Neste guia de início rápido, orientamos você na configuração do seu ambiente de desenvolvimento local com o SDK do Azure AI Foundry . Escrevemos um prompt, executamo-lo como parte do código do seu aplicativo, rastreamos as chamadas LLM que estão sendo feitas e executamos uma avaliação básica nas saídas do LLM.
Sugestão
O restante deste artigo mostra como usar um projeto baseado em hub. Selecione Projeto de fundição na parte superior deste artigo se quiser usar um projeto de fundição.
De que tipo de projeto necessito?
Pré-requisitos
- Uma assinatura do Azure. Se você não tiver uma assinatura do Azure, crie uma conta gratuita antes de começar.
- Um projeto baseado em hub. Se você é novo no Azure AI Foundry e não tem um projeto baseado em hub, selecione Projeto do Foundry na parte superior deste artigo para usar um projeto do Foundry.
Configurar seu ambiente de desenvolvimento
Configure seu ambiente de desenvolvimento
Instale estes pacotes.
pip install azure-ai-inference azure-identity azure-ai-projects==1.0.0b10
Observação
Diferentes tipos de projeto requerem diferentes versões do azure-ai-projects
pacote. Para evitar conflitos, crie ambientes Python separados: use a versão 1.0.0b10
para projetos baseados em hub e a versão mais recente para projetos do Foundry.
Implementar um modelo
Sugestão
Como você pode personalizar o painel esquerdo no portal do Azure AI Foundry, poderá ver itens diferentes dos mostrados nestas etapas. Se não encontrar o que procura, selecione ... Mais informações na parte inferior do painel esquerdo.
Entre no Azure AI Foundry.
Selecione um projeto baseado em hub. Se não tiveres um projeto baseado em hub, seleciona projeto Foundry no topo deste artigo para usares um projeto Foundry.
Selecione Catálogo de modelos no painel esquerdo.
Selecione o modelo gpt-4o-mini na lista de modelos. Você pode usar a barra de pesquisa para encontrá-lo.
Na página de detalhes do modelo, selecione Implantar.
Deixe o nome de implantação padrão. Selecione Implantar.
Depois de o modelo ter sido implementado, selecione Abrir no playground para testar o modelo.
Crie seu aplicativo de bate-papo
Crie um arquivo chamado chat.py. Copie e cole o seguinte código nele.
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)
Insira a cadeia de conexão
Sua cadeia de conexão de projeto é necessária para chamar o Azure OpenAI no Azure AI Foundry Models a partir do seu código.
Encontre sua cadeia de conexão no projeto do Azure AI Foundry que você criou no início rápido do playground do Azure AI Foundry. Abra o projeto e localize a cadeia de conexão na página Visão geral .
Copie a cadeia de conexão e substitua <your-connection-string-goes-here>
no arquivo chat.py .
Execute seu script de bate-papo
Execute o script para ver a resposta do modelo.
python chat.py
O script usa mensagens de entrada e saída codificadas. Em um aplicativo real, você receberia a entrada de um aplicativo cliente, geraria uma mensagem do sistema com instruções internas para o modelo e, em seguida, chamaria o LLM com todas as mensagens.
Vamos alterar o script para obter entrada de um aplicativo cliente e gerar uma mensagem do sistema usando um modelo de prompt.
Remova a última linha do script que imprime uma resposta.
Agora defina uma get_chat_response
função que recebe mensagens e contexto, gera uma mensagem do sistema usando um modelo de prompt e chama um modelo. Adicione este código ao seu ficheiro de chat.py existente:
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,
)
Observação
O modelo de prompt usa o formato bigode.
A função get_chat_response pode ser facilmente adicionada como uma rota para um aplicativo FastAPI ou Flask para permitir chamar essa função a partir de um aplicativo Web front-end.
Agora simule a passagem de informações de um aplicativo frontend para essa função. Adicione o seguinte código ao final do arquivo chat.py . Sinta-se à vontade para brincar com a mensagem e adicionar seu próprio nome.
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)
Execute o script revisado para ver a resposta do modelo com essa nova entrada.
python chat.py
Limpeza de recursos
Se você não precisar mais de nenhum dos recursos criados, exclua o grupo de recursos associado ao seu projeto.
