Agentes hospedados da Foundry

• SDK do .NET 10 ou posterior

Instale o pacote NuGet de hospedagem:

dotnet add package Microsoft.Agents.AI.Foundry.Hosting --prerelease
dotnet add package Azure.AI.Projects --prerelease

• Python 3.10 ou posterior

Instale o pacote de Python de hospedagem:

pip install agent-framework agent-framework-foundry-hosting

using Azure.AI.AgentServer.Core;
using Azure.AI.Projects;
using Azure.Identity;
using Microsoft.Agents.AI;
using Microsoft.Agents.AI.Foundry.Hosting;

var projectEndpoint = new Uri(Environment.GetEnvironmentVariable("FOUNDRY_PROJECT_ENDPOINT")
    ?? throw new InvalidOperationException("FOUNDRY_PROJECT_ENDPOINT is not set."));
var deployment = Environment.GetEnvironmentVariable("AZURE_AI_MODEL_DEPLOYMENT_NAME") ?? "gpt-4o";

AIAgent agent = new AIProjectClient(projectEndpoint, new DefaultAzureCredential())
    .AsAIAgent(
        model: deployment,
        instructions: "You are a helpful AI assistant.",
        name: "my-agent");

var builder = AgentHost.CreateBuilder(args);
builder.Services.AddFoundryResponses(agent);
builder.RegisterProtocol("responses", endpoints => endpoints.MapFoundryResponses());

var app = builder.Build();
app.Run();

O AgentHost.CreateBuilder cria um host de aplicativo pré-configurado para o ambiente de hospedagem do Foundry. AddFoundryResponses registra seu agente com o manipulador de protocolo Responses e MapFoundryResponses mapeia o /responses endpoint HTTP.

import os

from agent_framework import Agent
from agent_framework.foundry import FoundryChatClient
from agent_framework_foundry_hosting import ResponsesHostServer
from azure.identity import DefaultAzureCredential

client = FoundryChatClient(
    project_endpoint=os.environ["FOUNDRY_PROJECT_ENDPOINT"],
    model=os.environ["AZURE_AI_MODEL_DEPLOYMENT_NAME"],
    credential=DefaultAzureCredential(),
)

agent = Agent(
    client=client,
    instructions="You are a helpful AI assistant.",
    default_options={"store": False},
)

server = ResponsesHostServer(agent)
server.run()

O ResponsesHostServer encapsula seu agente e o expõe por meio do protocolo Foundry Responses. A configuração de store para False em default_options evita a duplicação do histórico de conversas, já que a infraestrutura de hospedagem gerencia o histórico automaticamente.

Com o protocolo Invocações em C#, você implementa uma classe personalizada InvocationHandler para processar solicitações de entrada:

using Azure.AI.AgentServer.Core;
using Azure.AI.AgentServer.Invocations;
using Microsoft.Agents.AI;

var builder = AgentHost.CreateBuilder(args);

builder.Services.AddSingleton<AIAgent, MyAgent>();
builder.Services.AddInvocationsServer();
builder.Services.AddScoped<InvocationHandler, MyInvocationHandler>();

builder.RegisterProtocol("invocations", endpoints => endpoints.MapInvocationsServer());

var app = builder.Build();
app.Run();

O método AddInvocationsServer registra os serviços do Protocolo de Invocações. Você implementa InvocationHandler para definir como seu agente processa cada solicitação.

Para uma configuração leve, use InvocationsHostServer do pacote agent_framework_foundry_hosting. Ele encapsula seu agente da mesma forma ResponsesHostServer e manipula o gerenciamento de sessão automaticamente:

import os

from agent_framework import Agent
from agent_framework.foundry import FoundryChatClient
from agent_framework_foundry_hosting import InvocationsHostServer
from azure.identity import DefaultAzureCredential

client = FoundryChatClient(
    project_endpoint=os.environ["FOUNDRY_PROJECT_ENDPOINT"],
    model=os.environ["AZURE_AI_MODEL_DEPLOYMENT_NAME"],
    credential=DefaultAzureCredential(),
)

agent = Agent(
    client=client,
    instructions="You are a friendly assistant. Keep your answers brief.",
    default_options={"store": False},
)

server = InvocationsHostServer(agent)
server.run()

Para obter controle total sobre o tratamento de solicitações, use InvocationAgentServerHost diretamente do azure.ai.agentserver.invocations pacote e implemente seu próprio manipulador de invocação:

import os
from collections.abc import AsyncGenerator

from agent_framework import Agent, AgentSession
from agent_framework.foundry import FoundryChatClient
from azure.ai.agentserver.invocations import InvocationAgentServerHost
from azure.identity import DefaultAzureCredential
from starlette.requests import Request
from starlette.responses import JSONResponse, Response, StreamingResponse

_sessions: dict[str, AgentSession] = {}

client = FoundryChatClient(
    project_endpoint=os.environ["FOUNDRY_PROJECT_ENDPOINT"],
    model=os.environ["AZURE_AI_MODEL_DEPLOYMENT_NAME"],
    credential=DefaultAzureCredential(),
)

agent = Agent(
    client=client,
    instructions="You are a friendly assistant. Keep your answers brief.",
    default_options={"store": False},
)

app = InvocationAgentServerHost()


@app.invoke_handler
async def handle_invoke(request: Request):
    """Handle streaming multi-turn chat."""
    data = await request.json()
    session_id = request.state.session_id
    stream = data.get("stream", False)
    user_message = data.get("message", None)

    if user_message is None:
        return Response(content="Missing 'message' in request", status_code=400)

    session = _sessions.setdefault(session_id, AgentSession(session_id=session_id))

    if stream:

        async def stream_response() -> AsyncGenerator[str]:
            async for update in agent.run(user_message, session=session, stream=True):
                yield update.text

        return StreamingResponse(
            stream_response(),
            media_type="text/event-stream",
            headers={"Cache-Control": "no-cache", "Connection": "keep-alive"},
        )

    response = await agent.run([user_message], session=session, stream=stream)
    return JSONResponse({"response": response.text})


if __name__ == "__main__":
    app.run()