get(0).getMessage().getContent());
// Display usage statistics
System.out.println("\nUsage Statistics:");
System.out.println("Prompt Tokens: " + completion.getUsage().getPromptTokens());
System.out.println("Completion Tokens: " + completion.getUsage().getCompletionTokens());
System.out.println("Total Tokens: " + completion.getUsage().getTotalTokens());
}
}
// Get the Azure AI endpoint and deployment name from environment variables
const endpoint = process.env.PROJECT_ENDPOINT;
const deployment = process.env.MODEL_DEPLOYMENT_NAME || 'gpt-4o';
// Create an Azure OpenAI Client
const project = new AIProjectClient(endpoint, new DefaultAzureCredential());
const client = await project.inference.azureOpenAI({
// The API version should match the version of the Azure OpenAI resource
apiVersion: "2024-12-01-preview"
});
// Create a chat completion
const chatCompletion = await client.chat.completions.create({
model: deployment,
messages: [
{ role: "system", content: "You are a helpful writing assistant" },
{ role: "user", content: "Write me a poem about flowers" },
],
});
console.log(`\n==================== 🌷 COMPLETIONS POEM ====================\n`);
console.log(chatCompletion.choices[0].message.content);
using Azure;
using Azure.Identity;
using Azure.AI.Projects;
using Azure.AI.Inference;
var projectEndpoint = new Uri(System.Environment.GetEnvironmentVariable("AZURE_AI_ENDPOINT"));
var modelDeploymentName = System.Environment.GetEnvironmentVariable("AZURE_AI_MODEL");
var credential = new DefaultAzureCredential();
AIProjectClient client = new AIProjectClient(projectEndpoint, credential);
ChatCompletionsClient chatClient = client.GetChatCompletionsClient();
var requestOptions = new ChatCompletionsOptions()
{
Messages =
{
new ChatRequestSystemMessage("You are a helpful assistant."),
new ChatRequestUserMessage("How many feet are in a mile?"),
},
Model = modelDeploymentName
};
Response<ChatCompletions> response = chatClient.Complete(requestOptions);
Console.WriteLine(response.Value.Content);
用您的值替换YOUR-FOUNDRY-RESOURCE-NAME
:
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2024-10-21' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
-d '{
"messages": [
{"role": "system",
"content": "You are a helpful writing assistant"},
{"role": "user",
"content": "Write me a poem about flowers"}
],
"model": "gpt-4o"
}'
创建并运行代理
代理允许使用工具提供更强大的功能。 首先,让我们使用代理编写相同的聊天完成代码。
- 在项目中的左窗格中,选择 “代理”。
- 选择与项目一起创建的代理。
- 如果未使用项目创建一个代理,请选择 “新建代理 ”以立即创建代理。
- 在右侧 “设置 ”窗格中,根据需要更改名称。
- 添加说明,例如“你是一个有用的写作助手”。
- 在“设置”窗格顶部,选择“在操场中试用”。
- 开始与代理聊天,例如,“给我写一首关于鲜花的诗”
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
from azure.ai.agents.models import ListSortOrder
project = AIProjectClient(
endpoint="https://your-foundry-resource-name.ai.azure.com/api/projects",
credential=DefaultAzureCredential(),
)
agent = project.agents.create_agent(
model="gpt-4o",
name="my-agent",
instructions="You are a helpful writing assistant")
thread = project.agents.threads.create()
message = project.agents.messages.create(
thread_id=thread.id,
role="user",
content="Write me a poem about flowers")
run = project.agents.runs.create_and_process(thread_id=thread.id, agent_id=agent.id)
if run.status == "failed":
# Check if you got "Rate limit is exceeded.", then you want to get more quota
print(f"Run failed: {run.last_error}")
# Get messages from the thread
messages = project.agents.messages.list(thread_id=thread.id)
# Get the last message from the sender
messages = project.agents.messages.list(thread_id=thread.id, order=ListSortOrder.ASCENDING)
for message in messages:
if message.run_id == run.id and message.text_messages:
print(f"{message.role}: {message.text_messages[-1].text.value}")
# Delete the agent once done
project.agents.delete_agent(agent.id)
print("Deleted agent")
package com.azure.ai.foundry.samples;
import com.azure.ai.foundry.samples.utils.ConfigLoader;
import com.azure.ai.projects.ProjectsClient;
import com.azure.ai.projects.ProjectsClientBuilder;
import com.azure.ai.projects.models.agent.Agent;
import com.azure.ai.projects.models.agent.AgentClient;
import com.azure.ai.projects.models.agent.AgentMessage;
import com.azure.ai.projects.models.agent.AgentOptions;
import com.azure.ai.projects.models.agent.AgentRole;
import com.azure.ai.projects.models.agent.AgentRun;
import com.azure.ai.projects.models.agent.AgentRunStatus;
import com.azure.ai.projects.models.agent.AgentThread;
import com.azure.identity.DefaultAzureCredential;
import java.util.List;
/**
* This sample demonstrates how to create and run an agent using the Azure AI Foundry SDK.