No portal do Azure AI Foundry, selecione o nome do seu projeto no canto superior direito. Em seguida, selecione o link para o grupo de recursos para abri-lo no portal do Azure. Selecione o grupo de recursos e, em seguida, selecione Excluir. Confirme que deseja excluir o grupo de recursos.
Próximo passo
Neste início rápido, você usa o Azure AI Foundry para:
- Criar um projeto
- Implementar um modelo
- Executar uma finalização de chat
- Criar e executar um agente
- Carregar arquivos para o agente
O SDK do Azure AI Foundry está disponível em várias linguagens, incluindo Python, Java, JavaScript e C#. Este guia de início rápido fornece instruções para cada um desses idiomas.
Sugestão
O restante deste artigo mostra como usar um projeto do Foundry. Selecione projeto baseado em hub na parte superior deste artigo se quiser usar um projeto baseado em hub.
De que tipo de projeto necessito?
Pré-requisitos
- Uma assinatura do Azure. Se você não tiver uma assinatura do Azure, crie uma conta gratuita antes de começar.
- Você deve ser o proprietário da assinatura para receber o controle de acesso apropriado necessário para usar seu projeto.
Importante
Os itens marcados como (pré-visualização) neste artigo estão neste momento em pré-visualização pública. Esta pré-visualização é fornecida sem um acordo de nível de serviço, e não a recomendamos para trabalhos em produção. Algumas funcionalidades poderão não ser suportadas ou poderão ter capacidades limitadas. Para obter mais informações, veja Termos Suplementares de Utilização para Pré-visualizações do Microsoft Azure.
Comece com um projeto e modelo
Inicie sessão no portal do Azure AI Foundry.
Na página inicial, pesquise e selecione o modelo gpt-4o .
Na página de detalhes do modelo, selecione Usar este modelo.
Preencha um nome para usar em seu projeto e selecione Criar.
Uma vez que os seus recursos são criados, você está na zona de chat.
Configurar o ambiente
Nenhuma instalação é necessária para usar o portal do Azure AI Foundry.
Instale o Python e o Azure CLI
Instale estes pacotes:
pip install openai azure-ai-projects azure-identity
-
Copie o ponto de extremidade do projeto Azure AI Foundry na seção Visão geral do seu projeto. Você vai usá-lo em um momento.
Sugestão
Caso não veja o ponto de extremidade do projeto Azure AI Foundry, está a utilizar um projeto baseado em hub. (Ver Tipos de projetos). Mude para um projeto do Foundry ou use as etapas anteriores para criar um.
Certifique-se de entrar usando o comando CLI az login
(ou az login --use-device-code
) para autenticar antes de executar seus scripts Python.
-
Instale o Java e a CLI do Azure.
- Copie o ponto de extremidade do projeto Azure AI Foundry na seção Visão geral do seu projeto. Você vai usá-lo daqui a pouco.
Sugestão
Caso não veja o ponto de extremidade do projeto Azure AI Foundry, está a utilizar um projeto baseado em hub. (Ver Tipos de projetos). Mude para um projeto do Foundry ou use as etapas anteriores para criar um.
- Certifique-se de fazer login usando o comando CLI
az login
(ou az login --use-device-code
) para autenticar antes de executar seus scripts Java.
- Faça o download POM.XML para o seu IDE Java.
Instalar o Node.js e a CLI do Azure
Certifique-se de entrar usando o comando CLI az login
(ou az login --use-device-code
) para autenticar antes de executar seus scripts JavaScript.
Baixar package.json.
Instale pacotes com npm install
-
Copie o ponto de extremidade do projeto Azure AI Foundry na seção Visão geral do seu projeto. Você vai usá-lo em um momento.
Sugestão
Caso não veja o ponto de extremidade do projeto Azure AI Foundry, está a utilizar um projeto baseado em hub. (Ver Tipos de projetos). Mude para um projeto do Foundry ou use as etapas anteriores para criar um.
Defina estas variáveis de ambiente para usar em seus scripts:
MODEL_DEPLOYMENT_NAME=gpt-4o
PROJECT_ENDPOINT=https://<your-foundry-resource-name>.services.ai.azure.com/api/projects/<your-foundry-project-name>
Instalar o C# e a CLI do Azure
Instalar pacotes:
Para trabalhar com os serviços de IA do Azure em seu projeto .NET, você precisará instalar vários pacotes NuGet. Adicione pacotes NuGet usando a CLI .NET no terminal integrado:
# 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
-
Copie o ponto de extremidade do projeto Azure AI Foundry na seção Visão geral do seu projeto. Você vai usá-lo em um momento.
