mirror of
https://github.com/microsoft/agent-framework.git
synced 2026-06-16 21:04:09 +08:00
fef4fd2c18
* Updating code interpreter samples * Small adjustments to ensure it works as expected * Update dotnet/samples/GettingStarted/Steps/Step03_ChatClientAgent_UsingCodeInterpreterTools.cs Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Address xmldoc * Addressing PR coment, external implementaions + ChatClients update, removing RawRepresentationFactory requirement * Address warnings * Update Fix warnings * Proposed changes for the OpenAIAssistantChatClient * Address PR comments --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Mark Wallace <127216156+markwallace-microsoft@users.noreply.github.com>
167 lines
7.2 KiB
C#
167 lines
7.2 KiB
C#
// Copyright (c) Microsoft. All rights reserved.
|
|
|
|
using System.ClientModel;
|
|
using Azure.AI.Agents.Persistent;
|
|
using Azure.AI.OpenAI;
|
|
using Azure.Identity;
|
|
using Microsoft.Extensions.AI;
|
|
using Microsoft.Extensions.AI.Agents;
|
|
using Microsoft.Shared.Samples;
|
|
using OpenAI.Assistants;
|
|
using OpenAI.Chat;
|
|
using OpenAI.Responses;
|
|
|
|
#pragma warning disable OPENAI001
|
|
|
|
namespace GettingStarted;
|
|
|
|
public class AgentSample(ITestOutputHelper output) : BaseSample(output)
|
|
{
|
|
/// <summary>
|
|
/// Represents the available providers for <see cref="IChatClient"/> instances.
|
|
/// </summary>
|
|
public enum ChatClientProviders
|
|
{
|
|
AzureOpenAI,
|
|
OpenAIChatCompletion,
|
|
OpenAIAssistant,
|
|
OpenAIResponses,
|
|
OpenAIResponses_InMemoryMessageThread,
|
|
OpenAIResponses_ConversationIdThread,
|
|
AzureAIAgentsPersistent
|
|
}
|
|
|
|
protected Task<IChatClient> GetChatClientAsync(ChatClientProviders provider, ChatClientAgentOptions options, CancellationToken cancellationToken = default)
|
|
=> provider switch
|
|
{
|
|
ChatClientProviders.OpenAIChatCompletion => GetOpenAIChatClientAsync(),
|
|
ChatClientProviders.OpenAIAssistant => GetOpenAIAssistantChatClientAsync(options, cancellationToken),
|
|
ChatClientProviders.AzureOpenAI => GetAzureOpenAIChatClientAsync(),
|
|
ChatClientProviders.AzureAIAgentsPersistent => GetAzureAIAgentPersistentClientAsync(options, cancellationToken),
|
|
ChatClientProviders.OpenAIResponses or
|
|
ChatClientProviders.OpenAIResponses_InMemoryMessageThread or
|
|
ChatClientProviders.OpenAIResponses_ConversationIdThread
|
|
=> GetOpenAIResponsesClientAsync(),
|
|
_ => throw new NotSupportedException($"Provider {provider} is not supported.")
|
|
};
|
|
|
|
protected ChatOptions? GetChatOptions(ChatClientProviders? provider)
|
|
=> provider switch
|
|
{
|
|
ChatClientProviders.OpenAIResponses_InMemoryMessageThread => new() { RawRepresentationFactory = static (_) => new ResponseCreationOptions() { StoredOutputEnabled = false } },
|
|
ChatClientProviders.OpenAIResponses_ConversationIdThread => new() { RawRepresentationFactory = static (_) => new ResponseCreationOptions() { StoredOutputEnabled = true } },
|
|
_ => null
|
|
};
|
|
|
|
/// <summary>
|
|
/// For providers that store the agent and the thread on the server side, this will clean and delete
|
|
/// any sample agent and thread that was created during this execution.
|
|
/// </summary>
|
|
/// <param name="provider">The chat client provider type that determines the cleanup process.</param>
|
|
/// <param name="agent">The agent instance to be cleaned up.</param>
|
|
/// <param name="thread">Optional thread associated with the agent that may also need to be cleaned up.</param>
|
|
/// <param name="cancellationToken">Cancellation token to monitor for cancellation requests. The default is <see cref="CancellationToken.None"/>.</param>
|
|
/// <remarks>
|
|
/// Ideally for faster execution and potential cost savings, server-side agents should be reused.
|
|
/// </remarks>
|
|
protected Task AgentCleanUpAsync(ChatClientProviders provider, ChatClientAgent agent, AgentThread? thread = null, CancellationToken cancellationToken = default)
|
|
{
|
|
return provider switch
|
|
{
|
|
ChatClientProviders.AzureAIAgentsPersistent => AzureAIAgentsPersistentAgentCleanUpAsync(agent, thread, cancellationToken),
|
|
|
|
// For other remaining provider sample types, no cleanup is needed as they don't offer a server-side agent/thread clean-up API.