*
* Agents in Azure AI Foundry are specialized AI assistants that can be customized with
* specific instructions and capabilities to perform particular tasks. They maintain conversation
* history in threads and can be deployed for various use cases.
*
* This sample shows:
* 1. How to authenticate with Azure AI Foundry using DefaultAzureCredential
* 2. How to create an agent with specific instructions and capabilities
* 3. How to create a thread for conversation with the agent
* 4. How to send messages to the agent and run it
* 5. How to wait for the agent to complete its execution
* 6. How to retrieve and display the agent's response
*
* Prerequisites:
* - An Azure account with access to Azure AI Foundry
* - Azure CLI installed and logged in ('az login')
* - Environment variables set in .env file (AZURE_ENDPOINT, AZURE_DEPLOYMENT)
*/
public class AgentSample {
public static void main(String[] args) {
// Load configuration values from the .env file
// These include the service endpoint and the deployment name of the model to use
String endpoint = ConfigLoader.getAzureEndpoint();
String deploymentName = ConfigLoader.getAzureDeployment();
// Get DefaultAzureCredential for authentication
// This uses the most appropriate authentication method based on the environment
// For local development, it will use your Azure CLI login credentials
DefaultAzureCredential credential = ConfigLoader.getDefaultCredential();
// Create a projects client to interact with Azure AI Foundry services
// The client requires an authentication credential and an endpoint
ProjectsClient client = new ProjectsClientBuilder()
.credential(credential)
.endpoint(endpoint)
.buildClient();
// Get an agent client, which provides operations for working with AI agents
// This includes creating, configuring, and running agents
AgentClient agentClient = client.getAgentClient();
// Create a new agent with specialized capabilities and instructions
// The agent is configured with a name, description, instructions, and underlying model
System.out.println("Creating agent...");
Agent agent = agentClient.createAgent(new AgentOptions()
.setName("Research Assistant") // Descriptive name for the agent
.setDescription("An agent that helps with research tasks") // Brief description of the agent's purpose
.setInstructions("You are a research assistant. Help users find information and summarize content.") // Detailed instructions for the agent's behavior
.setModel(deploymentName)); // The underlying AI model to power the agent
System.out.println("Agent created: " + agent.getName() + " (ID: " + agent.getId() + ")");
// Create a thread for the conversation with the agent
// Threads maintain conversation history and state across multiple interactions
System.out.println("Creating thread...");
AgentThread thread = agentClient.createThread();
System.out.println("Thread created: " + thread.getId());
// Create a user message
AgentMessage userMessage = new AgentMessage()
.setRole(AgentRole.USER)
.setContent("Explain what cloud computing is and list three benefits.");
// Run the agent
System.out.println("Running agent...");
AgentRun run = agentClient.createRun(thread.getId(), agent.getId(), userMessage);
System.out.println("Run created: " + run.getId());
// Wait for the run to complete
AgentRun completedRun = waitForRunCompletion(agentClient, thread.getId(), run.getId());
System.out.println("Run completed with status: " + completedRun.getStatus());
// Get messages from the thread
List<AgentMessage> messages = agentClient.getMessages(thread.getId());
// Display the assistant's response
System.out.println("\nConversation:");
for (AgentMessage message : messages) {
System.out.println(message.getRole() + ": " + message.getContent());
}
}
private static AgentRun waitForRunCompletion(AgentClient agentClient, String threadId, String runId) {