Sugestão
Caso não veja o ponto de extremidade do projeto Azure AI Foundry, está a utilizar um projeto baseado em hub. (Ver Tipos de projetos). Mude para um projeto do Foundry ou use as etapas anteriores para criar um.
Defina estas variáveis de ambiente para usar em seus scripts:
AZURE_AI_ENDPOINT=https://your.services.ai.azure.com/api/projects/project
AZURE_AI_MODEL=your_model_name
Certifique-se de entrar usando o comando CLI az login
(ou az login --use-device-code
) para autenticar antes de executar seus scripts C#.
Instalar a CLI do Azure
Certifique-se de entrar usando o comando CLI az login
(ou az login --use-device-code
) para autenticar antes de executar o próximo comando.
Obtenha um token de acesso temporário. Ele expirará em 60-90 minutos, você precisará atualizar depois disso.
az account get-access-token --scope https://ai.azure.com/.default
Salve os resultados como a variável AZURE_AI_AUTH_TOKEN
de ambiente .
Executar uma finalização de chat
A conclusão do bate-papo é o bloco de construção básico dos aplicativos de IA. Utilizando as funcionalidades de conclusão de chat, pode enviar uma lista de mensagens e obter uma resposta do modelo.
- No playground de bate-papo, preencha o prompt e selecione o botão Enviar .
- O modelo retorna uma resposta no painel Resposta .
Substitua pelo endpoint
o seu ponto de extremidade neste código.
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
project = AIProjectClient(
endpoint="https://your-foundry-resource-name.ai.azure.com/api/projects/project-name",
credential=DefaultAzureCredential(),
)
models = project.inference.get_azure_openai_client(api_version="2024-10-21")
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)
Defina estas variáveis de ambiente:
-
AZURE_ENDPOINT: Obrigatório. O ponto de extremidade base para o seu serviço de IA do Azure.
- AZURE_AI_API_KEY: Opcional. A chave de API para autenticação (retorna para DefaultAzureCredential se não for fornecida).
- AZURE_MODEL_DEPLOYMENT_NAME: Opcional. O nome da implantação do modelo (o padrão é "phi-4").
- AZURE_MODEL_API_PATH: Opcional. O segmento de caminho da API (o padrão é "implantações").
- CHAT_PROMPT: Opcional. O prompt para enviar para o modelo (usa um padrão se não for fornecido).
package com.azure.ai.foundry.samples;
import com.azure.ai.inference.ChatCompletionsClient;
import com.azure.ai.inference.ChatCompletionsClientBuilder;
import com.azure.ai.inference.models.ChatCompletions;
import com.azure.core.credential.AzureKeyCredential;
import com.azure.core.credential.TokenCredential;
import com.azure.core.exception.HttpResponseException;
import com.azure.core.util.logging.ClientLogger;
import com.azure.identity.DefaultAzureCredential;
import com.azure.identity.DefaultAzureCredentialBuilder;
/**
* Sample demonstrating non-streaming chat completion functionality using the Azure AI Inference SDK.
*
* This sample shows how to:
* - Set up authentication with either API key or Azure credentials
* - Configure a custom endpoint for any Azure AI model
* - Make a simple chat completion request with a single prompt
* - Process and handle the synchronous response
* - Work with the ChatCompletionsClient for basic AI interactions
*
* Environment variables:
* - AZURE_ENDPOINT: Required. The base endpoint for your Azure AI service.
* - AZURE_AI_API_KEY: Optional. The API key for authentication (falls back to DefaultAzureCredential if not provided).
* - AZURE_MODEL_DEPLOYMENT_NAME: Optional. The model deployment name (defaults to "phi-4").
* - AZURE_MODEL_API_PATH: Optional. The API path segment (defaults to "deployments").
* - CHAT_PROMPT: Optional. The prompt to send to the model (uses a default if not provided).
