基于OpenAI-DotNet的Assistant多轮对话续聊实现疑问

我正在用ChatGPT做相关实验，发现OpenAI-DotNet这个NuGet包使用起来较为简便。我希望实现用户与AI的对话功能，了解到Assistant是合适的实现方式，因此编写了如下代码：

var request = new CreateAssistantRequest(Model.GPT4o);
assistant = await client.AssistantsEndpoint.CreateAssistantAsync(
    new CreateAssistantRequest(
        name: "English Teacher",
        instructions: "You are a personal English Teacher.",
        model: Model.GPT4o));

var messages = new List<Message> { "Good morning! How is the weather?" };
var threadRequest = new CreateThreadRequest(messages);

chats = await assistant.CreateThreadAndRunAsync(threadRequest);

但我不清楚创建Thread后，如何添加更多消息以继续对话，难道每次添加消息都需要重新创建chats？

不需要每次重新创建线程或chats，你只需要保留线程ID，后续直接向该线程添加消息并触发新的Run即可，具体步骤如下：

1. 保存线程ID
   首次创建线程并运行后，从返回的Run对象中提取线程ID并存储，后续对话都基于这个ID操作：

   string threadId = chats.ThreadId;
   

2. 向已有线程添加新消息
   使用ThreadsEndpoint.CreateMessageAsync方法，将用户的新消息添加到指定线程：

   var newMessageRequest = new CreateMessageRequest(
       role: Role.User,
       content: "Teach me some daily English sentences about weather.");
   await client.ThreadsEndpoint.CreateMessageAsync(threadId, newMessageRequest);
   

3. 触发Assistant处理新消息
   调用CreateRunAsync让Assistant处理线程中的最新消息：

   var runRequest = new CreateRunRequest(assistant.Id);
   var newRun = await client.ThreadsEndpoint.CreateRunAsync(threadId, runRequest);
   

4. 等待处理完成并获取对话结果
   轮询Run的状态，直到其完成后拉取线程中的所有消息查看完整对话：

   // 等待Run执行完成
   while (newRun.Status is RunStatus.Queued or RunStatus.InProgress)
   {
       await Task.Delay(1000);
       newRun = await client.ThreadsEndpoint.RetrieveRunAsync(threadId, newRun.Id);
   }
   
   // 获取线程中的所有消息并按时间排序
   var messagesList = await client.ThreadsEndpoint.ListMessagesAsync(threadId);
   foreach (var msg in messagesList.OrderBy(m => m.CreatedAt))
   {
       Console.WriteLine($"{msg.Role}: {msg.Content.First().Text.Value}");
   }
   

完整对话示例代码 #

// 首次创建Assistant和初始线程
var assistant = await client.AssistantsEndpoint.CreateAssistantAsync(
    new CreateAssistantRequest(
        name: "English Teacher",
        instructions: "You are a personal English Teacher.",
        model: Model.GPT4o));

var initialMessages = new List<Message> { new Message(Role.User, "Good morning! How is the weather?") };
var threadRequest = new CreateThreadRequest(initialMessages);
var initialRun = await assistant.CreateThreadAndRunAsync(threadRequest);
string threadId = initialRun.ThreadId;

// 等待首次回复生成
while (initialRun.Status is RunStatus.Queued or RunStatus.InProgress)
{
    await Task.Delay(1000);
    initialRun = await client.ThreadsEndpoint.RetrieveRunAsync(threadId, initialRun.Id);
}

// 添加新消息并触发新一轮对话
var newMessage = new CreateMessageRequest(Role.User, "Teach me some daily English sentences about weather.");
await client.ThreadsEndpoint.CreateMessageAsync(threadId, newMessage);

var newRun = await client.ThreadsEndpoint.CreateRunAsync(threadId, new CreateRunRequest(assistant.Id));

// 等待新回复生成并输出完整对话
while (newRun.Status is RunStatus.Queued or RunStatus.InProgress)
{
    await Task.Delay(1000);
    newRun = await client.ThreadsEndpoint.RetrieveRunAsync(threadId, newRun.Id);
}

var allMessages = await client.ThreadsEndpoint.ListMessagesAsync(threadId);
foreach (var msg in allMessages.OrderBy(m => m.CreatedAt))
{
    Console.WriteLine($"{msg.Role}: {msg.Content.First().Text.Value}");
}

关键要点 #

• 线程（Thread）是对话的持久化容器，所有历史消息都存储在其中，无需重复创建
• 每次用户发送新消息，只需添加到对应线程，再创建新的Run即可触发Assistant的回复
• 必须等待Run状态变为Completed后，才能获取到最新的AI回复消息

内容的提问来源于stack exchange，提问作者Enrico