AgentRun run = agentClient.getRun(threadId, runId);
while (run.getStatus() == AgentRunStatus.QUEUED || run.getStatus() == AgentRunStatus.IN_PROGRESS) {
try {
System.out.println("Run status: " + run.getStatus() + " - waiting...");
Thread.sleep(1000); // Wait for 1 second before checking again
run = agentClient.getRun(threadId, runId);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new RuntimeException("Thread was interrupted", e);
}
}
return run;
}
}
const endpoint = process.env.PROJECT_ENDPOINT;
const deployment = process.env.MODEL_DEPLOYMENT_NAME || 'gpt-4o';
const client = new AIProjectClient(endpoint, new DefaultAzureCredential());
// Create an Agent
const agent = await client.agents.createAgent(deployment, {
name: 'my-agent',
instructions: 'You are a helpful agent'
});
console.log(`\n==================== 🕵️ POEM AGENT ====================`);
// Create a thread and message
const thread = await client.agents.threads.create();
const prompt = 'Write me a poem about flowers';
console.log(`\n---------------- 📝 User Prompt ---------------- \n${prompt}`);
await client.agents.messages.create(thread.id, 'user', prompt);
// Create run
let run = await client.agents.runs.create(thread.id, agent.id);
// Wait for run to complete
console.log(`\n---------------- 🚦 Run Status ----------------`);
while (['queued', 'in_progress', 'requires_action'].includes(run.status)) {
// Avoid adding a lot of messages to the console
await new Promise((resolve) => setTimeout(resolve, 1000));
run = await client.agents.runs.get(thread.id, run.id);
console.log(`Run status: ${run.status}`);
}
console.log('\n---------------- 📊 Token Usage ----------------');
console.table([run.usage]);
const messagesIterator = await client.agents.messages.list(thread.id);
const assistantMessage = await getAssistantMessage(messagesIterator);
console.log('\n---------------- 💬 Response ----------------');
printAssistantMessage(assistantMessage);
// Clean up
console.log(`\n---------------- 🧹 Clean Up Poem Agent ----------------`);
await client.agents.deleteAgent(agent.id);
console.log(`Deleted Agent, Agent ID: ${agent.id}`);
using Azure;
using Azure.Identity;
using Azure.AI.Agents.Persistent;
// Creating the Client for agents
var projectEndpoint = System.Environment.GetEnvironmentVariable("AZURE_AI_ENDPOINT");
var modelDeploymentName = System.Environment.GetEnvironmentVariable("AZURE_AI_MODEL");
PersistentAgentsClient client = new(projectEndpoint, new DefaultAzureCredential());
// Create an Agent with toolResources and process Agent run
PersistentAgent agent = client.Administration.CreateAgent(
model: modelDeploymentName,
name: "SDK Test Agent - Tutor",
instructions: "You are a personal electronics tutor. Write and run code to answer questions.",
tools: new List<ToolDefinition> { new CodeInterpreterToolDefinition() });
// Create thread for communication
PersistentAgentThread thread = client.Threads.CreateThread();
// Create message to thread
PersistentThreadMessage messageResponse = client.Messages.CreateMessage(
thread.Id,
MessageRole.User,
"I need to solve the equation `3x + 11 = 14`. Can you help me?");
// Run the Agent
ThreadRun run = client.Runs.CreateRun(thread, agent);
// Wait for the run to complete
do
{
Thread.Sleep(TimeSpan.FromMilliseconds(500));
run = client.Runs.GetRun(thread.Id, run.Id);
}
while (run.Status == RunStatus.Queued
|| run.Status == RunStatus.InProgress);
Pageable<PersistentThreadMessage> messages = client.Messages.GetMessages(
threadId: thread.Id,
order: ListSortOrder.Ascending
);
// Print the messages in the thread
WriteMessages(messages);
// Delete the thread and agent after use
client.Threads.DeleteThread(thread.Id);
client.Administration.DeleteAgent(agent.Id);
// Temporary function to use a list of messages in the thread and write them to the console.