*
* SDK Features Demonstrated:
* - Using the Azure AI Inference SDK (com.azure:azure-ai-inference:1.0.0-beta.5)
* - Creating a ChatCompletionsClient with Azure or API key authentication
* - Configuring endpoint paths for different model deployments
* - Using the simplified complete() method for quick completions
* - Accessing response content through strongly-typed objects
* - Implementing proper error handling for service requests
* - Choosing between DefaultAzureCredential and AzureKeyCredential
*/
public class ChatCompletionSample {
private static final ClientLogger logger = new ClientLogger(ChatCompletionSample.class);
public static void main(String[] args) {
// Load environment variables with proper error handling
String endpoint = System.getenv("AZURE_ENDPOINT");
String apiKey = System.getenv("AZURE_AI_API_KEY");
String deploymentName = System.getenv("AZURE_MODEL_DEPLOYMENT_NAME");
String apiPath = System.getenv("AZURE_MODEL_API_PATH");
String prompt = System.getenv("CHAT_PROMPT");
// Validate required environment variables
if (endpoint == null) {
String errorMessage = "Environment variable AZURE_ENDPOINT is required but not set";
logger.error("ERROR: {}", errorMessage);
logger.error("Please set your environment variables or create a .env file. See README.md for details.");
return;
}
// Set defaults for optional parameters
if (deploymentName == null) {
deploymentName = "phi-4"; // Default to phi-4
logger.info("No AZURE_MODEL_DEPLOYMENT_NAME provided, using default: {}", deploymentName);
}
// Set default API path if not provided
if (apiPath == null) {
apiPath = "deployments";
logger.info("No AZURE_MODEL_API_PATH provided, using default: {}", apiPath);
}
if (prompt == null) {
prompt = "What best practices should I follow when asking an AI model to review Java code?";
logger.info("No CHAT_PROMPT provided, using default prompt: {}", prompt);
}
try {
logger.info("Creating ChatCompletions client with endpoint: {}", endpoint);
// Construct the full endpoint URL including deployment name
String fullEndpoint = endpoint;
if (!fullEndpoint.endsWith("/")) {
fullEndpoint += "/";
}
fullEndpoint += apiPath + "/" + deploymentName;
logger.info("Using full endpoint URL: {}", fullEndpoint);
ChatCompletionsClient client;
// Create client using either API key or Azure credentials with proper error handling
if (apiKey != null && !apiKey.isEmpty()) {
logger.info("Using API key authentication");
client = new ChatCompletionsClientBuilder()
.credential(new AzureKeyCredential(apiKey))
.endpoint(fullEndpoint)
.buildClient();
} else {
logger.info("Using Azure credential authentication with DefaultAzureCredential");
DefaultAzureCredential credential = new DefaultAzureCredentialBuilder().build();
client = new ChatCompletionsClientBuilder()
.credential(credential)
.endpoint(fullEndpoint)
.buildClient();
}
logger.info("Sending chat completion request with prompt: {}", prompt);
// Call the API with the simple prompt interface
ChatCompletions completions = client.complete(prompt);
// Print response
String content = completions.getChoice().getMessage().getContent();
logger.info("Received response from model");
logger.info("\nResponse from AI assistant:\n{}", content);
} catch (HttpResponseException e) {
// Handle service-specific errors with detailed information
int statusCode = e.getResponse().getStatusCode();
logger.error("Service error {}: {}", statusCode, e.getMessage());
// Provide more helpful context based on error status code
if (statusCode == 401 || statusCode == 403) {
logger.error("Authentication error. Check your API key or Azure credentials.");
} else if (statusCode == 404) {
logger.error("Resource not found. Check if the deployment name and endpoint are correct.");
} else if (statusCode == 429) {
logger.error("Rate limit exceeded. Try again later or adjust your request rate.");
}
} catch (Exception e) {
// Handle general exceptions
logger.error("Error in chat completion: {}", e.getMessage(), e);
logger.error("Make sure the Azure AI Inference SDK dependency is correct (using beta.5)");
}
}
}
// 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);
Substitua YOUR-FOUNDRY-RESOURCE-NAME
pelos seus valores:
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"
}'
Conversar com um agente
Os agentes têm capacidades poderosas através do uso de ferramentas. Comece conversando com um agente.
Quando você estiver pronto para experimentar um agente, um agente padrão será criado para você. Para conversar com este agente:
- No painel esquerdo, selecione Playgrounds.
- No cartão Agents playground , selecione Let's go.
- Adicione instruções, como "Você é um assistente de escrita útil."
- Comece a conversar com seu agente, por exemplo: "Escreva-me um poema sobre flores".
Substitua pelo endpoint
o seu ponto de extremidade neste código.