|
|
_ => Task.CompletedTask
|
|
};
|
|
}
|
|
|
|
#region Private GetChatClient
|
|
|
|
private Task<IChatClient> GetOpenAIChatClientAsync()
|
|
=> Task.FromResult(
|
|
new ChatClient(TestConfiguration.OpenAI.ChatModelId, TestConfiguration.OpenAI.ApiKey)
|
|
.AsIChatClient());
|
|
|
|
private Task<IChatClient> GetAzureOpenAIChatClientAsync()
|
|
=> Task.FromResult(
|
|
((TestConfiguration.AzureOpenAI.ApiKey is null)
|
|
// Use Azure CLI credentials if API key is not provided.
|
|
? new AzureOpenAIClient(TestConfiguration.AzureOpenAI.Endpoint, new AzureCliCredential())
|
|
: new AzureOpenAIClient(TestConfiguration.AzureOpenAI.Endpoint, new ApiKeyCredential(TestConfiguration.AzureOpenAI.ApiKey)))
|
|
.GetChatClient(TestConfiguration.AzureOpenAI.DeploymentName)
|
|
.AsIChatClient());
|
|
|
|
private Task<IChatClient> GetOpenAIResponsesClientAsync()
|
|
=> Task.FromResult(
|
|
new OpenAIResponseClient(TestConfiguration.OpenAI.ChatModelId, TestConfiguration.OpenAI.ApiKey)
|
|
.AsIChatClient());
|
|
|
|
private async Task<IChatClient> GetAzureAIAgentPersistentClientAsync(ChatClientAgentOptions options, CancellationToken cancellationToken)
|
|
{
|
|
var persistentAgentsClient = new PersistentAgentsClient(
|
|
TestConfiguration.AzureAI.Endpoint,
|
|
new AzureCliCredential());
|
|
|
|
// If the Id is not provided, create a new agent.
|
|
if (options.Id is null)
|
|
{
|
|
var persistentAgentResponse = await persistentAgentsClient.Administration.CreateAgentAsync(
|
|
model: TestConfiguration.AzureAI.DeploymentName,
|
|
name: options.Name,
|
|
instructions: options.Instructions,
|
|
cancellationToken: cancellationToken);
|
|
|
|
var persistentAgent = persistentAgentResponse.Value;
|
|
|
|
return persistentAgentsClient.AsIChatClient(persistentAgent.Id);
|
|
}
|
|
|
|
return persistentAgentsClient.AsIChatClient(options.Id);
|
|
}
|
|
|
|
private async Task<IChatClient> GetOpenAIAssistantChatClientAsync(ChatClientAgentOptions options, CancellationToken cancellationToken)
|
|
{
|
|
var assistantClient = new AssistantClient(TestConfiguration.OpenAI.ApiKey);
|
|
|
|
// If the Id is not provided, create a new assistant.
|
|
if (options.Id is null)
|
|
{
|
|
Assistant assistant = await assistantClient.CreateAssistantAsync(
|
|
TestConfiguration.OpenAI.ChatModelId,
|
|
new()
|
|
{
|
|
Name = options.Name,
|
|
Instructions = options.Instructions
|
|
},
|
|
cancellationToken);
|
|
|
|
return new NewOpenAIAssistantChatClient(assistantClient, assistant.Id, null);
|
|
}
|
|
|
|
return new NewOpenAIAssistantChatClient(assistantClient, options.Id, null);
|
|
}
|
|
|
|
#endregion
|
|
|
|
#region Private AgentCleanUp
|
|
|
|
private async Task AzureAIAgentsPersistentAgentCleanUpAsync(ChatClientAgent agent, AgentThread? thread, CancellationToken cancellationToken)
|
|
{
|
|
var persistentAgentsClient = agent.ChatClient.GetService<PersistentAgentsClient>();
|
|
if (persistentAgentsClient is null)
|
|
{
|
|
throw new InvalidOperationException("The provided chat client is not a Persistent Agents Chat Client");
|
|
}
|
|
|
|
await persistentAgentsClient.Administration.DeleteAgentAsync(agent.Id, cancellationToken);
|
|
|
|
// If a thread is provided, delete it as well.
|
|
if (thread is not null)
|
|
{
|
|
await persistentAgentsClient.Threads.DeleteThreadAsync(thread.Id, cancellationToken);
|
|
}
|
|
}
|
|
|
|
#endregion
|
|
}
|