static void WriteMessages(IEnumerable<PersistentThreadMessage> messages)
{
foreach (PersistentThreadMessage threadMessage in messages)
{
Console.Write($"{threadMessage.CreatedAt:yyyy-MM-dd HH:mm:ss} - {threadMessage.Role,10}: ");
foreach (MessageContent contentItem in threadMessage.ContentItems)
{
if (contentItem is MessageTextContent textItem)
{
Console.Write(textItem.Text);
}
else if (contentItem is MessageImageFileContent imageFileItem)
{
Console.Write($"<image from ID: {imageFileItem.FileId}");
}
Console.WriteLine();
}
}
}
将 YOUR-FOUNDRY-RESOURCE-NAME
和 YOUR-PROJECT-NAME
替换为你的值:
# Create agent
curl --request POST --url "https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/assistants?api-version=v1" \
-h "authorization: Bearer $AZURE_AI_AUTH_TOKEN" \
-h "content-type: application/json" \
-d '{
"model": "gpt-4o",
"name": "my-agent",
"instructions": "You are a helpful writing assistant"
}'
#Lets say agent ID created is asst_123456789. Use this to run the agent
# Create thread
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json'
#Lets say thread ID created is thread_123456789. Use this in the next step
# Create message using thread ID
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads/thread_123456789/messages?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
-d '{
"role": "user",
"content": "Write me a poem about flowers"
}'
# Run thread with the agent - use both agent id and thread id
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads/thread_123456789/runs?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
--data '{
"assistant_id": "asst_123456789"
}'
# List the messages in the thread using thread ID
curl --request GET --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads/thread_123456789/messages?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json'
# Delete agent once done using agent id
curl --request DELETE --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/assistants/asst_123456789?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json'
将文件添加到代理
现在,让我们添加一个文件搜索工具,使我们能够进行知识检索。
- 下载 product_info_1.md 以提供给代理。
- 在代理的 “设置 ”窗格中,根据需要向下滚动以查找 知识。
- 选择 并添加。
- 选择“文件”以上传 product_info_1.md 文件。
- 选择“添加文件”下的“选择本地文件”。
- 选择“ 上传并保存”。
- 更改代理说明,例如,“你是一个有用的助手,可以从上传的文件搜索信息。
- 问一个问题,例如,“你好,你知道哪些 Contoso 产品?
- 若要添加更多文件,请在 AgentVectorStore 上选择 ... ,然后选择“ 管理”。
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
from azure.ai.agents.models import ListSortOrder, FileSearchTool
project = AIProjectClient(
endpoint="https://your-foundry-resource-name.ai.azure.com/api/projects",
credential=DefaultAzureCredential(),
)
# Upload file and create vector store
file = project.agents.files.upload(file_path="../../../data/product_info_1.md", purpose="assistants")
vector_store = project.agents.vector_stores.create_and_poll(file_ids=[file.id], name="my_vectorstore")
# Create file search tool and agent
file_search = FileSearchTool(vector_store_ids=[vector_store.id])
agent = project.agents.create_agent(
model="gpt-4o",
name="my-assistant",
instructions="You are a helpful assistant and can search information from uploaded files",
tools=file_search.definitions,
tool_resources=file_search.resources,
)
# Create thread and process user message
thread = project.agents.threads.create()
project.agents.messages.create(thread_id=thread.id, role="user", content="Hello, what Contoso products do you know?")