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential
from azure.ai.agents.models import ListSortOrder, FilePurpose
project = AIProjectClient(
endpoint="https://your-foundry-resource-name.ai.azure.com/api/projects/project-name",
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")
Defina estas variáveis de ambiente:
- AZURE_ENDPOINT: Alternativa opcional. O ponto de extremidade base para o serviço de IA do Azure caso PROJECT_ENDPOINT não seja fornecido.
-
PROJECT_ENDPOINT: Obrigatório. O endpoint para o seu projeto de IA do Azure.
- MODEL_DEPLOYMENT_NAME: Opcional. O nome da implantação do modelo (o padrão é "gpt-4o").
- AGENT_NAME: Opcional. O nome a ser dado ao agente criado (o padrão é "java-quickstart-agent").
- AGENT_INSTRUCTIONS: Opcional. As instruções para o agente (o padrão é um assistente útil).
package com.azure.ai.foundry.samples;
import com.azure.ai.agents.persistent.PersistentAgentsClient;
import com.azure.ai.agents.persistent.PersistentAgentsClientBuilder;
import com.azure.ai.agents.persistent.PersistentAgentsAdministrationClient;
import com.azure.ai.agents.persistent.models.CreateAgentOptions;
import com.azure.ai.agents.persistent.models.CreateThreadAndRunOptions;
import com.azure.ai.agents.persistent.models.PersistentAgent;
import com.azure.ai.agents.persistent.models.ThreadRun;
import com.azure.core.credential.TokenCredential;
import com.azure.core.exception.HttpResponseException;
import com.azure.core.util.logging.ClientLogger;
import com.azure.identity.DefaultAzureCredentialBuilder;
/**
* Sample demonstrating how to work with Azure AI Agents using the Azure AI Agents Persistent SDK.
*
* This sample shows how to:
* - Set up authentication with Azure credentials
* - Create a persistent agent with custom instructions
* - Start a thread and run with the agent
* - Access various properties of the agent and thread run
* - Work with the PersistentAgentsClient and PersistentAgentsAdministrationClient
*
* Environment variables:
* - AZURE_ENDPOINT: Optional fallback. The base endpoint for your Azure AI service if PROJECT_ENDPOINT is not provided.
* - PROJECT_ENDPOINT: Required. The endpoint for your Azure AI Project.
* - MODEL_DEPLOYMENT_NAME: Optional. The model deployment name (defaults to "gpt-4o").
* - AGENT_NAME: Optional. The name to give to the created agent (defaults to "java-quickstart-agent").
* - AGENT_INSTRUCTIONS: Optional. The instructions for the agent (defaults to a helpful assistant).
*
* Note: This sample requires proper Azure authentication. It uses DefaultAzureCredential which supports
* multiple authentication methods including environment variables, managed identities, and interactive login.
*
* SDK Features Demonstrated:
* - Using the Azure AI Agents Persistent SDK (com.azure:azure-ai-agents-persistent:1.0.0-beta.2)
* - Creating an authenticated client with DefaultAzureCredential
* - Using the PersistentAgentsClientBuilder pattern for client instantiation
* - Working with the PersistentAgentsAdministrationClient for agent management
* - Creating agents with specific configurations (name, model, instructions)
* - Starting threads and runs for agent conversations
* - Working with agent state and thread management
* - Accessing agent and thread run properties
* - Implementing proper error handling for Azure service interactions
*/
public class AgentSample {
private static final ClientLogger logger = new ClientLogger(AgentSample.class);
public static void main(String[] args) {
// Load environment variables with better error handling, supporting both .env and system environment variables
String endpoint = System.getenv("AZURE_ENDPOINT");
String projectEndpoint = System.getenv("PROJECT_ENDPOINT");
String modelName = System.getenv("MODEL_DEPLOYMENT_NAME");
String agentName = System.getenv("AGENT_NAME");
String instructions = System.getenv("AGENT_INSTRUCTIONS");
// Check for required endpoint configuration
if (projectEndpoint == null && endpoint == null) {
String errorMessage = "Environment variables not configured. Required: either PROJECT_ENDPOINT or AZURE_ENDPOINT must be set.";
logger.error("ERROR: {}", errorMessage);
logger.error("Please set your environment variables or create a .env file. See README.md for details.");
return;
}
// Use AZURE_ENDPOINT as fallback if PROJECT_ENDPOINT not set
if (projectEndpoint == null) {
projectEndpoint = endpoint;
logger.info("Using AZURE_ENDPOINT as PROJECT_ENDPOINT: {}", projectEndpoint);
}
// Set defaults for optional parameters with informative logging
if (modelName == null) {
modelName = "gpt-4o";
logger.info("No MODEL_DEPLOYMENT_NAME provided, using default: {}", modelName);
}
if (agentName == null) {
agentName = "java-quickstart-agent";
logger.info("No AGENT_NAME provided, using default: {}", agentName);
}
if (instructions == null) {
instructions = "You are a helpful assistant that provides clear and concise information.";
logger.info("No AGENT_INSTRUCTIONS provided, using default instructions");
}
// Create Azure credential with DefaultAzureCredentialBuilder
// This supports multiple authentication methods including environment variables,
// managed identities, and interactive browser login
logger.info("Building DefaultAzureCredential");
TokenCredential credential = new DefaultAzureCredentialBuilder().build();
try {
// Build the general agents client
logger.info("Creating PersistentAgentsClient with endpoint: {}", projectEndpoint);
PersistentAgentsClient agentsClient = new PersistentAgentsClientBuilder()
.endpoint(projectEndpoint)
.credential(credential)
.buildClient();
// Derive the administration client
logger.info("Getting PersistentAgentsAdministrationClient");
PersistentAgentsAdministrationClient adminClient =
agentsClient.getPersistentAgentsAdministrationClient();
// Create an agent
logger.info("Creating agent with name: {}, model: {}", agentName, modelName);
PersistentAgent agent = adminClient.createAgent(
new CreateAgentOptions(modelName)
.setName(agentName)
.setInstructions(instructions)
);
logger.info("Agent created: ID={}, Name={}", agent.getId(), agent.getName());
logger.info("Agent model: {}", agent.getModel());
// Start a thread/run on the general client
logger.info("Creating thread and run with agent ID: {}", agent.getId());
ThreadRun runResult = agentsClient.createThreadAndRun(
new CreateThreadAndRunOptions(agent.getId())
);
logger.info("ThreadRun created: ThreadId={}", runResult.getThreadId());
// List available getters on ThreadRun for informational purposes
logger.info("\nAvailable getters on ThreadRun:");
for (var method : ThreadRun.class.getMethods()) {
if (method.getName().startsWith("get")) {
logger.info(" - {}", method.getName());
}
}
logger.info("\nDemo completed successfully!");
} catch (HttpResponseException e) {
// Handle service-specific errors with detailed information
int statusCode = e.getResponse().getStatusCode();
logger.error("Service error {}: {}", statusCode, e.getMessage());
logger.error("Refer to the Azure AI Agents documentation for troubleshooting information.");
} catch (Exception e) {
// Handle general exceptions
logger.error("Error in agent sample: {}", e.getMessage(), e);
}
}
}
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();
}
}
}
Substitua YOUR-FOUNDRY-RESOURCE-NAME
e YOUR-PROJECT-NAME
pelos seus valores:
# 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'
Adicionar arquivos ao agente
Agora vamos adicionar uma ferramenta de pesquisa de arquivos que nos permite fazer a recuperação de conhecimento.
- No painel Configuração do agente, desloque-se para baixo, se necessário, para encontrar Conhecimento.
- Selecione Adicionar.
- Selecione Arquivos para carregar o arquivo product_info_1.md .
- Selecione Selecionar arquivos locais em Adicionar arquivos.
- Selecione Carregar e guardar.
- Altere as instruções dos agentes, como "Você é um assistente útil e pode pesquisar informações de arquivos carregados".
- Faça uma pergunta, como "Olá, quais produtos da Contoso você conhece?"
- Para adicionar mais arquivos, selecione o ... no AgentVectorStore e, em seguida, selecione Gerenciar.
Substitua pelo endpoint
o seu ponto de extremidade neste código.
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/project-name",
credential=DefaultAzureCredential(),
)
# Upload file and create vector store
file = project.agents.files.upload(file_path="./product_info_1.md", purpose=FilePurpose.AGENTS)
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)
Defina estas variáveis de ambiente:
- AZURE_ENDPOINT: Fallback opcional. O ponto de extremidade base para o seu serviço de IA do Azure, caso o PROJECT_ENDPOINT não seja fornecido.
-
PROJECT_ENDPOINT: Obrigatório. O ponto de extremidade para o seu projeto de Inteligência Artificial do Azure.