run = project.agents.runs.create_and_process(thread_id=thread.id, agent_id=agent.id)
# Handle run status
if run.status == "failed":
print(f"Run failed: {run.last_error}")
# Print thread messages
messages = project.agents.messages.list(thread_id=thread.id, order=ListSortOrder.ASCENDING)
for message in messages:
if message.run_id == run.id and message.text_messages:
print(f"{message.role}: {message.text_messages[-1].text.value}")
# Cleanup resources
project.agents.vector_stores.delete(vector_store.id)
project.agents.files.delete(file_id=file.id)
project.agents.delete_agent(agent.id)
package com.azure.ai.foundry.samples;
import com.azure.ai.foundry.samples.utils.ConfigLoader;
import com.azure.ai.projects.ProjectsClient;
import com.azure.ai.projects.ProjectsClientBuilder;
import com.azure.ai.projects.models.agent.Agent;
import com.azure.ai.projects.models.agent.AgentClient;
import com.azure.ai.projects.models.agent.AgentMessage;
import com.azure.ai.projects.models.agent.AgentOptions;
import com.azure.ai.projects.models.agent.AgentRole;
import com.azure.ai.projects.models.agent.AgentRun;
import com.azure.ai.projects.models.agent.AgentRunStatus;
import com.azure.ai.projects.models.agent.AgentThread;
import com.azure.ai.projects.models.agent.FileTool;
import com.azure.ai.projects.models.file.File;
import com.azure.ai.projects.models.file.FileClient;
import com.azure.identity.DefaultAzureCredential;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.List;
/**
* This sample demonstrates how to create an agent with file search capability using the Azure AI Foundry SDK.
*
* File search agents can analyze and extract information from documents, allowing users to ask
* questions about the content of those documents. This extends the capabilities of regular agents
* by giving them access to specific information contained in files.
*
* This sample shows:
* 1. How to authenticate with Azure AI Foundry using DefaultAzureCredential
* 2. How to create and upload a sample document
* 3. How to create an agent with file search capabilities
* 4. How to configure the agent with access to specific files
* 5. How to run the agent and ask questions about the document content
* 6. How to retrieve and display the agent's responses
*
* Use cases for file search agents include:
* - Document analysis and summarization
* - Question answering from technical documentation
* - Data extraction from structured files
* - Research assistance across multiple documents
*
* Prerequisites:
* - An Azure account with access to Azure AI Foundry
* - Azure CLI installed and logged in ('az login')
* - Environment variables set in .env file (AZURE_ENDPOINT, AZURE_DEPLOYMENT)
*/
public class FileSearchAgentSample {
public static void main(String[] args) {
// Load configuration values from the .env file
// These include the service endpoint and the deployment name of the model to use
String endpoint = ConfigLoader.getAzureEndpoint();
String deploymentName = ConfigLoader.getAzureDeployment();
// Get DefaultAzureCredential for authentication
// This uses the most appropriate authentication method based on the environment
// For local development, it will use your Azure CLI login credentials
DefaultAzureCredential credential = ConfigLoader.getDefaultCredential();
// Create a projects client to interact with Azure AI Foundry services
// The client requires an authentication credential and an endpoint
ProjectsClient client = new ProjectsClientBuilder()
.credential(credential)
.endpoint(endpoint)
.buildClient();
try {
// Create a sample document containing information about cloud computing
// In a real application, you would use your own existing documents
Path tempFile = createSampleDocument();
// Get a file client to handle file operations
// This is used to upload files that the agent will search through
FileClient fileClient = client.getFileClient();
// Upload the file to Azure AI Foundry
// The uploaded file will be available for the agent to search and analyze
System.out.println("Uploading file: " + tempFile);
File uploadedFile = fileClient.uploadFile(tempFile.toString());
System.out.println("File uploaded with ID: " + uploadedFile.getId());
// Get an agent client
AgentClient agentClient = client.getAgentClient();
// Create an agent with file search capability
System.out.println("Creating agent with file search capability...");
Agent agent = agentClient.createAgent(new AgentOptions()
.setName("Document Assistant")
.setDescription("An agent that helps with document searching")
.setInstructions("You are a document assistant. Help users find information in their documents.")