- MODEL_DEPLOYMENT_NAME: Opcional. O nome da implantação do modelo (o padrão é "gpt-4o").
- AGENT_NAME: Opcional. O nome a ser dado ao agente criado (o padrão é "java-file-search-agent").
- AGENT_INSTRUCTIONS: Opcional. As instruções para o agente (padrão para instruções focadas em documento).
package com.azure.ai.foundry.samples;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import com.azure.ai.agents.persistent.PersistentAgentsClient;
import com.azure.ai.agents.persistent.PersistentAgentsClientBuilder;
import com.azure.ai.agents.persistent.PersistentAgentsAdministrationClient;
import com.azure.ai.agents.persistent.models.CreateAgentOptions;
import com.azure.ai.agents.persistent.models.CreateThreadAndRunOptions;
import com.azure.ai.agents.persistent.models.PersistentAgent;
import com.azure.ai.agents.persistent.models.ThreadRun;
import com.azure.core.exception.HttpResponseException;
import com.azure.core.util.logging.ClientLogger;
import com.azure.identity.DefaultAzureCredentialBuilder;
/**
* Sample demonstrating agent creation with document capabilities using Azure AI Agents Persistent SDK.
*
* This sample shows how to:
* - Set up authentication with Azure credentials
* - Create a temporary document file for demonstration purposes
* - Create a persistent agent with custom instructions for document search
* - Start a thread and run with the agent that can access document content
* - Work with file-based knowledge sources for agent interactions
*
* Environment variables:
* - AZURE_ENDPOINT: Optional fallback. The base endpoint for your Azure AI service if PROJECT_ENDPOINT is not provided.
* - PROJECT_ENDPOINT: Required. The endpoint for your Azure AI Project.
* - MODEL_DEPLOYMENT_NAME: Optional. The model deployment name (defaults to "gpt-4o").
* - AGENT_NAME: Optional. The name to give to the created agent (defaults to "java-file-search-agent").
* - AGENT_INSTRUCTIONS: Optional. The instructions for the agent (defaults to document-focused instructions).
*
* Note: This sample demonstrates the creation of an agent that can process document content.
* In a real-world scenario, you might want to integrate with Azure AI Search or similar services
* for more advanced document processing capabilities.
*
* SDK Features Demonstrated:
* - Using the Azure AI Agents Persistent SDK (com.azure:azure-ai-agents-persistent:1.0.0-beta.2)
* - Creating an authenticated client with DefaultAzureCredential
* - Using the PersistentAgentsClientBuilder for client instantiation
* - Working with the PersistentAgentsAdministrationClient for agent management
* - Creating temporary document files for agent access
* - Adding document knowledge sources to agents
* - Creating document-aware agents that can search and reference content
* - Starting threads and runs for document-based Q&A
* - Error handling for Azure service and file operations
*/
public class FileSearchAgentSample {
private static final ClientLogger logger = new ClientLogger(FileSearchAgentSample.class);
public static void main(String[] args) {
// Load environment variables with proper error handling
String endpoint = System.getenv("AZURE_ENDPOINT");
String projectEndpoint = System.getenv("PROJECT_ENDPOINT");
String modelName = System.getenv("MODEL_DEPLOYMENT_NAME");
String agentName = System.getenv("AGENT_NAME");
String instructions = System.getenv("AGENT_INSTRUCTIONS");
// Check for required endpoint configuration
if (projectEndpoint == null && endpoint == null) {
String errorMessage = "Environment variables not configured. Required: either PROJECT_ENDPOINT or AZURE_ENDPOINT must be set.";
logger.error("ERROR: {}", errorMessage);
logger.error("Please set your environment variables or create a .env file. See README.md for details.");
return;
}
// Set defaults for optional parameters
if (modelName == null) {
modelName = "gpt-4o";
logger.info("No MODEL_DEPLOYMENT_NAME provided, using default: {}", modelName);
}
if (agentName == null) {
agentName = "java-file-search-agent";
logger.info("No AGENT_NAME provided, using default: {}", agentName);
}
if (instructions == null) {
instructions = "You are a helpful assistant that can answer questions about documents.";
logger.info("No AGENT_INSTRUCTIONS provided, using default instructions: {}", instructions);
}
logger.info("Building DefaultAzureCredential");
var credential = new DefaultAzureCredentialBuilder().build();
// Use AZURE_ENDPOINT as fallback if PROJECT_ENDPOINT not set