.setModel(deploymentName)
.addTool(new FileTool()
.addFile(uploadedFile.getId())));
System.out.println("Agent created: " + agent.getName() + " (ID: " + agent.getId() + ")");
// Create a thread for the conversation
System.out.println("Creating thread...");
AgentThread thread = agentClient.createThread();
System.out.println("Thread created: " + thread.getId());
// Create a user message
AgentMessage userMessage = new AgentMessage()
.setRole(AgentRole.USER)
.setContent("Find and list the benefits of cloud computing from my document.");
// Run the agent
System.out.println("Running agent...");
AgentRun run = agentClient.createRun(thread.getId(), agent.getId(), userMessage);
System.out.println("Run created: " + run.getId());
// Wait for the run to complete
AgentRun completedRun = waitForRunCompletion(agentClient, thread.getId(), run.getId());
System.out.println("Run completed with status: " + completedRun.getStatus());
// Get messages from the thread
List<AgentMessage> messages = agentClient.getMessages(thread.getId());
// Display the assistant's response
System.out.println("\nConversation:");
for (AgentMessage message : messages) {
System.out.println(message.getRole() + ": " + message.getContent());
}
// Clean up the temporary file
Files.deleteIfExists(tempFile);
} catch (IOException e) {
System.err.println("Error working with files: " + e.getMessage());
e.printStackTrace();
}
}
private static Path createSampleDocument() throws IOException {
String content = "# Cloud Computing Overview\n\n" +
"Cloud computing is the delivery of computing services—including servers, storage, databases, networking, software, analytics, and intelligence—over the Internet ("the cloud") to offer faster innovation, flexible resources, and economies of scale.\n\n" +
"## Benefits of Cloud Computing\n\n" +
"1. **Cost Savings**: Cloud computing eliminates the capital expense of buying hardware and software and setting up and running on-site data centers.\n\n" +
"2. **Scalability**: Cloud services can be scaled up or down based on demand, providing businesses with flexibility as their needs change.\n\n" +
"3. **Performance**: The biggest cloud computing services run on a worldwide network of secure data centers, which are regularly upgraded to the latest generation of fast and efficient computing hardware.\n\n" +
"4. **Reliability**: Cloud computing makes data backup, disaster recovery, and business continuity easier and less expensive because data can be mirrored at multiple redundant sites on the cloud provider's network.\n\n" +
"5. **Security**: Many cloud providers offer a broad set of policies, technologies, and controls that strengthen your security posture overall.";
Path tempFile = Files.createTempFile("cloud-computing-info-", ".md");
Files.writeString(tempFile, content);
return tempFile;
}
private static AgentRun waitForRunCompletion(AgentClient agentClient, String threadId, String runId) {
AgentRun run = agentClient.getRun(threadId, runId);
while (run.getStatus() == AgentRunStatus.QUEUED || run.getStatus() == AgentRunStatus.IN_PROGRESS) {
try {
System.out.println("Run status: " + run.getStatus() + " - waiting...");
Thread.sleep(1000); // Wait for 1 second before checking again
run = agentClient.getRun(threadId, runId);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new RuntimeException("Thread was interrupted", e);
}
}
return run;
}
}
// Upload a file named product_info_1.md
console.log(`\n==================== 🕵️ FILE AGENT ====================`);
const __dirname = path.dirname(fileURLToPath(import.meta.url));
const filePath = path.join(__dirname, '../data/product_info_1.md');
const fileStream = fs.createReadStream(filePath);
fileStream.on('data', (chunk) => {
console.log(`Read ${chunk.length} bytes of data.`);
});
const file = await client.agents.files.upload(fileStream, 'assistants', {
fileName: 'product_info_1.md'
});
console.log(`Uploaded file, ID: ${file.id}`);
const vectorStore = await client.agents.vectorStores.create({
fileIds: [file.id],
name: 'my_vectorstore'
});
console.log('\n---------------- 🗃️ Vector Store Info ----------------');
console.table([
{
'Vector Store ID': vectorStore.id,
'Usage (bytes)': vectorStore.usageBytes,
'File Count': vectorStore.fileCounts?.total ?? 'N/A'