String finalEndpoint = projectEndpoint != null ? projectEndpoint : endpoint;
logger.info("Using endpoint: {}", finalEndpoint);
try {
// Build the general agents client with proper error handling
logger.info("Creating PersistentAgentsClient with endpoint: {}", finalEndpoint);
PersistentAgentsClient agentsClient = new PersistentAgentsClientBuilder()
.endpoint(finalEndpoint)
.credential(credential)
.buildClient();
// Derive the administration client
logger.info("Getting PersistentAgentsAdministrationClient");
PersistentAgentsAdministrationClient adminClient =
agentsClient.getPersistentAgentsAdministrationClient();
// Create sample document for demonstration
Path tmpFile = createSampleDocument();
logger.info("Created sample document at: {}", tmpFile);
String filePreview = Files.readString(tmpFile).substring(0, 200) + "...";
logger.info("{}", filePreview);
// Create the agent with proper configuration
logger.info("Creating agent with name: {}, model: {}", agentName, modelName);
PersistentAgent agent = adminClient.createAgent(
new CreateAgentOptions(modelName)
.setName(agentName)
.setInstructions(instructions)
);
logger.info("Agent ID: {}", agent.getId());
logger.info("Agent model: {}", agent.getModel());
// Start a thread and run on the general client
logger.info("Creating thread and run with agent ID: {}", agent.getId());
ThreadRun threadRun = agentsClient.createThreadAndRun(
new CreateThreadAndRunOptions(agent.getId())
);
logger.info("ThreadRun ID: {}", threadRun.getThreadId());
// Display success message
logger.info("\nDemo completed successfully!");
} catch (HttpResponseException e) {
// Handle service-specific errors with detailed information
int statusCode = e.getResponse().getStatusCode();
logger.error("Service error {}: {}", statusCode, e.getMessage());
logger.error("Refer to the Azure AI Agents documentation for troubleshooting information.");
} catch (IOException e) {
// Handle IO exceptions specifically for file operations
logger.error("I/O error while creating sample document: {}", e.getMessage(), e);
} catch (Exception e) {
// Handle general exceptions
logger.error("Error in file search agent sample: {}", e.getMessage(), e);
}
}
/**
* Creates a sample markdown document with cloud computing information.
*
* This method demonstrates:
* - Creating a temporary file that will be automatically deleted when the JVM exits
* - Writing structured markdown content to the file
* - Logging file creation and preview of content
*
* In a real application, you might read existing files or create more complex documents.
* You could also upload them to a document storage service for persistent access.
*
* @return Path to the created temporary file
* @throws IOException if an I/O error occurs during file creation or writing
*/
private static Path createSampleDocument() throws IOException {
logger.info("Creating sample document");
String content = """
# Cloud Computing Overview
Cloud computing is the delivery of computing services over the internet, including servers, storage,
databases, networking, software, analytics, and intelligence. Cloud services offer faster innovation,
flexible resources, and economies of scale.
## Key Cloud Service Models
1. **Infrastructure as a Service (IaaS)** - Provides virtualized computing resources
2. **Platform as a Service (PaaS)** - Provides hardware and software tools over the internet
3. **Software as a Service (SaaS)** - Delivers software applications over the internet
## Major Cloud Providers
- Microsoft Azure
- Amazon Web Services (AWS)
- Google Cloud Platform (GCP)
- IBM Cloud
## Benefits of Cloud Computing
- Cost efficiency
- Scalability
- Reliability
- Performance
- Security
""";
Path tempFile = Files.createTempFile("cloud-doc", ".md");
Files.writeString(tempFile, content);
logger.info("Sample document created at: {}", tempFile);
return tempFile;
}
}
// 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}]");
}
Substitua YOUR-FOUNDRY-RESOURCE-NAME
e YOUR-PROJECT-NAME
pelos seus valores:
#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'
Limpeza de recursos
Se você não precisar mais de nenhum dos recursos criados, exclua o grupo de recursos associado ao seu projeto.
No portal do Azure AI Foundry, selecione o nome do seu projeto no canto superior direito. Em seguida, selecione o link para o grupo de recursos para abri-lo no portal do Azure. Selecione o grupo de recursos e, em seguida, selecione Excluir. Confirme que deseja excluir o grupo de recursos.
Próximo passo