}
]);
// Create an Agent and a FileSearch tool
const fileSearchTool = ToolUtility.createFileSearchTool([vectorStore.id]);
const fileAgent = await client.agents.createAgent(deployment, {
name: 'my-file-agent',
instructions: 'You are a helpful assistant and can search information from uploaded files',
tools: [fileSearchTool.definition],
toolResources: fileSearchTool.resources,
});
// Create a thread and message
const fileSearchThread = await client.agents.threads.create({ toolResources: fileSearchTool.resources });
const filePrompt = 'What are the steps to setup the TrailMaster X4 Tent?';
console.log(`\n---------------- 📝 User Prompt ---------------- \n${filePrompt}`);
await client.agents.messages.create(fileSearchThread.id, 'user', filePrompt);
// Create run
let fileSearchRun = await client.agents.runs.create(fileSearchThread.id, fileAgent.id).stream();
for await (const eventMessage of fileSearchRun) {
if (eventMessage.event === DoneEvent.Done) {
console.log(`Run completed: ${eventMessage.data}`);
}
if (eventMessage.event === ErrorEvent.Error) {
console.log(`An error occurred. ${eventMessage.data}`);
}
}
const fileSearchMessagesIterator = await client.agents.messages.list(fileSearchThread.id);
const fileAssistantMessage = await getAssistantMessage(fileSearchMessagesIterator);
console.log(`\n---------------- 💬 Response ---------------- \n`);
printAssistantMessage(fileAssistantMessage);
// Clean up
console.log(`\n---------------- 🧹 Clean Up File Agent ----------------`);
client.agents.vectorStores.delete(vectorStore.id);
client.agents.files.delete(file.id);
client.agents.deleteAgent(fileAgent.id);
console.log(`Deleted VectorStore, File, and FileAgent. FileAgent ID: ${fileAgent.id}`);
using Azure;
using Azure.Identity;
using Azure.AI.Agents.Persistent;
// Creating the Client for agents and vector stores
var projectEndpoint = System.Environment.GetEnvironmentVariable("AZURE_AI_ENDPOINT");
var modelDeploymentName = System.Environment.GetEnvironmentVariable("AZURE_AI_MODEL");
PersistentAgentsClient client = new(projectEndpoint, new DefaultAzureCredential());
PersistentAgentFileInfo uploadedAgentFile = client.Files.UploadFile(
filePath: "product_info_1.md",
purpose: PersistentAgentFilePurpose.Agents);
// Create a vector store with the file and wait for it to be processed.
// If you do not specify a vector store, create_message will create a vector store with a default expiration policy of seven days after they were last active
Dictionary<string, string> fileIds = new()
{
{ uploadedAgentFile.Id, uploadedAgentFile.Filename }
};
PersistentAgentsVectorStore vectorStore = client.VectorStores.CreateVectorStore(
name: "my_vector_store");
// Add file ID to vector store.
VectorStoreFile vctFile = client.VectorStores.CreateVectorStoreFile(
vectorStoreId: vectorStore.Id,
fileId: uploadedAgentFile.Id
);
Console.WriteLine($"Added file to vector store. The id file in the vector store is {vctFile.Id}.");
FileSearchToolResource fileSearchToolResource = new FileSearchToolResource();
fileSearchToolResource.VectorStoreIds.Add(vectorStore.Id);
// Create an Agent with toolResources and process Agent run
PersistentAgent agent = client.Administration.CreateAgent(
model: modelDeploymentName,
name: "SDK Test Agent - Retrieval",
instructions: "You are a helpful agent that can help fetch data from files you know about.",
tools: new List<ToolDefinition> { new FileSearchToolDefinition() },
toolResources: new ToolResources() { FileSearch = fileSearchToolResource });
// Create thread for communication
PersistentAgentThread thread = client.Threads.CreateThread();
// Create message to thread
PersistentThreadMessage messageResponse = client.Messages.CreateMessage(
thread.Id,
MessageRole.User,
"Can you give me information on how to mount the product?");
// Run the Agent
ThreadRun run = client.Runs.CreateRun(thread, agent);
// Wait for the run to complete
// This is a blocking call, so it will wait until the run is completed
do
{
Thread.Sleep(TimeSpan.FromMilliseconds(500));
run = client.Runs.GetRun(thread.Id, run.Id);
}
while (run.Status == RunStatus.Queued
|| run.Status == RunStatus.InProgress);
// Create a list of messages in the thread and write them to the console.
Pageable<PersistentThreadMessage> messages = client.Messages.GetMessages(
threadId: thread.Id,
order: ListSortOrder.Ascending
);
WriteMessages(messages, fileIds);
// Delete the thread and agent after use
client.VectorStores.DeleteVectorStore(vectorStore.Id);
client.Files.DeleteFile(uploadedAgentFile.Id);
client.Threads.DeleteThread(thread.Id);
client.Administration.DeleteAgent(agent.Id);
// Helper method to write messages to the console
static void WriteMessages(IEnumerable<PersistentThreadMessage> messages, Dictionary<string, string> fileIds)
{
foreach (PersistentThreadMessage threadMessage in messages)
{
Console.Write($"{threadMessage.CreatedAt:yyyy-MM-dd HH:mm:ss} - {threadMessage.Role,10}: ");
foreach (MessageContent contentItem in threadMessage.ContentItems)
{
if (contentItem is MessageTextContent textItem)
{
if (threadMessage.Role == MessageRole.Agent && textItem.Annotations.Count > 0)
{
string strMessage = textItem.Text;
foreach (MessageTextAnnotation annotation in textItem.Annotations)
{
if (annotation is MessageTextFilePathAnnotation pathAnnotation)
{
strMessage = replaceReferences(fileIds, pathAnnotation.FileId, pathAnnotation.Text, strMessage);
}
else if (annotation is MessageTextFileCitationAnnotation citationAnnotation)
{
strMessage = replaceReferences(fileIds, citationAnnotation.FileId, citationAnnotation.Text, strMessage);
}
}
Console.Write(strMessage);
}
else
{
Console.Write(textItem.Text);
}
}
else if (contentItem is MessageImageFileContent imageFileItem)
{
Console.Write($"<image from ID: {imageFileItem.FileId}");
}
Console.WriteLine();
}
}
}
// Helper method to replace file references in the text
static string replaceReferences(Dictionary<string, string> fileIds, string fileID, string placeholder, string text)
{
if (fileIds.TryGetValue(fileID, out string replacement))
return text.Replace(placeholder, $" [{replacement}]");
else
return text.Replace(placeholder, $" [{fileID}]");
}
将 YOUR-FOUNDRY-RESOURCE-NAME
和 YOUR-PROJECT-NAME
替换为你的值:
#Upload the file
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/files?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-f purpose="assistant" \
-f file="@product_info_1.md" #File object (not file name) to be uploaded.
#Lets say file ID created is assistant-123456789. Use this in the next step
# create vector store
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/vector_stores?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
-d '{
"name": "my_vectorstore",
"file_ids": ["assistant-123456789"]
}'
#Lets say Vector Store ID created is vs_123456789. Use this in the next step
# Create Agent for File Search
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/assistants?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
-d '{
"model": "gpt-4o",
"name": "my-assistant",
"instructions": "You are a helpful assistant and can search information from uploaded files",
"tools": [{"type": "file_search"}],
"tool_resources": {"file_search": {"vector_store_ids": ["vs_123456789"]}}
}'
#Lets say agent ID created is asst_123456789. Use this to run the agent
# Create thread
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json'
#Lets say thread ID created is thread_123456789. Use this in the next step
# Create message using thread ID
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads/thread_123456789/messages?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
-d '{
"role": "user",
"content": "Hello, what Contoso products do you know?"
}'
# Run thread with the agent - use both agent id and thread id
curl --request POST --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads/thread_123456789/runs?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json' \
--data '{
"assistant_id": "asst_123456789"
}'
# List the messages in the thread using thread ID
curl --request GET --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/threads/thread_123456789/messages?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json'
# Delete agent once done using agent id
curl --request DELETE --url 'https://YOUR-FOUNDRY-RESOURCE-NAME.services.ai.azure.com/api/projects/YOUR-PROJECT-NAME/assistants/asst_123456789?api-version=v1' \
-h 'authorization: Bearer $AZURE_AI_AUTH_TOKEN' \
-h 'content-type: application/json'
清理资源
如果不再需要它们,请删除与项目关联的资源组。
在 Azure AI Foundry 门户中,选择右上角的项目名称。 然后选择资源组的链接,在 Azure 门户中将其打开。 选择资源组,然后选择“ 删除”。 确认要删除资源组。
相关内容
反馈
此页面是否有帮助?